http://www.lsoft.com/images/listserv-community.gifhttp://www.lsoft.com/images/listserv_small.gifSPM ListSPM List Archiveshttps://www.jiscmail.ac.uk/cgi-bin/webadmin?RSS&L=SPM&v=ATOM1.0

This is an Atom formatted XML site feed. It is intended to be viewed in a Newsreader.
Alternatively you can view the HTML archives at https://www.jiscmail.ac.uk/cgi-bin/webadmin?A0=SPM.

LISTSERV Web Interface 16.52019-07-22T09:38:38ZMarko Wilke2019-07-22T11:38:30+02:002019-07-22T11:38:30+02:00Re: Cat12 DARTEL template for pediatric sampleshttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;55f8128d.1907Hello Ariadna,<br><br>you could do that, but if you have a small pediatric sample (not in the<br>triple digits) it may be worthwhile using custom Cerebromatic priors for<br>both segmentation and for DARTEL. You will find more information at<br><br>https://doi.org/10.3389/fncom.2017.00005<br><br>and at<br><br>https://doi.org/10.1016/j.dib.2017.12.001<br><br>The toolbox and the dataset (in your case, datasets) described therein<br>are available at http://www.medizin.uni-tuebingen.de/kinder/epn/<br><br>Hope this helps<br>Marko<br><br>Ariadna Albajara Saenz schrieb:<br>> Dear cat experts,<br>><br>> When creating a customised DARTEL-Template, in the first step in the manual (page 39 segment data, grey matter cartel export affine, white matter dartel export affine), is it recommended to enter at this point the customised tissue probability maps created with TOM8?<br>><br>> Thank you in advance,<br>><br>> Ariadna<br>><br><br>--Renew Andrade2019-07-22T10:07:23+02:002019-07-22T10:07:23+02:00VBM CAT12 SPM12https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;9b90a089.1907Dear SPM experts:<br>I am trying to do a group analysis of a few subjects to find a ROI related to those brains. I do not know how to obtain the image with the coloured region though. I can segment , smooth, make the statistical model, estimate, obtain spmT, spmF, morphometrics of different atlas, and do the same for surface analysis. But still can’t find how to obtain the images with the regions highlighted. Sorry for asking such a silly question but I am just learning.<br><br>Thanks in advance!Adeel Razi2019-07-22T16:18:15+10:002019-07-22T16:18:15+10:00Re: Discrepancy in connectivity priors for spDCMhttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;e51cdc47.1907Dear Onur,<br><br>There is no discrepancy here. The self-connections – the leading diagonal<br>of the A matrix – specify log scaling parameters. This means that these<br>parameters encode a self-connections of −1/2∗exp(A); where A has a prior<br>mean of 0 and −1/2∗exp(0) = −1/2. I hope this clarifies the confusion.<br><br>Best wishes,<br>Adeel<br><br>On Thu, Jul 18, 2019 at 5:58 AM Onur Cezmi Mutlu <cezmi@stanford.edu> wrote:<br><br>> Dear DCM experts,<br>><br>> I am currently working on fMRI data and trying to fit spectral DCM models.<br>> While exploring the effects of different priors, I noticed a discrepancy<br>> between the SPM12 priors for spDCM and 2014 spDCM paper "A DCM for resting<br>> state fMRI". In the paper, in Table 1, prior expectations for inhibitory<br>> self connections are set to ln(1/2). On the other hand,<br>> spm_dcm_fmri_priors.m sets them to 0 and spm_dcm_fmri_csd script does not<br>> further modify them to the values in the paper.<br>><br>> Am I missing something or should I modify the prior expectations according<br>> to the paper before fitting a spDCM?<br>><br>> Best regards,<br>> Cezmi<br>>Adeel Razi2019-07-22T16:12:50+10:002019-07-22T16:12:50+10:00Re: Regarding AR model order in spDCMhttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;c4dfde3d.1907Dear Cezmi,<br><br>The order of the AR model is to calculate the data feature that we will<br>eventually fit. When you define AR order, we use this to calculate the AR<br>coefficients (stored in DCM.mar.lag) from which we calculate the `sample'<br>cross spectra (see line 84 and 85 in spm_dcm_fmri_csd_data). We then use<br>this sample cross spectra as a data feature to fit our generative model<br>with an assumption that `predicted' cross spectra comes from an 1/f process<br>(modeled by amplitude and exponent terms). This means that the predicted<br>cross spectra will always have two parameters irrespective of the order of<br>the AR model used to generate sample cross spectra.<br><br>I hope this helps.<br><br>Best wishes,<br>Adeel<br><br>On Wed, Jul 17, 2019 at 5:12 AM Onur Cezmi Mutlu <cezmi@stanford.edu> wrote:<br><br>> Dear Dr. Razi,<br>><br>> Thank you very much for your response, that was very helpful. But I have<br>> two follow-up questions regarding your answer.<br>><br>> As far as I understand from the spm12 convention, Ep.a and Ep.b correspond<br>> to the posterior estimates of neuronal fluctuation and global channel noise<br>> spectral parameters, respectively. They are both arrays of length 2,<br>> regardless of options.order parameter. If we were to model the spectra with<br>> power law or AR(1) then two parameters would be enough to explain it. But<br>> for higher AR orders we need more.<br>><br>> My questions are as follows:<br>> - If changing options.order creates the same effect, why doesn't it change<br>> the number of spectral parameters calculated?<br>> - Unless it is set in the model definition, options.order is set to 8 by<br>> default. But regardless of its value, spm_csd_fmri_mtf.m script performs a<br>> power law modeling followed by AR modeling at the very end (last step of<br>> the script : y = spm_mar2csd(spm_csd2mar(y,M.Hz,M.p - 1),M.Hz);). What is<br>> the idea behind this two leveled modeled and what type of modeling am I<br>> performing if I use the default settings (form='1/f' and options.order=8)?<br>><br>> Best regards,<br>> Cezmi<br>><br>><br>><br>>Ariadna Albajara Saenz2019-07-21T14:15:46+01:002019-07-21T14:15:46+01:00Cat12 DARTEL template for pediatric sampleshttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;f60206f3.1907Dear cat experts,<br><br>When creating a customised DARTEL-Template, in the first step in the manual (page 39 segment data, grey matter cartel export affine, white matter dartel export affine), is it recommended to enter at this point the customised tissue probability maps created with TOM8?<br><br>Thank you in advance,<br><br>AriadnaXinyuan Yan2019-07-21T20:59:39+08:002019-07-21T20:59:39+08:00CAT12-how to get the significant clusters or peak coordinate for further analysishttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;6ab43c87.1907Dear CAT experts,<br>I got the surface results such as thickness in CAT, next, I aimed to use<br>these significant clusters as masks to do further analysis. But I did not<br>know:<br>(1) How to extract the significant clusters and used as a mask?<br>(2) How to get the coordinates of these significant clusters?<br>Looking forward to your reply!<br>Thanks!<br>Best,<br>xinyuan<br><br>Xinyuan Yan<br>Graduate student<br>State Key Laboratory of Cognitive Neuroscience and Learning<br>Xinjiekouwai Street 19, Beijing Haidian. Postcode:100875Bernadette Rusch2019-07-21T10:15:05+02:002019-07-21T10:15:05+02:00Full Factorial - ROI extract main effect & covariates (marsbar)https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;81ec7f1.1907dear all<br><br>I have a full factorial model with two groups (patient/control) and 3<br>conditions (A/B/C) and 3 covariates. We find a main effect of group<br>(patients<controls), but no differences between the conditions (A/B/C). So<br>independently of the condition the patients have less activation in one<br>area.<br><br>I want to know whether this activation difference is driven by the numbers<br>of symptoms (the severity of the disorder). For that reason I want to<br>extract mean paraemter estimates from my region of interest to look at more<br>closely. The extraction will automatically be individually for the 3<br>conditions when using the SPM full factorial design to extract from using<br>marsbar. In order to get the main effect of group (main activation per<br>group independent of the condiion) is it correct to average the extracted<br>mean parameter estimates from my region for condition A, B, C per group?<br><br>Secondly, are the extracted estimates adjusted for the covariates within<br>the model or do I have to adjust for this post-hoc and if so how?<br>I aim to particularly quantify the effect of one specific covariate on our<br>main effect (whether the main effect changes through the inclusion or<br>exclusion of this covariate). Is there a way to get to the average score<br>for my ROI that is either corrected or uncorrected for my covariate?<br><br>Thanks a lot for any input/thoughts!<br>Best, BernadetteYin Wang2019-07-20T22:57:25-04:002019-07-20T22:57:25-04:00Postdoc position at Temple Universityhttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;7356a8aa.1907Full message available at: <a href="https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;7356a8aa.1907">Postdoc position at Temple University</a>Wen-Ze Shao2019-07-21T06:33:42+08:002019-07-21T06:33:42+08:00unsubscribehttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;2055432.1907Dear SPM Team, please unsubscribe me. kind regards, Vinc Peter Indefrey2019-07-20T17:10:11+00:002019-07-20T17:10:11+00:00unsubscribehttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;d35bfa77.1907Dear SPM Team,<br>please unsubscribe me.<br><br>kind regards,<br>Peter Indefrey<br>--<br>Univ. Prof. Dr. phil. Dr. med. Peter Indefrey<br><br>Abt. für Allgemeine Sprachwissenschaft<br>Institut für Sprache und Information<br>Heinrich-Heine-Universität Düsseldorf<br><br>Universitätsstr. 1 Phone: +49 (0)211 81 15464<br>40225 Düsseldorf Fax: +49 (0)211 81 11325<br>Germany indefrey@phil.uni-duesseldorf.de<br><br>Donders Institute for Brain, Cognition and Behaviour,<br>Centre for Cognitive Neuroimaging<br><br>P.O. Box 9101 //204 Phone +31-24-366-6272<br>NL-6500 HB Nijmegen<br>The NetherlandsYAN Chao-Gan2019-07-20T07:36:43+08:002019-07-20T07:36:43+08:00Re: image overlay toolhttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;c312f6f4.1907Or DPABI_VIEW (for NIfTI volume images) or DPABISurf_VIEW (for .gii surface<br>images).<br><br>http://rfmri.org/dpabi or https://github.com/Chaogan-Yan/DPABI/<br><br>Best,<br><br>Chao-Gan<br><br>On Sat, Jul 20, 2019 at 7:30 AM Hupfeld,Kathleen E <khupfeld@ufl.edu> wrote:<br><br>> I'd suggest MRIcroGL (or MRIcron), BSPMview, or FSLeyes.<br>><br>> Best,<br>> Kathleen<br>><br>> ------------------------------<br>> *From:* SPM (Statistical Parametric Mapping) <SPM@JISCMAIL.AC.UK> on<br>> behalf of Wang Ping <pwang_xjtu@HOTMAIL.COM><br>> *Sent:* Friday, July 19, 2019 6:47:39 PM<br>> *To:* SPM@JISCMAIL.AC.UK<br>> *Subject:* [SPM] image overlay tool<br>><br>> could anyone recommend a free software to do image overlay? i.e., to<br>> overlay a colormap of a parameter map within a selected region on the<br>> entire gray scale structural image? ideally display the colorbar as well.<br>><br>> Thanks a lot<br>><br>>Hupfeld,Kathleen E2019-07-19T23:30:23+00:002019-07-19T23:30:23+00:00Re: image overlay toolhttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;522a3eb7.1907I'd suggest MRIcroGL (or MRIcron), BSPMview, or FSLeyes.<br><br>Best,<br>Kathleen Wang Ping2019-07-19T22:47:39+00:002019-07-19T22:47:39+00:00image overlay toolhttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;774cdc3e.1907could anyone recommend a free software to do image overlay? i.e., to overlay a colormap of a parameter map within a selected region on the entire gray scale structural image? ideally display the colorbar as well.<br><br>Thanks a lot David Hofmann2019-07-18T20:27:52+02:002019-07-18T20:27:52+02:00Re: issue to save SPM.mat due to huge number of contrastshttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;f4eb264a.1907Hi Simon,<br><br>just replace: save('SPM.mat', 'SPM', spm_get_defaults('mat.format'));<br>with save('SPM.mat', 'SPM','-v7.3');<br><br>that should work.<br><br>all the best,<br><br>David<br><br>Am Do., 18. Juli 2019 um 18:13 Uhr schrieb simon thibault <<br>simon.thibault@inserm.fr>:<br><br>> Hi,<br>><br>> I am currently trying to specify and estimate a model with lot of<br>> contrasts : each single trial among a total of 9 runs is considered as a<br>> single condition. I would like to do that in order to make easier MVPA<br>> analysis.<br>><br>> However it does not seem to work from a certain amount of contrasts.<br>><br>><br>><br>> Indeed, I get the following warning :<br>><br>><br>><br>> “Warning: Variable 'SPM' cannot be saved to a MAT-file whose version is<br>> older than 7.3.<br>><br>> To save this variable, use the -v7.3 switch.<br>><br>> Skipping...<br>><br>> > In spm_contrasts (line 328)<br>><br>> In spm_run_con (line 277)<br>><br>> In cfg_run_cm (line 29)<br>><br>> In cfg_util>local_runcj (line 1688)<br>><br>> In cfg_util (line 959)<br>><br>> In spm_jobman>fill_run_job (line 469)<br>><br>> In spm_jobman (line 247)”<br>><br>><br>><br>><br>><br>> The SPM.mat file cannot be read. I tried to specify the option –v7.3 in<br>> the spm codes mentioned above but it does not seem to work and I got the<br>> following error message :<br>><br>><br>><br>> “The -V6 option may not be used in combination with the -V7.3 option.<br>><br>> In file "D:\SPM\spm12\spm_contrasts.m" (v7029), function "spm_contrasts"<br>> at line 328.<br>><br>> In file "D:\SPM\spm12\config\spm_run_con.m" (v7093), function<br>> "spm_run_con" at line 277.<br>><br>> The following modules did not run:<br>><br>> Failed: Contrast Manager”<br>><br>><br>><br>><br>><br>> Did someone already have this issue ?<br>><br>><br>><br>> Best,<br>><br>><br>><br>> Simon<br>><br>><br>>simon thibault2019-07-18T18:08:57+02:002019-07-18T18:08:57+02:00issue to save SPM.mat due to huge number of contrastshttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;e371a144.1907Hi,<br><br>I am currently trying to specify and estimate a model with lot of contrasts<br>: each single trial among a total of 9 runs is considered as a single<br>condition. I would like to do that in order to make easier MVPA analysis.<br><br>However it does not seem to work from a certain amount of contrasts.<br><br>Indeed, I get the following warning :<br><br>"Warning: Variable 'SPM' cannot be saved to a MAT-file whose version is<br>older than 7.3.<br><br>To save this variable, use the -v7.3 switch.<br><br>Skipping...<br><br>> In spm_contrasts (line 328)<br><br>In spm_run_con (line 277)<br><br>In cfg_run_cm (line 29)<br><br>In cfg_util>local_runcj (line 1688)<br><br>In cfg_util (line 959)<br><br>In spm_jobman>fill_run_job (line 469)<br><br>In spm_jobman (line 247)"<br><br>The SPM.mat file cannot be read. I tried to specify the option -v7.3 in the<br>spm codes mentioned above but it does not seem to work and I got the<br>following error message :<br><br>"The -V6 option may not be used in combination with the -V7.3 option.<br><br>In file "D:\SPM\spm12\spm_contrasts.m" (v7029), function "spm_contrasts" at<br>line 328.<br><br>In file "D:\SPM\spm12\config\spm_run_con.m" (v7093), function "spm_run_con"<br>at line 277.<br><br>The following modules did not run:<br><br>Failed: Contrast Manager"<br><br>Did someone already have this issue ?<br><br>Best,<br><br>Simon Iris Proff2019-07-18T17:33:53+02:002019-07-18T17:33:53+02:002nd level design matrix and principal eigenvariateshttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;ca11fcf9.1907Dear experts,<br><br>I have two questions concerning 2nd level fmri results in SPM.<br><br>1) Design matrix<br><br>Which quantity exactly is plotted in the 2nd level design matrix? For one flexible factorial analysis (35 subjects, 12 conditions), I get one outlier (i.e. one condition of one subject) notable as a black stripe (with a corresponding value of 0) in the design matrix. This outlier seems to prevent me from viewing the results for my defined contrasts, as SPM throws the following warning:<br><br>Warning: Returning NaN for out of range arguments<br><br>I have a hard time figuring out what is wrong with that subject/condition, as I am not sure how to interpret the design matrix values.<br><br>2) Principal eigenvariates extraction<br><br>In another analysis (35 subjects, 4 conditions), I ran the extraction of eigenvariates in a voxel of interest on my 2nd level results using the following steps:<br><br>Navigate to voxel<br>Eigenvariates<br>Don’t adjust<br>Sphere radius = 0 mm<br><br>When I average the principal eigenvariates I get from this procedure over my four conditions, the results appear to be completely unrelated to the results I get when I plot my average contrast estimates at the same voxel. Why is that? In order to find if activation of a certain voxel correlates with a covariate, is it more appropriate to use eigenvariates or the individual contrast estimates (obtained by plot - fitted responses)?<br><br>Very much appreciate your help!<br><br>Kind regards,<br>Iris Proff<br><br>Max Planck Institute for Metabolism ResearchVarun Arunachalam Chandran2019-07-18T14:13:58+01:002019-07-18T14:13:58+01:00CAT12 segmentation errorhttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;cb284f79.1907Hi everyone,<br><br>I'm facing an issue in performing cortical thickness analysis using CAT12. In the first step when I tried running the segmentation using the default pipeline, it is throwing off an error which is shown below. Could anyone please suggest me some solution? Any help on this would be very much appreciated.<br><br>thanks,<br><br>Print 'Graphics' figure to:<br>/storage/shared/research/cinn/2009/emprew/Varun_ExploreDTI/Cortical_thickness_analysis_CAT_91subjs/report/catreport_Con_P001_T1.pdf<br>Failed 'CAT12: Segmentation'<br>Error using cat_check_system_output (line 27)<br>ln: failed to create symbolic link ''/storage/shared/research/cinn/2009/emprew/Varun_ExploreDTI/Cortical_thickness_analysis_CAT_91subjs/err/cat_check_system_output.line27.CAT.system_error/Con_P001_T1.nii'': File exists<br><br>In file "/storage/shared/research/cinn/2009/emprew/Varun_ExploreDTI/Softwares/spm12/toolbox/cat12/cat_check_system_output.m" (v1022), function "cat_check_system_output" at line 27.<br>In file "/storage/shared/research/cinn/2009/emprew/Varun_ExploreDTI/Softwares/spm12/toolbox/cat12/cat_run_newcatch.m" (???), function "cat_run_newcatch" at line 97.<br>In file "/storage/shared/research/cinn/2009/emprew/Varun_ExploreDTI/Softwares/spm12/toolbox/cat12/cat_run.m" (v1439), function "run_job" at line 720.<br>In file "/storage/shared/research/cinn/2009/emprew/Varun_ExploreDTI/Softwares/spm12/toolbox/cat12/cat_run.m" (v1439), function "cat_run" at line 434.<br><br>The following modules did not run:<br>Failed: CAT12: Segmentation<br><br>best regards,<br>Varun<br>PhD scholar in Neurosciences<br>School of Psychology and Clinical Language Sciences<br>University of Reading, Whiteknights campus,<br>United KingdomVarun Arunachalam Chandran2019-07-18T14:03:46+01:002019-07-18T14:03:46+01:00CAT12 segmentation errorhttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;61ab5e88.1907Hi everyone,<br><br>I'm facing an issue in performing cortical thickness analysis using CAT12. In the first step when I tried running the segmentation using the default pipeline, it is throwing off an error which is shown below. Could anyone please suggest me some solution? Any help on this would be very much appreciated.<br><br>thanks,<br><br>Print 'Graphics' figure to:<br>/storage/shared/research/cinn/2009/emprew/Varun_ExploreDTI/Cortical_thickness_analysis_CAT_91subjs/report/catreport_Con_P001_T1.pdf<br>Failed 'CAT12: Segmentation'<br>Error using cat_check_system_output (line 27)<br>ln: failed to create symbolic link ''/storage/shared/research/cinn/2009/emprew/Varun_ExploreDTI/Cortical_thickness_analysis_CAT_91subjs/err/cat_check_system_output.line27.CAT.system_error/Con_P001_T1.nii'': File exists<br><br>In file "/storage/shared/research/cinn/2009/emprew/Varun_ExploreDTI/Softwares/spm12/toolbox/cat12/cat_check_system_output.m" (v1022), function "cat_check_system_output" at line 27.<br>In file "/storage/shared/research/cinn/2009/emprew/Varun_ExploreDTI/Softwares/spm12/toolbox/cat12/cat_run_newcatch.m" (???), function "cat_run_newcatch" at line 97.<br>In file "/storage/shared/research/cinn/2009/emprew/Varun_ExploreDTI/Softwares/spm12/toolbox/cat12/cat_run.m" (v1439), function "run_job" at line 720.<br>In file "/storage/shared/research/cinn/2009/emprew/Varun_ExploreDTI/Softwares/spm12/toolbox/cat12/cat_run.m" (v1439), function "cat_run" at line 434.<br><br>The following modules did not run:<br>Failed: CAT12: Segmentation<br><br>best regards,<br>Varun<br>PhD scholar in Neurosciences<br>School of Psychology and Clinical Language Sciences<br>University of Reading, Whiteknights campus,<br>United KingdomVladimir Litvak2019-07-18T11:40:48+01:002019-07-18T11:40:48+01:00Re: resting state EEG datahttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;3bef9747.1907Hi Deborah,<br><br>This is an old function which was introduced in SPM8 before there was an<br>option to do the same thing in TF module. I wouldn't use it now but would<br>use the 'Spectrum' option in TF instead where you can also do multitaper.<br>The output is scalp x frequency but at the time there was no option to<br>convert this to images so as a hack the frequency axis was treated as time.<br>I don't want to confuse you with further, explanations, just use TF. There<br>you can do the log if you want with the rescale function.<br><br>Vladimir<br><br>On Thu, Jul 18, 2019 at 11:28 AM Deborah Talmi <<br>deborah.talmi@manchester.ac.uk> wrote:<br><br>> hi there,<br>><br>> we have a bit of resting state EEG data and are interested in exploring<br>> them. i've used<br>><br>> spm_eeg_ft_multitaper_powermap<br>><br>> on a converted, filtered, downsampled dataset of about 5 minutes.<br>><br>> - this worked. but what are the x/y axes on the output?<br>><br>> I then used<br>><br>> selectdata(D, 'Fz', [0.2 0.3], [], [])<br>><br>> - this worked. but what are the x/y axes on the output?<br>> - assuming the Y axis is power (or is it log(power?) I tried to look at a<br>> specified frequency range, e.g. 4-7Hz<br>><br>> selectdata(D, 'Fz', [4 7], [], [])<br>><br>> based on the help for this function which states res = selectdata(D,<br>> chanlabel, freqborders, timeborders, condition)<br>><br>> - but here I get an error, where the freqborders are assumed to be time<br>> borders. the error is<br>> Warning: Could not find an index matching the requested time 4 sec<br>> > In meeg.indsample at 21<br>> In meeg.selectdata at 55<br>><br>> hope you can help!<br>> best wishes, Deborah<br>>Deborah Talmi2019-07-18T11:28:40+01:002019-07-18T11:28:40+01:00resting state EEG datahttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;bdef8c5.1907hi there,<br><br>we have a bit of resting state EEG data and are interested in exploring them. i've used<br><br>spm_eeg_ft_multitaper_powermap<br><br>on a converted, filtered, downsampled dataset of about 5 minutes.<br><br>- this worked. but what are the x/y axes on the output?<br><br>I then used<br><br>selectdata(D, 'Fz', [0.2 0.3], [], [])<br><br>- this worked. but what are the x/y axes on the output?<br>- assuming the Y axis is power (or is it log(power?) I tried to look at a specified frequency range, e.g. 4-7Hz<br><br>selectdata(D, 'Fz', [4 7], [], [])<br><br>based on the help for this function which states res = selectdata(D, chanlabel, freqborders, timeborders, condition)<br><br>- but here I get an error, where the freqborders are assumed to be time borders. the error is<br>Warning: Could not find an index matching the requested time 4 sec<br>> In meeg.indsample at 21<br>In meeg.selectdata at 55<br><br>hope you can help!<br>best wishes, DeborahElia Valentini2019-07-18T09:30:30+01:002019-07-18T09:30:30+01:00Cross-departmental PhD bursary available at the University of Essexhttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;8fe6f415.1907Dear colleagues,<br><br>I would be very grateful if you could circulate the following call within<br>the list:<br><br>https://www.essex.ac.uk/postgraduate-research-degrees/opportunities/towards-a-novel-biomarker-for-pain<br><br>Prospective candidates can contact Dr. Elia Valentini (evalent@essex.ac.uk) for<br>informal enquiries.<br><br>Best wishes,<br><br>*Dr Elia Valentini* PhD, FHEA<br><br>Lecturer<br><br>Department of Psychology & Centre for Brain Science<br><br>University of Essex<br><br>*T* +44 (0)1206 873710<br><br>*E* evalent@essex.ac.uk<br><br>*►* https://www.essex.ac.uk/psychology/staff/profile.aspx?ID=4600<br><br>*Google Scholar:*<br>https://scholar.google.co.uk/citations?user=oLKwJFUAAAAJ&hl=en&oi=ao<br><br>*Researchgate:* https://www.researchgate.net/profile/Elia_Valentini<br><br>*Linkedin:* https://www.linkedin.com/in/eliavalentini/<br><br>[image: image.png]<br><br>[image: image.png]Onur Cezmi Mutlu2019-07-17T20:58:37+01:002019-07-17T20:58:37+01:00Discrepancy in connectivity priors for spDCMhttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;f943969c.1907Dear DCM experts,<br><br>I am currently working on fMRI data and trying to fit spectral DCM models. While exploring the effects of different priors, I noticed a discrepancy between the SPM12 priors for spDCM and 2014 spDCM paper "A DCM for resting state fMRI". In the paper, in Table 1, prior expectations for inhibitory self connections are set to ln(1/2). On the other hand, spm_dcm_fmri_priors.m sets them to 0 and spm_dcm_fmri_csd script does not further modify them to the values in the paper.<br><br>Am I missing something or should I modify the prior expectations according to the paper before fitting a spDCM?<br><br>Best regards,<br>CezmiHupfeld,Kathleen E2019-07-17T19:39:55+00:002019-07-17T19:39:55+00:00FieldMap Batch Script Problemhttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;b606604d.1907Hi all,<br><br>I have two blip up/down SE field map images (not phase + magnitude images) for fMRI data. I would like to use the FieldMap toolbox to create a vdm for each subject for later use in Realign & Unwarp.<br><br>So far:<br><br>1) Followed steps here to create a fieldmap in FSL topup: https://lcni.uoregon.edu/kb-articles/kb-0003<br><br>2) Opening FieldMap toolbox and doing this works OK to unwarp an example 1 fMRI volume I load in (or it seems):<br><br>-Precalculated-->Load, selecting my topup outputted field map<br><br>-EPI based fieldmap: no<br><br>-Polarity: -ve<br><br>-Apply Jacobian modulation: no<br><br>-Total EPI readout time: 23.1 ms<br><br>However, if I access FieldMap toolbox through batch mode and select Calculate VDM + Precalculated FieldMap (in Hz), it requires that you enter both a precalculated field map and a magnitude image in the same space as field map.<br>-->Is there a way to bypass needing to enter magnitude image here?<br><br>Alternatively, if I use the FieldMap_ngui.m script that comes with the toolbox (& for now manually load in the fpm file using LoadFieldMap, it throws an error trying to create the vdm file:<br>%----------------------------------------------------------------------<br>% Or you may want to load a precalculated Hz phase map instead...<br>%----------------------------------------------------------------------<br>[IP.fm, IP.pP] = FieldMap('LoadFieldMap');<br><br>%----------------------------------------------------------------------<br>% Create field map (in Hz) - this routine calls the unwrapping<br>%----------------------------------------------------------------------<br>IP.fm = FieldMap('CreateFieldMap',IP);<br>Struct contents reference from a non-struct array object.<br><br>Any advice here would be greatly appreciated.<br><br>Best,<br>Kathleen<br><br>Kathleen Hupfeld<br><br>PhD Candidate, Biobehavioral Science<br>Neuromotor Behavior Lab<br>Department of Applied Physiology & Kinesiology<br>University of Florida Eugenio Abela2019-07-17T19:27:56+01:002019-07-17T19:27:56+01:00Re: CAT12: get CSF maps and suppress results imagehttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;418e76c7.1907That did the trick, Simon! Thank you very much!<br><br>Bw<br><br>Eugenio<br><br>Von meinem iPhone gesendet<br><br>> Am 17.07.2019 um 17:26 schrieb <spj24@cam.ac.uk> <spj24@cam.ac.uk>:<br>><br>> It took me a while to work this out. Try putting these lines before your cat batch job<br>><br>> %Set up the desired spm if necessary<br>> addpath('/applications/spm/spm12/')<br>> spm('defaults','PET');<br>> spm_jobman('initcfg');<br>><br>> %Set CAT12 defaults<br>> cat_get_defaults('extopts.expertgui',1);<br>> cat_get_defaults('output.CSF.native',1);<br>> cat_get_defaults('output.CSF.mod',1);<br>><br>> ...<br>> Your batch job from the gui here<br>> ...<br>><br>> and this at the end of script<br>> spm_jobman('run',matlabbatch);<br>><br>> But changing the defaults files as you suggest should work as well I guess so perhaps something else is wrong (above also outputs native map to be modulated)?<br>><br>> Simon<br>><br>> -----Original Message-----<br>> From: SPM (Statistical Parametric Mapping) <SPM@JISCMAIL.AC.UK> On Behalf Of Eugenio Abela<br>> Sent: 17 July 2019 15:03<br>> To: SPM@JISCMAIL.AC.UK<br>> Subject: [SPM] CAT12: get CSF maps and suppress results image<br>><br>> Hi Chistian and others<br>><br>> I'm running CAT12 within a MATLAB batch job. The CAT12 part is preceded by fMRI preprocessing, and the anatomical data are coregistered to the fMRI before CAT12 segmentation. I want to achieve two things: get CAT12 to write CSF maps (mwp3*) and to not print any image.<br>><br>> I tried setting these options in the matlabbatch directly, i.e.<br>><br>> matlabbatch{5}.spm.tools.cat.estwrite.extopts.admin.print = 0; to suppress the figure<br>> matlabbatch{5}.spm.tools.cat.estwrite.output.CSF.mod = 1; To write modulated noramlised CSF maps<br>><br>> That didn't work; I therefore tried to change the cat_defaults.m on my system, i.e.<br>><br>> cat.extopts.print = 0;<br>> cat.output.CSF.mod = 1;<br>><br>> Also, because I read somewhere on the list that another user had done this, I also set the field "expertgui" to 1:<br>><br>> cat.extopts.expertgui = 1;<br>><br>> Again, that didn't seem to change anything, i.e. the image still got printed, and I still did not obtain a CSF map.<br>><br>><br>> How can I make this work?<br>><br>> Thank you very much!<br>><br>> Eugenio<br>> Philipp Kanske2019-07-17T19:53:20+02:002019-07-17T19:53:20+02:00PostDoc position in clinical social neurosciencehttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;e303e316.1907Dear colleagues,<br><br>We are offering a PostDoc position at the Chair of Clinical Psychology and Behavioral Neuroscience at Technische Universität Dresden.<br>https://tu-dresden.de/karriere/stellenangebote/stellenangebote-der-tud/stellenangebote?style=cms2&amp;amp;strukturId=fakpsy&amp;amp;set_language=en<br><br>I would greatly appreciate if you could forward this message to potentially interested colleagues.<br><br>All the best,<br>Philipp Kanske<>2019-07-17T17:26:26+01:002019-07-17T17:26:26+01:00Re: CAT12: get CSF maps and suppress results imagehttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;ad3c8349.1907It took me a while to work this out. Try putting these lines before your cat batch job<br><br>%Set up the desired spm if necessary<br>addpath('/applications/spm/spm12/')<br>spm('defaults','PET');<br>spm_jobman('initcfg');<br><br>%Set CAT12 defaults<br>cat_get_defaults('extopts.expertgui',1);<br>cat_get_defaults('output.CSF.native',1);<br>cat_get_defaults('output.CSF.mod',1);<br><br>...<br>Your batch job from the gui here<br>...<br><br>and this at the end of script<br>spm_jobman('run',matlabbatch);<br><br>But changing the defaults files as you suggest should work as well I guess so perhaps something else is wrong (above also outputs native map to be modulated)?<br><br>Simon<br><br>-----Original Message-----<br>From: SPM (Statistical Parametric Mapping) <SPM@JISCMAIL.AC.UK> On Behalf Of Eugenio Abela<br>Sent: 17 July 2019 15:03<br>To: SPM@JISCMAIL.AC.UK<br>Subject: [SPM] CAT12: get CSF maps and suppress results image<br><br>Hi Chistian and others<br><br>I'm running CAT12 within a MATLAB batch job. The CAT12 part is preceded by fMRI preprocessing, and the anatomical data are coregistered to the fMRI before CAT12 segmentation. I want to achieve two things: get CAT12 to write CSF maps (mwp3*) and to not print any image.<br><br>I tried setting these options in the matlabbatch directly, i.e.<br><br>matlabbatch{5}.spm.tools.cat.estwrite.extopts.admin.print = 0; to suppress the figure<br>matlabbatch{5}.spm.tools.cat.estwrite.output.CSF.mod = 1; To write modulated noramlised CSF maps<br><br>That didn't work; I therefore tried to change the cat_defaults.m on my system, i.e.<br><br>cat.extopts.print = 0;<br>cat.output.CSF.mod = 1;<br><br>Also, because I read somewhere on the list that another user had done this, I also set the field "expertgui" to 1:<br><br>cat.extopts.expertgui = 1;<br><br>Again, that didn't seem to change anything, i.e. the image still got printed, and I still did not obtain a CSF map.<br><br>How can I make this work?<br><br>Thank you very much!<br><br>EugenioMiranda Elizabeth2019-07-17T11:35:45-04:002019-07-17T11:35:45-04:00Paired-sample t-test for 2 different scan sessionshttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;d61e7ab4.1907Hi everyone,<br><br>We have data from the same subjects who performed two different (but<br>conceptually related) tasks in two different scan sessions (anywhere from a<br>few days to a few months apart). I have created a separate GLM for each<br>task and each subject. I would like to do a whole-brain paired-sample tests<br>comparing a contrast from task 1 (e.g., A > B) and a contrast from task 2<br>(e.g., C > D). So basically, [(A > B) - (C > D)]. I am unsure whether this<br>test is valid due to different scan sessions.<br><br>I would appreciate any insights on the issue!<br><br>Thank you,<br><br>M.E.SUBSCRIBE SPM YingYing Wang2019-07-17T15:56:38+01:002019-07-17T15:56:38+01:00Looking for highly motivated Postdoc Fellow for a newly funded NIH projecthttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;f6f014c4.1907Full message available at: <a href="https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;f6f014c4.1907">Looking for highly motivated Postdoc Fellow for a newly funded NIH project</a>Eugenio Abela2019-07-17T15:03:23+01:002019-07-17T15:03:23+01:00CAT12: get CSF maps and suppress results imagehttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;fc377e96.1907Hi Chistian and others<br><br>I'm running CAT12 within a MATLAB batch job. The CAT12 part is preceded by fMRI preprocessing, and the anatomical data are coregistered to the fMRI before CAT12 segmentation. I want to achieve two things: get CAT12 to write CSF maps (mwp3*) and to not print any image.<br><br>I tried setting these options in the matlabbatch directly, i.e.<br><br>matlabbatch{5}.spm.tools.cat.estwrite.extopts.admin.print = 0; to suppress the figure<br>matlabbatch{5}.spm.tools.cat.estwrite.output.CSF.mod = 1; To write modulated noramlised CSF maps<br><br>That didn't work; I therefore tried to change the cat_defaults.m on my system, i.e.<br><br>cat.extopts.print = 0;<br>cat.output.CSF.mod = 1;<br><br>Also, because I read somewhere on the list that another user had done this, I also set the field "expertgui" to 1:<br><br>cat.extopts.expertgui = 1;<br><br>Again, that didn't seem to change anything, i.e. the image still got printed, and I still did not obtain a CSF map.<br><br>How can I make this work?<br><br>Thank you very much!<br><br>EugenioAndreas Voldstad2019-07-17T14:56:49+01:002019-07-17T14:56:49+01:00Comparing PET imageshttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;ae6a2675.1907Dear SPM-users,<br><br>I would like to compare the differences between preprocessed resting state PET images based on reconstructions using different attenuation corrections:<br>3-4 different conditions (attenuation correction methods used), 2-4 sessions (images) to compare per subject.<br>I would like to enter these as contrasts in a SPM model.<br><br>I am new to SPM and neuroimaging statistics.<br>Do you have any suggestions about the correct statistical test to choose in the SPM factorial design batch, at the single-subject level and then at the group level?<br><br>I would be very grateful for your help.<br><br>Sincerely,<br><br>AndreasChristine Bastin2019-07-17T09:52:30+02:002019-07-17T09:52:30+02:00RFN2019-Call for abstracts - reminderhttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;d3d0e94b.1907In collaboration with the University of Liège, F.R.S-FNRS, and EDT<br>Neuroscience,<br><br>we are pleased to announce the 2nd edition of<br><br>Cognitive Neuroscience of Memory:<br><br>The Recollection, Familiarity, and Novelty Detection conference<br><br>Dates: 3-4 October 2019<br><br>Location: University of Liège, Liège, Belgium<br><br>Keynote speakers: Bernhard Staresina, Rik Henson, Magdalena Sauvage, Yana<br>Fandakova, Louis Renoult and Roni Tibon.<br><br>Call for abstracts: Now opened; deadline: the 2nd of August 2019.<br><br>We welcome presentations on any topic relevant to the field of memory.<br>Please submit a 250 word abstract, stating preference for oral or poster<br>presentation.<br><br>Registration (from July 1st) and full information on<br><https://events.uliege.be/rfn> https://events.uliege.be/rfn<br><br>We look forward to seeing you in Liège.<br><br>The organizing committee: Christine Bastin, Eric Salmon, Arnaud dArgembeau,<br>Laurent Lefebvre, Isabelle Simoes Loureiro, and Adrian Ivanoiu André Schmidt2019-07-17T05:42:12+00:002019-07-17T05:42:12+00:00AW: first levelhttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;7b09b96d.1907Dear SPM experts,<br><br>let me specify my question in more detail. Our participants underwent two sessions, one with placebo and the second with active treatment. Each session had two runs with 3 different experimental conditions. Including the 6 motion parameters I intend to model the first level as follows:<br><br>PLA1condition1 PLA1condition2 PLA1condition3 motion1-6 PLA2condition1 PLA2condition2 PLA2condition3 motion1-6 ACT1condition1 ACT1condition2 ACT1condition3 motion1-6 ACT2condition1 ACT2condition2 ACT2condition3 motion1-6<br><br>One of my contrast of interest is condition 2 > condition1.<br><br>To see this contrast over all 4 runs, it is right to model it like this or do I need to attach some weights?<br><br>-1 1 0 0 0 0 0 0 0 -1 1 0 0 0 0 0 0 0 -1 1 0 0 0 0 0 0 0 -1 1 0 0 0 0 0 0 0 -1 1 0 0 0 0 0 0 0 -1 1 0 0 0 0 0 0 0<br><br>I also want to see if activation for this contrast is higher in the PLA than ACT treatment condition. I can model this as such:<br><br>-1 1 0 0 0 0 0 0 0 -1 1 0 0 0 0 0 0 0 -1 1 0 0 0 0 0 0 0 1 -1 0 0 0 0 0 0 0 1 -1 0 0 0 0 0 0 0 1 -1 0 0 0 0 0 0 0<br><br>and shall I add weights also here?<br><br>Many thanks for your help and best wishes,<br><br>André Bihong Beth Chen MD2019-07-17T02:45:16+00:002019-07-17T02:45:16+00:00Re: looking for postdoc in LA areahttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;6b08ff89.1907Hi SPM list,<br><br>Our neuroimaging lab in the LA area is looking for a postdoc.<br><br>Please email us your CV if interested at bechen@coh.org<mailto:bechen@coh.org><br><br>Preferred qualifications:<br><br>1. Ph.D. in Psychology, Neuroscience, Biomedical Engineering or a related field if possible.<br>2. Training and expertise in fMRI data processing, statistical analysis, and analysis software (e.g., SPM, FSL, and AFNI)<br>3. Some programming experience in python, C++, Java or similar languages.<br>4. Some experience in tumor segmentation and machine learning.<br>5. Proficient English language skills Shengwei Zhang2019-07-16T20:10:27-05:002019-07-16T20:10:27-05:00cat12 standalone versionhttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;dd1b4a5e.1907Hi Prof. Gaser and SPM experts,<br><br>I'm trying to compile cat12 with Matlab compiler, but I noticed that in<br>cat_run.m, it calls matlab with system() command to run it in background.<br><br>Just curious if there's any other way to get around this so that a matlab<br>license is not needed? SPM12 has a standalone version that takes<br>matlabbatch jobs, although I'm not sure if it can run across multiple cores<br>without the singleThread option during compilation.<br><br>Any help would be appreciated.<br>ShengweiCPSY Events2019-07-16T16:28:05-04:002019-07-16T16:28:05-04:00Registration is Open! Next Generation Computational Psychiatry Event, October 17-18 (Chicago, IL, USA)https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;c14383eb.1907Hello all,<br><br>We are pleased to invite you to attend the Next Generation Computational<br>Psychiatry event, featuring a day on invasive Computational Psychiatry!<br><br>This 2-day event will be held in *Chicago, IL *(USA) on *October 17 & 18,<br>2019 *(just before Neuroscience 2019). This year's event will focus on two<br>big data-driven approaches to Computational Psychiatry--<br><br>Day 1: Behavioral modeling and neuroimaging across states of mental disease.<br><br>Day 2: Invasive approaches<br><br>We are honored to have Helen Mayberg<br><https://www.mountsinai.org/profiles/helen-s-mayberg> (Icahn School of<br>Medicine at Mount Sinai) and Martin Paulus<br><http://www.laureateinstitute.org/martin-paulus.html> (Laureate Institute<br>for Brain Research) as our keynote speakers.<br><br>Registration includes interactive *lectures*, *data blitz sessions*,<br>informal *networking opportunities, *and refreshments during morning and<br>afternoon breaks. The event is very interactive and offers participants a<br>taste of modern human neuroscience and its growing connection to our<br>understanding of mental disease. We welcome participants at all stages of<br>training and experience.<br><br>Confirmed presenters include: Cameron Carter, Pearl Chiu, Klaas Enno<br>Stephan, Xiaosi Gu, Talma Hendler, Ken Kishida, Ifat Levy, Carrie McAdams,<br>Read Montague and Rosalyn Moran.<br><br>Visit computationalpsychiatry.org <http://bit.ly/cpCHI2019event> for more<br>information and to register. Space is limited.<br><br>*Official Satellite Event, Neuroscience 2019, Chicago, IL, October 19-23*<br><br>We hope to see you there!<br><br>Best wishes,<br><br>Course Organizers<br><br>events@computationalpsychiatry.org<br><events@computationalpsychiatry.org?subject=2018%20Computational%20Psychiatry%20Course><br><br>*Peter Dayan, Ph.D.*<br><br>*Scientific Member & Director*<br><br>*Max Planck Institute for Biological Cybernetics*<br><br>*Xiaosi Gu, Ph.D.Director, Computational Psychiatry UnitAssistant<br>Professor, Psychiatry & NeurosciencePrinciple Investigator, Friedman Brain<br>Institute & Addiction Institute*<br><br>*Icahn School of Medicine at Mount Sinai*<br><br>*Read Montague, Ph.D.*<br><br>*Director, Computational Psychiatry Unit, Virginia Tech Carilion Research<br>Institute*<br><br>*Director, Human Neuroimaging Laboratory, Virginia Tech Carilion Research<br>Institute*<br><br>*Virginia Tech Carilion Vernon Mountcastle Research Professor*<br><br>*Professor, Virginia Tech Carilion Research Institute*<br><br>*Professor, Department of Physics, College of Science, Virginia Tech*<br><br>*Professor of Psychiatry and Behavioral Medicine, Virginia Tech Carilion<br>School of Medicine*<br><br>*Honorary Professor, Wellcome Centre for Human Neuroimaging, University<br>College London*Onur Cezmi Mutlu2019-07-16T20:11:46+01:002019-07-16T20:11:46+01:00Re: Regarding AR model order in spDCMhttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;11bfb3a9.1907Dear Dr. Razi,<br><br>Thank you very much for your response, that was very helpful. But I have two follow-up questions regarding your answer.<br><br>As far as I understand from the spm12 convention, Ep.a and Ep.b correspond to the posterior estimates of neuronal fluctuation and global channel noise spectral parameters, respectively. They are both arrays of length 2, regardless of options.order parameter. If we were to model the spectra with power law or AR(1) then two parameters would be enough to explain it. But for higher AR orders we need more.<br><br>My questions are as follows:<br>- If changing options.order creates the same effect, why doesn't it change the number of spectral parameters calculated?<br>- Unless it is set in the model definition, options.order is set to 8 by default. But regardless of its value, spm_csd_fmri_mtf.m script performs a power law modeling followed by AR modeling at the very end (last step of the script : y = spm_mar2csd(spm_csd2mar(y,M.Hz,M.p - 1),M.Hz);). What is the idea behind this two leveled modeled and what type of modeling am I performing if I use the default settings (form='1/f' and options.order=8)?<br><br>Best regards,<br>CezmiLéa Domain2019-07-16T18:52:12+02:002019-07-16T18:52:12+02:00SBM- statistical analysis with CAT12https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;1c43a298.1907Dear SPM experts,<br><br>I have a question regarding SBM analysis with the CAT12 toolbox I hope you can help with.<br><br>I’m doing statistical analysis (two-sample T-Test) on cortical thickness between 2 groups of subjects (30 depressed women (MDD), and 31 healthy controls (HC) matched for age).<br><br>As parameters in the statistical model, I defined « age » as a covariate.<br><br>After estimating Surface Model, I wanted to get some results with the ROI tool « Analyze ROIs ».<br><br>As contrats, I defined two t-contrast : for group HC > MDD (coded as 1's and -1's in design matrix) and MDD > HC (coded as 1's and -1's in design matrix).<br>Then, I set the p-value at 5% and selected apart_DK40 as atlas.<br><br>When I want to look for the results of the analysis, the only proposition made is « uncorrected » results (and it is the same for the others atlas) providing some significant results, in each hemisphere, but I would like to check corrected for multiple comparisons ones (FDR). I was wondering: if the proposition is not made does it only mean that there is no result surviving the correction ? Or is there another way to look for corrected results ?<br><br>Thanks for the help and thanks for the tutorials!<br><br>Best,<br>Léa<br><br>Psychiatry department<br>University of Rennes, FranceAndré Schmidt2019-07-16T14:07:36+00:002019-07-16T14:07:36+00:00first levelhttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;bbd2ff79.1907Dear experts,<br><br>I would glad to receive advice regarding a first level model with multiple sessions and treatments.<br><br>For each of the two treatments I have two runs with 3 conditions + 6 movement parameters. The first two runs refer to the placebo and the third and fourth to the active treatment. The contrast of interest is condition 2 > condition 1 for each run. I would like to see this effect now over all 4 runs. Is it correct in this way?<br><br>-1 1 0 0 0 0 0 0 0 -1 1 0 0 0 0 0 0 0 -1 1 0 0 0 0 0 0 0 -1 1 0 0 0 0 0 0 0 -1 1 0 0 0 0 0 0 0 -1 1 0 0 0 0 0 0 0<br><br>Or should I weight the contrast?<br><br>Many thanks for your help,<br><br>André MRI More2019-07-16T13:55:29+01:002019-07-16T13:55:29+01:00Re: AW: Realign & Unwarp - not including motion parameters as nuisance covariateshttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;23e83909.1907Dear Manuel,<br><br>Please apologize the late reply! In the Anderson 2001 Neuroimage paper, several sources (or explanations) for remaining motion-related variance are mentioned in the discussion, and they also focus a bit on certain limitations. To give a practical example, if a subject were to move by almost one voxel size along one dimension across the session (with usual cut-off values being < one voxel size), then the interpolation error should be largest somewhere during the midpoint, and becoming closer to the end of the session again, as a simple shift by one voxel should become sufficient (however, geometric distortions due to the inhomogeneities should become larger the further the subject moves, although not necessarily in a linear fashion).<br><br>Concerning Jesper's recommendations on the list, at least some of the discussion back then focused on instances with task-correlated motion. When using the RP file as predictors, the parameters would account for both, signal changes due to motion and due to brain activation. Unwarp would be an option to "account" for some of the motion-related variance in the data without having to use predictors that also account for brain activations. As long as you don't have any problems with task-correlated motion this should be irrelevant. In that case, it would boil down to the question whether unwarp makes sense if you plan to include the RP files, with the conclusion that the unwarp would be unneccessary then, but there is no strong reason to believe that it would lead to confounds (apart from losing a few df for the single-subject models).<br><br>Surprisingly, there is one trivial aspect that has not been discussed back then as far as I can see, but would be relevant in your case: Considering the default settings, unwarp only takes into account pitch and roll and adjusts the data accordingly. However, with realignment-only, one would use all six parameters as nuisance predictors in the GLM (possibly also squared versions, derivatives, composite scores). So either, there is no meaningful variance that can be explained by the other four, then there would be no need to ever use them in the GLM, or there is, at least in theory, then this should also hold for unwarped data. In other words, it should be a question whether to go with unwarp + four predictors for the translations and yaw (possibly all six due to other sources like the interpolation issue, although in that case one would probably prefer to use the adjusted ones for pitch and roll as stored in ds.P(n,1).mat - note that they are in degrees, not rad, which would only be relevant when calculating some composite score though) vs. no unwarp + all six predictors. However, as the focus seems to have been on instances with task-correlated motion back then, one would obviously not want to include any RP at all.<br><br>Hope this helps a bit & best regards<br><br>HelmutTamara Tavares2019-07-15T23:18:57-04:002019-07-15T23:18:57-04:00Cat12 Errorhttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;a09ea4e6.1907Hello,<br>I am very new to SPM and hoping to obtain GM volumes from my data to use as<br>a covariate in my functional analysis I ran in AFNI.<br><br>I am following the online manual for CAT12 and I came across an error when<br>I attempted to estimate the total intracranial volume. I read suggestions<br>online that I should re-run the segmentations...is this correct?<br><br>15-Jul-2019 20:48:22 - Failed 'Estimate TIV and global tissue volumes'<br>Reference to non-existent field 'subjectmeasures'.<br>In file "/usr/local/spm12/toolbox/cat12/cat_stat_TIV.m" (v1234), function<br>"cat_stat_TIV" at line 26.<br><br>Thank you very much for your assistance.<br><br>Best,<br>TamaraEugenio Abela2019-07-15T22:22:36+01:002019-07-15T22:22:36+01:00Re: dicom2niix issueshttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;b87e5b1a.1907Hi Lydia<br><br>sounds like something the developers might be able to answer: https://www.nitrc.org/forum/?group_id=880 <https://www.nitrc.org/forum/?group_id=880><br><br>Cheers<br><br>Eugenio<br><br>On 15 Jul 2019, at 21:19, lydia fang <lydiafang1025@GMAIL.COM> wrote:<br><br>Hello,<br><br>Recently we are working on a multi-band perfusion ASL dataset acquired from Siemens scanner. However, we had some issues while converting the DICOM data to .nii.<br><br>We used dcm2niix (latest version) to convert the data. The 4D .nii images look fine, but we got a waring:<br><br>"slices stacked despite varying acquisition numbers (if this is not desired recompile with ‘mySegmentByAcq’")<br><br>Has anyone else had the same issue before? Should we worry about this warning?<br><br>Another issue is if we use the dcm2niix syntax, there is no slice timing information at all. But if using GUI, the slice timing is 0.<br><br>These issues make us a little worried about the images per se, although the images look normal, but we still hope to make it clear, especially the slice timing information.<br><br>Thank you in advance.<br>Lydialydia fang2019-07-15T16:19:56-04:002019-07-15T16:19:56-04:00dicom2niix issueshttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;6694eacb.1907Hello,<br><br>Recently we are working on a multi-band perfusion ASL dataset acquired from<br>Siemens scanner. However, we had some issues while converting the DICOM<br>data to .nii.<br><br>We used dcm2niix (latest version) to convert the data. The 4D .nii images<br>look fine, but we got a waring:<br><br>"slices stacked despite varying acquisition numbers (if this is not desired<br>recompile with ‘mySegmentByAcq’")<br><br>Has anyone else had the same issue before? Should we worry about this<br>warning?<br><br>Another issue is if we use the dcm2niix syntax, there is no slice timing<br>information at all. But if using GUI, the slice timing is 0.<br><br>These issues make us a little worried about the images per se, although the<br>images look normal, but we still hope to make it clear, especially the<br>slice timing information.<br><br>Thank you in advance.<br>LydiaPo-Yu Fong2019-07-15T18:55:57+01:002019-07-15T18:55:57+01:00Quesitons about source reconstructionhttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;2fc22ed0.1907Dear SPM experts<br>I have some questions about source reconstruction of EEG.<br>1. In a ERP study, can I merge EEG data of all trials from all subjects into a single mat file and average all trials, then do source reconstruction for an averaged source localization? Is it reasonable? Or we can only look at the source reconstruction one subject by one subject?<br><br>2. When I get a mesh image of source reconstruction, what is the unit of the scale of colorbar? And what is the unit of the 'T' in a data cursor? (For example, when I use the function of data cursor in Matlab to point the reddest point on the mesh and then a small table, displaying positions on X, Y, Z axis and a value named 'T', shows up in right-lower corner in the window. What is this T?).<br><br>3. When I invert EEG model to mip figure, is it necessary to cover the time period before event like -100ms or -200ms, and then to choose a interested time window after mip figure showed up? Or I can just cut the time period like 50-150ms to make a mip figure and use the same time period to get a source image on MRI template or mesh?<br><br>4. When doing group statistics on EEG source reconstruction, is it absolutely needed to use FWE p<0.05 to avoid false positive result? If some areas, known as reasonable regions based in previous studies, were significance by using uncorrected p value at 0.05, but were not significant by corrected P value or FWE. Could this be a kind of false negative result?<br><br>5. What is the exactly physical meaning of the "source"? Does it mean a generator responding to whole brain activity in a specific time period, or an area with biggest amplitude in a time period, or a node in a neural circuit?<br><br>Thank you very much<br><br>with best wishes<br>Po-YuJi Hyun Ko2019-07-15T17:44:28+00:002019-07-15T17:44:28+00:00postdoc on fMRI and HD-tDCShttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;85430681.1907POST-DOCTORAL FELLOWSHIP IN BRAIN IMAGING & BRAIN STIMULATION<br>Ko Lab at University of Manitoba is seeking a full-time post-doctoral fellow who will lead imaging-guided brain stimulation studies. Multiple projects are on-going in Ko Lab (https://www.kolabneuro.com/). The candidate will analyze the already collected brain imaging data (fMRI, DTI, PET) and conduct HD-tDCS studies.<br><br>Requirements<br>The applicant should possess a PhD in Neuroscience or a related discipline. Preference will be given toward the individuals with 1) experiences in data analysis of neuroimaging data: resting-state fMRI and/or DTI (graph theory application is preferred), 2) experience in the neuropsychological assessment, and/or 3) experiences in transcranial direct current stimulation (tDCS) and/or HD-tDCS applications.<br>The position is initially for 1 year, and it may be extended subject to competency and funding availability. The position may start as soon as possible but the start date is negotiable.<br><br>Interested applicants should submit the following application materials via e-mail (ji.ko@umanitoba.ca<mailto:ji.ko@umanitoba.ca>).<br><br>(i) A 1-2 page personal statement or cover letter outlining professional goals, research interests, and relevant background experience.<br><br>(ii) A full curriculum vitae with a list of publications<br><br>(iii) Work sample, such as a published manuscript on which you are first author or other written product that highlights your work relevant to the brain imaging and/or PTSD emphasis areas.<br><br>(iv) The names and contact information of at least two references.<br><br>Benefits<br>Salary: Total funding will be based on experience. Minimum funding for this position is 35,000 CAD/year (+benefits)<br>For an informal discussion about the post, please contact Dr. Ji Hyun Ko (ji.ko@umanitoba.ca<mailto:ji.ko@umanitoba.ca> or +1-204-318-2566).<br><br>Best,<br><br>Ji Hyun<br><br>Ji Hyun Ko, PhD<br>Assistant Professor<br>Department of Human Anatomy and Cell Science<br>Max Rady College of Medicine, Rady Faculty of Health Sciences<br>University of Manitoba<br><br>Principal Investigator<br>Neuroscience Research Program<br>Kleysen Institute for Advanced Medicine<br>Health Science Centre<br><br>Mailing Address:<br>SR452 - 710 William Avenue<br>Winnipeg, MB R3E 0Z3<br>Canada<br><br>Tel: +1-204-318-2566<br>Email: ji.ko@umanitoba.ca<mailto:ji.ko@umanitoba.ca><br><br>http://www.kolabneuro.com/ Nicolas Villain2019-07-15T19:32:29+02:002019-07-15T19:32:29+02:00Job Offer : "Engineer / Software developer: Analysis of multimodal longitudinal neuroimaging data in Alzheimer’s disease"https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;41342201.1907Dear colleagues,<br><br>You will find enclosed a Job Offer for an Engineer / Software developer<br>position in Paris, France regarding a project on "Analysis of multimodal<br>longitudinal neuroimaging data in Alzheimer’s disease".<br>Please let us know your interest.<br><br>Best regards,<br><br>Nicolas Villain, MD, PhD<br>Fellow in Cognitive and Behavioural Neurology<br>Institute of Memory and Alzheimer's Disease<br>Department of Neurology<br>Pitié-Salpêtrière University Hospital<br>Pars, France<br><br>GRC n°21 Alzheimer Precision Medicine<br>Institute of Spine and Brain<br>Pitié-Salpêtrière University Hospital<br>Pars, FranceChristian N.2019-07-15T16:43:17+01:002019-07-15T16:43:17+01:00Re: CAT12 : batch estimate surface modelshttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;960c5454.1907Sorry for the duplicate posting, but I'm not sure whether the original post is displayed in my message.<br><br>The original message was this:<br><br>"Dear Christian,<br><br>Is there any way of launching "estimate surface models" from batch commands ?<br><br>Best regards,<br>Matthieu"<br><br>Thank you.Christian N.2019-07-15T16:35:47+01:002019-07-15T16:35:47+01:00Re: CAT12 : batch estimate surface modelshttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;e80c6b6e.1907Dear all,<br><br>I know this is an old topic, but does anyone know the answer to this question?<br><br>Thanks.Maral Ye2019-07-15T10:46:23-04:002019-07-15T10:46:23-04:00unsubscribe pleasehttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;545fc621.1907Hi,<br><br>Would you please unsubscribe me from this list?<br><br>Thank you,<br><br>Maral YeganehOyetunde Gbadeyan2019-07-15T14:33:59+00:002019-07-15T14:33:59+00:00MRI Technologist position available at The Ohio State University.https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;3958c28c.1907Magnetic Resonance Imaging (MRI) Technologist – OHIO STATE UNIVERSITY.<br><br>The Center for Cognitive and Behavioral Brain Imaging<https://ccbbi.osu.edu/> (CCBBI), in the College of Arts and Sciences, seeks an MRI Technologist to join our team. The College of Arts and Sciences is the largest college and the academic heart of the university. The College hosts 81 majors. With 38 departments, 20+ world-class research centers, and more than 2,000 faculty and staff members, students have the unique opportunity to study with the best artists, scholars, and scientists in their field. The College values diversity and offers a supportive, open, and inclusive community.<br><br>The CCBBI is directed by Dr. Ruchika Prakash and the center is dedicated to the study of brain mechanisms underlying individuals’ cognitive capacity and subjective well-being, as well as dysfunctions of these brain mechanisms in normal aging and mental disorders. It aims to contribute to the development of future brain imaging modalities and to create and disseminate knowledge about brain, mind, and imaging research.<br><br>Successful candidates must have earned a bachelor’s degree in radiology, biomedical engineering, or physics, or a related field or an equivalent combination of education and experience, with a certification as an MRI Technologist; experience working in a clinical or research MRI environment, including knowledge of proper safety screening and scanning procedures. Additionally, American Registry of Radiologic Technologists (ARRT) Radiography certification and Ohio Department of Health radiology license is desired.<br><br>If you are interested in being considered for the position, review the full job description and qualifications and apply online at https://www.jobsatosu.com, under job listing #451397 and submit a cover letter detailing background and experiences, a CV, the expected date of availability, and the names and contact of three references.<br><br>If you have questions about the position or the center, please contact Dr. Ruchika Prakash at prakash.30@osu.edu<br><br>CCBBI Website: https://ccbbi.osu.edu/<br><br>To build a diverse workforce, The Ohio State University encourages applications from all individuals. Ohio State is an Equal Opportunity/Affirmative Action Employer. All qualified applicants will receive consideration for employment without regard to age, race, color, religion, sex, sexual orientation, gender identity, national origin, disability status, or protected veteran status.David Hofmann2019-07-15T12:10:08+02:002019-07-15T12:10:08+02:00Fwd: DCM questionshttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;55f063c4.1907Hi Peter,<br><br>I have a few questions and hope you don't mind:<br><br>1. I calculated two DCMs. One with 2 ROIs but uncoupled, another with one<br>of the same ROI as the DCM before and compared the results. I looked at the<br>self-connections. I noticed that the parameters had different values for<br>the ROI that was the same in both DCM. This is strange since in the first<br>DCM the ROI was the same as in the second. But I expected the values of the<br>self-connections to be the same in both DCMs. Do you have any idea how<br>these differences arise?<br><br>2. Is it possible to replace the DCM bilinear model by any other<br>(nonlinear) model? As I understood the parameter estimation algorithm can<br>be applied to any nonlinear system, however, I do not know where the<br>"interface" is, in order to replace the equations.<br><br>all the best,<br><br>DavidRoberto Viviani2019-07-13T16:04:33+02:002019-07-13T16:04:33+02:00Tenure-track position at the University of Innsbruckhttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;1ce7fad4.1907The University of Innsbruck is seeking a scientist for a tenure-tack<br>position in the computational neurosciences, with a focus on<br>functional or structural imaging approaches. This position may be of<br>interest to a neuroimager in the early stages of her career, who would<br>like to complement the post-doc stage with an independent position<br>that, after confirmation, leads to a permanent employment at professor<br>level.<br><br>The position will be partly located in a newly-founded<br>interdisciplinary research centre in the digital humanities, thus<br>giving the opportunity to contribute to shaping its development.<br><br>Teaching activities optionally in English in master courses.<br><br>For more details, see https://euraxess.ec.europa.eu/jobs/416967 Munir Elias2019-07-12T20:49:28+01:002019-07-12T20:49:28+01:00Dartel with VBMhttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;d2de0b75.1907Dear Administrator/Guillaume,<br>Can I have data to the subject Dartel VBM because it is not available and as I saw that, your guill_vb_dell_scr_1366x768_hp.m belonging to you.<br>Need your help to proceed with tutorials.<br>Thanks in advance.<br>Munir EliasAdeel Razi2019-07-13T00:25:01+10:002019-07-13T00:25:01+10:00Re: Regarding AR model order in spDCMhttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;4ddd06ac.1907Dear Cezmi,<br><br>I would use the first approach as we now have the possibility to assign the<br>order of AR model during the DCM specification (using DCM.options.order) as<br>you pointed out.<br><br>The seond approach is a long-handed equivalent of the first. See this paper<br>which followed the second approach here:<br>https://www.omicsonline.org/open-access/altered-structuralfunctional-maturation-of-the-right-amygdala-in-healthyadolescents-exposed-to-traumatic-events-2375-4494-1000248.pdf<br><br>Best wishes,<br>Adeel<br><br>On Fri, Jul 12, 2019 at 9:39 AM Onur Cezmi Mutlu <cezmi@stanford.edu> wrote:<br><br>> Dear DCM experts,<br>><br>> I have a question regarding autoregressive model orders in spectral DCM.<br>><br>> I am trying to fit spectral DCM to fMRI data and want to try different AR<br>> model orders to monitor the effect of changing model complexity. I'm<br>> planning the build 4 models with AR orders 1 to 4 for neuronal fluctuations<br>> and observation noise. For this purpose I tried to set my models<br>> accordingly but came across with two different AR modeling sections in the<br>> spm codebase that might make the changes that I want. I couldn't decide on<br>> which one to follow and wanted to seek guidance from SPM community.<br>><br>> My first approach was to set DCM.options.order parameter to the AR model<br>> order that I want, since it effects the last line of spm_csd_fmri_mtf.m<br>> script and used as an input to spm_csd2mar function.<br>><br>> My second approach was to play with the "spectrum of neuronal fluctuations<br>> (Gu) and observation noise (Gn)" section of spm_csd_fmri_mtf.m script. I<br>> first changed the form variable to something else other than "1/f" so that<br>> it would make AR modeling and then I changed<br>><br>> G = spm_mar2csd(exp(P.a(2,1)),w); line in "neuronal fluctuations (Gu)<br>> (1/f or AR(1) form)" section to G = spm_mar2csd(exp(P.a(2:end,1)),w);<br>><br>> and<br>><br>> G = spm_mar2csd(exp(P.b(2,1))/2,w); lines in "region specific<br>> observation noise (1/f or AR(1) form)" and "global components" sections to<br>> G = spm_mar2csd(exp(P.b(2:end,1))/2,w);<br>><br>> I also modified "add prior on spectral density of fluctuations (amplitude<br>> and exponent)" section of spm_dcm_fmri_priors.m script so that I would have<br>> enough number of coefficients for each type of model. (I changed pE.a =<br>> sparse(2,1); to pE.a = sparse(p,1); and so on where p is the AR model<br>> order I want)<br>><br>> Which of the approaches is the correct way to do what I have in mind? And<br>> if one of them is correct what is the effect of the other one?<br>><br>> Thanks in advance,<br>><br>> Cezmi<br>>Fanghella, Martina2019-07-12T14:20:56+00:002019-07-12T14:20:56+00:00CoSAN PhD call 2019https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;2cca394e.1907Dear all,<br><br>I hope this email finds you well.<br><br>The international Sapienza (Rome) CoSAN PhD call is open and the deadline for the application is July 18th.<br><br>Most of the positions are in Cognitive, Social and Affective Neuroscience.<br>Importantly, however, at least one of the positions is funded by the Italian Institute of Technology and is dedicated to applicants with a degree in Biomedical Engineering, Electronic Engineering, Physics, Informatics, Computer graphics.<br><br>I would be most grateful if you may circulate this message among prospective students who are invited to contact Prof. Aglioti (salvatoremaria.aglioti@uniroma1.it<mailto:salvatoremaria.aglioti@uniroma1.it>) for details on this specific issue or any other scientific aspects of the call. For any requests concerning the logistics feel free to contact our lab manager, Dr Michela La Padula (michela.lapadula@uniroma1.it<mailto:michela.lapadula@uniroma1.it>).<br><br>How to apply: https://phd.uniroma1.it/dottorati/cartellaDocumentiWeb/18aff3e7-f327-4acb-a4b2-a43046a2504e.pdf<br>For details on the call visit: https://phd.uniroma1.it/web/pagina.aspx?s=&i=3533&m=&l=EN&p=48&a=<br><br>I thank you very much and send my warmest wishes,<br><br>Martina<br><br>Martina Fanghella, Ph.D. Student | Joint PhD program in Social Neuroscience<br>Cognitive Neuroscience Research Unit | CNRU<http://www.city.ac.uk/arts-social-sciences/psychology/research/cognitive-neuroscience-research-unit> - Psychology Department - City, University of London<br>Agliotilab | Department of Psychology - La Sapienza, University of Rome<br>Tel: +393460920429 | Email: martina.fanghella@city.ac.uk<mailto:martina.fanghella@city.ac.uk> - martina.fanghella@uniroma1.it<mailto:martina.fanghella@uniroma1.it><br>https://agliotilab.org/lab-staff/phd-students/1st-year/martina-fanghella#anchor Alistair Perry2019-07-12T15:43:41+02:002019-07-12T15:43:41+02:00disproportionate shrunken parameter estimates after DCM PEBhttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;e01bc877.1907Hi Peter / SPM users,<br><br>I'm running the DCM PEB framework on fMRI data involving attentional<br>modulations during a random dot-motion task. I've identified strong<br>non-zero associations between certain connectivity parameters and<br>second-level effects of interest (i.e. interaction, group effects,<br>behavioral associations).<br><br>However, I'm having difficulty interpreting the nature of these effects, as<br>the shrunken connectivity estimates (after dcm_peb) are well, rather<br>erroneous..<br><br>For instance, for a certain subject after group-level shrinking has<br>posterior means (i.e. Ep.B) at values around *70-100*.. But, the raw<br>estimates of the fully-connected DCM model for this subject range between<br>-1 and 1. What even more telling is that for the behavioral associations<br>identified in the PEB model, I observe similar relationships (and<br>frequentisticly significant) in the raw estimates, but not the shrunken<br>parameters.<br><br>I construct the values using the following commands:<br><br>M = struct();<br>M.alpha = 1;<br>M.beta = 16;<br>M.hE = 0;<br>M.hC = 1/16;<br>M.Q = 'all';<br>M.X(1:46,1) = ones(46,1);<br>field = {'B'};<br>[PEB, rGCM] = spm_dcm_peb(GCM,M,field)<br><br>I'm not sure whether these issues can be attributable to model complexity<br>or noisy data, as the space comprises a fully-connected model involving 10<br>regions. I do have plans to trim down the number of circuit-regions.<br><br>Best,<br><br>AlistairKai-Hsiang Chuang2019-07-12T15:29:14+10:002019-07-12T15:29:14+10:00Postdoc on fMRI, calcium and electrophysiologyhttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;54e71f32.1907A postdoctoral research fellow position is immediately available at<br>the Queensland Brain Institute, The University of Queensland,<br>Brisbane, Australia on developing technologies for functional imaging<br>of the brain.<br><br>The project<br><br>Large-scale synchronous network has been identified in mammalian brain<br>at the resting state which implicates the spontaneous organization of<br>the brain. However the neural basis and function of these<br>resting-state networks are largely unknown. The lab focuses on<br>understanding the neural basis of resting-state network of the brain<br>and its roles in learning, memory and dementia, and the translation<br>between animal models and humans. This project will develop advanced<br>imaging techniques that combines functional Magnetic Resonance Imaging<br>(MRI), high-density fibre photometry of calcium recording and<br>electrophysiology in rodent models on a 9.4T MRI to investigate the<br>mechanisms of the functional connectivity and its dynamics. The<br>results will be validated by axonal tract tracing, dendritic spine<br>imaging and optogenetic/chemogenetic interventions. This will require<br>extensive system design, programming, data processing and in vivo<br>imaging on ultrahigh field 9.4T animal MRI and two-photon microscopy.<br><br>The candidate<br><br>· PhD in the area of neuroscience, biomedical engineering,<br>physics, chemistry.<br>· Demonstrated expert knowledge in the area of MRI pulse<br>sequence design/programming, in vivo calcium<br>imaging/electrophysiology.<br>· Experience in functional MRI, MRI data analysis using SPM or<br>FSL; programming language (Matlab or C/C++ and shell script).<br>· Experience with animal fMRI or simultaneous MRI/EEG/calcium<br>recording are desirable but not required.<br><br>Application<br><br>Interested candidates should submit their application online:<br>http://search.jobs.uq.edu.au/caw/en/job/508003/postdoctoral-research-fellow<br><br>Any question can be addressed to A/Prof Kai-Hsiang Chuang <k.chuang@uq.edu.au>.<br><br>The University of Queensland is committed to equity, diversity and inclusion.Onur Cezmi Mutlu2019-07-12T00:28:21+01:002019-07-12T00:28:21+01:00Regarding AR model order in spDCMhttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;b2834bd7.1907Dear DCM experts,<br><br>I have a question regarding autoregressive model orders in spectral DCM.<br><br>I am trying to fit spectral DCM to fMRI data and want to try different AR model orders to monitor the effect of changing model complexity. I'm planning the build 4 models with AR orders 1 to 4 for neuronal fluctuations and observation noise. For this purpose I tried to set my models accordingly but came across with two different AR modeling sections in the spm codebase that might make the changes that I want. I couldn't decide on which one to follow and wanted to seek guidance from SPM community.<br><br>My first approach was to set DCM.options.order parameter to the AR model order that I want, since it effects the last line of spm_csd_fmri_mtf.m script and used as an input to spm_csd2mar function.<br><br>My second approach was to play with the "spectrum of neuronal fluctuations (Gu) and observation noise (Gn)" section of spm_csd_fmri_mtf.m script. I first changed the form variable to something else other than "1/f" so that it would make AR modeling and then I changed<br><br>G = spm_mar2csd(exp(P.a(2,1)),w); line in "neuronal fluctuations (Gu) (1/f or AR(1) form)" section to G = spm_mar2csd(exp(P.a(2:end,1)),w);<br><br>and<br><br>G = spm_mar2csd(exp(P.b(2,1))/2,w); lines in "region specific observation noise (1/f or AR(1) form)" and "global components" sections to G = spm_mar2csd(exp(P.b(2:end,1))/2,w);<br><br>I also modified "add prior on spectral density of fluctuations (amplitude and exponent)" section of spm_dcm_fmri_priors.m script so that I would have enough number of coefficients for each type of model. (I changed pE.a = sparse(2,1); to pE.a = sparse(p,1); and so on where p is the AR model order I want)<br><br>Which of the approaches is the correct way to do what I have in mind? And if one of them is correct what is the effect of the other one?<br><br>Thanks in advance,<br><br>CezmiChannelle Tham2019-07-11T22:03:21+01:002019-07-11T22:03:21+01:00Flexible Factorial Design on CAT12https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;ab825eb7.1907Hello Experts,<br><br>I am attempting to perform a longitudinal analysis of 1 patient over 2 time points. So far, I've followed the flexible factorial design based on the instructions from the CAT12 manual (pg 33). I would like to get statistics of the thickness loss/gain overtime. I've created and estimated the design, but viewing the results is where there's an issue. Below is the error message I receive:<br><br>SPM12: spm_results_ui (v7204) 16:54:28 - 11/07/2019Marina Bedny2019-07-11T17:50:19+01:002019-07-11T17:50:19+01:00Post-doc position at Neuroplasticity & Development Lab Johns Hopkins Universityhttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;86ced983.1907Post-doc position at Neuroplasticity & Development Lab Johns Hopkins University, PI: Marina Bedny<br>The Neuroplasticity & Development Lab invites applicants for a post-doctoral fellowship position. The postdoctoral fellow will develop a research program investigating how experience shapes human cognitive and neural development and plasticity. Our laboratory compares cognition in adults and children with different development experiences using techniques such as fMRI, TMS and behavioral measures. A key research direction compares neurocognitive function among individuals who are blind and sighted. We investigate visual cortex function in blindness as a window into mechanisms of brain development and plasticity in humans. We also study how sensory and linguistic experience contributes to concepts and their neural basis (e.g. neural basis of concrete entity and event representations in blind and sighted individuals). For more information about our research go to http://bednylab.com/<br><br>The post-doctoral fellow will be encouraged to take part in the intellectually rich cognitive science and neuroscience communities of Johns Hopkins University.<br><br>Qualifications: A Ph.D. in Psychology, Cognitive Neuroscience or related field is required. Qualified applicants will have a strong interest in neuroplasticity and cognition. Expertise in fMRI is preferred but candidates with proficiency in other relevant domains/techniques will also be considered (e.g. developmental psychology, psycholinguistics, TMS/EEG/DTI). Excellent computational skills are strongly preferred.<br><br>How to Apply: Candidates should submit a letter of interest, a CV and contact information for three references by emailing the lab director at marina.bedny@jhu.edu. The letter of interest should be no more than a single page long and should include a statement of what research the applicant wishes to pursue during their post-doctoral training and why they believe this job is a good fit to their specific research interests and career goals. Applications will be evaluated on an ongoing basis until the position is filled. The starting date is flexible and available candidates could begin work immediately.Jeff Browndyke2019-07-11T11:37:00-04:002019-07-11T11:37:00-04:00Re: [CAT12] Description of longitudinal segmentation pipelinehttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;e51316a8.1907Just to kick this up the queue, I, too, would like to know more about the most recent longitudinal segmentation pipeline in CAT12.<br><br>Are there recent papers out there that better describe what is going on “under the hood?”<br><br>Thanks,<br>Jeff Browndyke<br><br>> On Jul 8, 2019, at 4:37 AM, Paul Zhutovsky <paul.zhutovsky@GMAIL.COM> wrote:<br>><br>> Dear all,<br>><br>> I used CAT12's (r1447) longitudinal segmentation pipeline and I would like to get a basic intuition about how it is computed. I've read the manual but the recent toolbox update (r1444) indicates that the procedure for performing the segmentation has changed.<br>> I am therefore wondering if the description of the longitudinal segmentation provided in the manual (page 30 onward) is out-of-date and if a more detailed updated description could be provided?<br>><br>> Thank you for your help!<br>> Paulyael jacob2019-07-11T10:31:55-04:002019-07-11T10:31:55-04:00Postdoctoral fellow positions at Icahn School of Medicine at Mount Sinai, New Yorkhttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;45c3ff86.1907The Translational and Molecular Imaging Institute (TMII) at The Icahn<br>School of Medicine at Mount Sinai New York currently offers two<br>Postdoctoral scholar positions.<br>Please see attached announcements.Feng Xu2019-07-11T10:10:52-04:002019-07-11T10:10:52-04:00Re: Does SPM handle complex value images?https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;ee701bc8.1907Thank you, Dr. Ashburner and Dr. Lund for your kind response.<br><br>Best regards,<br>Susan<br><br>On Thu, Jul 11, 2019 at 9:57 AM Torben Lund <torbenelund@me.com> wrote:<br><br>> As a compromise, with a bit of programming, using spm functions and<br>> matlab, if you store Magnitude an Phase in diffrent images, you could:<br>><br>> 1. Create the corresponding set of Real and Imaginary images<br>> 2. Calculate the realignment parameters based on the magnitude images<br>> 3. Apply the transformaions to the Real and Imaginary images, and reslice<br>> the images<br>> 4. Create the corresponding Magnitude and Phase images<br>><br>> Possibly you could also skip the conversion steps between Magnitude/Phase<br>> Real/Imaginary, but I am not sure interpolation of the phase images is well<br>> behaved.<br>><br>> Best<br>> Torben<br>><br>><br>><br>><br>> > Den 9. jul. 2019 kl. 21.26 skrev Feng Xu <susanfengxu@GMAIL.COM>:<br>> ><br>> > Dear SPM users,<br>> > I am wondering if SPM handle images that have values in complex form. I<br>> need to do motion correction and registration. I am not sure if how to save<br>> complex image in .nii file either.<br>> ><br>> > Thanks,<br>> > Susan<br>><br>>Torben Lund2019-07-11T15:56:56+02:002019-07-11T15:56:56+02:00Re: Does SPM handle complex value images?https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;3ec3efe8.1907As a compromise, with a bit of programming, using spm functions and matlab, if you store Magnitude an Phase in diffrent images, you could:<br><br>1. Create the corresponding set of Real and Imaginary images<br>2. Calculate the realignment parameters based on the magnitude images<br>3. Apply the transformaions to the Real and Imaginary images, and reslice the images<br>4. Create the corresponding Magnitude and Phase images<br><br>Possibly you could also skip the conversion steps between Magnitude/Phase Real/Imaginary, but I am not sure interpolation of the phase images is well behaved.<br><br>Best<br>Torben<br><br>> Den 9. jul. 2019 kl. 21.26 skrev Feng Xu <susanfengxu@GMAIL.COM>:<br>><br>> Dear SPM users,<br>> I am wondering if SPM handle images that have values in complex form. I need to do motion correction and registration. I am not sure if how to save complex image in .nii file either.<br>><br>> Thanks,<br>> Susan Sophie Wilcox2019-07-11T13:41:25+01:002019-07-11T13:41:25+01:00Postdoctoral position at GIGA-neuroscience in fMRI and neuroendocrinologyhttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;53f10c33.1907*** Applications are still open for this position. For any further questions or to apply please contact Dr. Julie Bakker (jbakker@uliege.be) ***<br><br>A position is open for a post-doctoral research fellow within the Laboratory of Neuroendocrinology (Head PI Dr. Julie Bakker) at the GIGA Neurosciences, University of Liège.<br><br>The lab has a strong research focus on understanding of how gonadal hormones induce sex differences in the brain and behavior during fetal development but also how they further affect brain structure and functioning during puberty and later in adulthood. Current projects include a multimodal (structural and functional) imaging study investigating unravel the role of pubertal gonadal hormones in brain sexual differentiation using Kallmann Syndrome as model. Other projects include the potential role of pheromones in influencing cognition and mood via limbic pathways and pharmacological interventions in women with a disorder of low sexual desire.<br><br>We are seeking a candidate with a PhD in a relevant field (neuroscience, neuropsychology, etc.) to join our lively and dynamic research group. The successful candidate will be highly motivated and with a strong background in neuroimaging. The candidate is expected to have an interest in behavioral neuroscience and neuroendocrinology, preferably reflected in their research background. Proficiency in all stages of MRI research, including study design and management, recruitment, acquisition and analysis is required. Experience in multiply MRI modalities (structural, task and resting-state fMRI, DTI) is considered a plus. The successful candidate will have demonstrated academic productivity in the form of peer reviewed scientific publications, presentations at scientific congresses and a proven record of grant/fellowship awards.<br><br>The research fellow will work in a collaborative, multi-disciplinary environment at the GIGA, a major centre for research and development in biotechnology and is one of a very few centers in Europe that have excelled at integrating academic research, collaborations with companies, technology transfer and training facilities. The GIGA Neurosciences unit carries out top-level research on the development, functioning and disorders of the nervous system. The MRI acquisitions will take place at the Cyclotron Research Center (CRC), a dedicated research unit within the GIGA (GIGA-CRC in vivo imaging).<br><br>The position is temporary for a period of 1 year (commencing October 2019 - October 2020) with the potential for renewal depending on grant/fellowship outcomes.<br><br>Candidates should send their application, including a curriculum vitae, or requests for additional information to Dr. Julie Bakker via email at jbakker@uliege.beXiaowei Song2019-07-10T22:48:55-04:002019-07-10T22:48:55-04:00Re: Inference on GLM parameters when design matrix is derived from datahttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;bd7689dd.1907Dear Simon,<br><br>> For the subsequent inference, my approach was to fix the X that was found<br>> from this matrix factorization problem, and then solve the GLM (i.e. linear<br>> regression) N times, i.e. for N reshufflings of the data, to find the null<br>> distribution of the parameters. From that, the percentile of the original<br>> parameters indicates their significance. (Alternatively, I could<br>> re-estimate both X and S for every reshuffle, but the factorization is<br>> quite computationally intensive due to several aspects that I'll omit here<br>> for brevity, so this would be prohibitive.)<br><br>I did the permutation tests as you are doing, we are same on how to do the<br>inference. We also found significant correlation of columns in X with some<br>behavioral measures as shown in the ISBI 2018 conference paper, also like<br>you did.<br>One slight difference is that we picked one stable run of ICA among many<br>runs.<br><br>Best,<br>Xiaowei<br><br>On Wed, Jul 10, 2019 at 4:16 PM Simon Van Eyndhoven <<br>Simon.VanEyndhoven@esat.kuleuven.be> wrote:<br><br>> Dear Xiaowei,<br>><br>> Thanks for the clarification. I agree that for the unshuffled data, the<br>> beta parameters are taken from S, indeed (I used both the notation with<br>> Beta to draw the connection with the GLM, and the notation with S to make<br>> the connection with ICA-style methods).<br>><br>> For the subsequent inference, my approach was to fix the X that was found<br>> from this matrix factorization problem, and then solve the GLM (i.e. linear<br>> regression) N times, i.e. for N reshufflings of the data, to find the null<br>> distribution of the parameters. From that, the percentile of the original<br>> parameters indicates their significance. (Alternatively, I could<br>> re-estimate both X and S for every reshuffle, but the factorization is<br>> quite computationally intensive due to several aspects that I'll omit here<br>> for brevity, so this would be prohibitive.)<br>><br>> However, from what I understand, you have performed inference differently?<br>> Did you also aim to determine the significance of the coefficients / scores<br>> / loadings / parameters / ... (whichever nomenclature fits the<br>> interpretation) in S, or was your objective different (e.g. correlation of<br>> the columns in X or the rows of S with a behavioral measure)? I am eager to<br>> know more about how you performed it concretely, if you are willing to go<br>> in more depth! :)<br>><br>> All the best, and thanks for your help so far,<br>><br>> Simon<br>><br>> On 2019-07-10 19:01, Xiaowei Song wrote:<br>><br>> From your "3. Problem statement", the "parameters Beta" you mentioned,<br>> is actually S in "Y = X*S + residual" if the design matrix "X" is given<br>> in GLM. Do you agree?<br>> In ICA methods, with some abuse of notation, X is the optimization<br>> variable instead of being a user-defined design-matrix in GLM. The ICA cost<br>> function is usually derived using maximum likelihood estimation, which has<br>> many desired properties (For example, Lecture 8: Properties of Maximum<br>> Likelihood Estimation (MLE)<br>> <https://engineering.purdue.edu/ChanGroup/ECE645Notes/StudentLecture08.pdf><br>> ). I don't think I am a victim to the fallacy you mentioned.<br>> I guess you have the point our paper tries to make: "these parameters<br>> correspond to regressors (in X) that were* themselves tuned to explain<br>> the data well*. ".<br>> By the way, the comparison was done in a journal version.<br>> I hope I answered your question.<br>><br>><br>><br>><br>> On Wed, Jul 10, 2019 at 12:19 PM Simon Van Eyndhoven <<br>> Simon.VanEyndhoven@esat.kuleuven.be> wrote:<br>><br>>> Dear Xiaowei,<br>>><br>>> Thanks for your reply! Do you mean that you fell victim to the same<br>>> fallacy, i.e. that the higher sensitivity was due to the self-fulfilling<br>>> prophecy of a data-derived design matrix, or do you mean that the higher<br>>> sensitivity was unbiased and therefore good news?<br>>><br>>> PS: I consulted the paper, but I don't seem to find the comparison with<br>>> permutation analysis of linear models (PALM)?<br>>><br>>> Best regards,<br>>><br>>> Simon<br>>><br>>> On 2019-07-10 17:20, Xiaowei Song wrote:<br>>><br>>> Dear Simon,<br>>><br>>> Is this intuition correct, and if so, what can be done to adjust the<br>>>> inference procedure, and perform unbiased statistical tests in such case?<br>>>> In my understanding, the problem is 'performing inference on data-driven<br>>>> factor models', and I've looked for literature on inference on e.g.<br>>>> ICA-derived time courses (the same problem would also occur in that case)<br>>>> but have found nothing relevant... Any advice is welcome!<br>>><br>>> I had similar intuitions:) We had one paper using EBM-ICA to estimate<br>>> "design-matrix", and compared to PALM with user input design matrix, the<br>>> estimated "design-matrix" shows more sensitivity.<br>>><br>>> Reference:<br>>> Xiaowei Song, Suchita Bhinge, Raimi Quiton, Tulay Adali (2018) A<br>>> two-level ICA approach reveals important differences in the female brain<br>>> responses to thermal pain. In: 2018 IEEE International Symposium on<br>>> Biomedical Imaging. Omni Shoreham Hotel, Washington, DC, USA. Available at:<br>>> https://ieeexplore.ieee.org/document/8363828<br>>> <https://embs.papercept.net/conferences/conferences/ISBI18/program/ISBI18_ProgramAtAGlanceWeb.html#weam21><br>>> .<br>>><br>>><br>>> Best,<br>>> Xiaowei<br>>><br>>> On Wed, Jul 10, 2019 at 8:48 AM Simon Van Eyndhoven <<br>>> Simon.VanEyndhoven@esat.kuleuven.be> wrote:<br>>><br>>>> Dear neuroimaging experts,<br>>>><br>>>> I would like to solicit your advice on a statistical inference problem.<br>>>><br>>>> 1. Goal: perform (non-parametric) inference on betas in linear model<br>>>><br>>>> Assume fMRI data in the form of time courses for a large set of regions<br>>>> of interest (ROIs), represented as a matrix Y (scans x ROIs).<br>>>> Equipped with a design matrix X, I estimate beta parameters for R<br>>>> regressors via the model Y = X*Beta + residual, where Beta = (regressors x<br>>>> ROIs). Then, I wish to perform non-parametric inference: for every entry in<br>>>> Beta, I wish to answer the question "is the Beta(r,k)-th value<br>>>> significantly different from zero?".<br>>>><br>>>> 2. Method: based on permutation testing<br>>>><br>>>> My attempt is to use a permutation testing procedure, as described in<br>>>> [1], with a wavelet-based data reshuffling scheme as in [2]. Essentially, I<br>>>> thus temporally reshuffle the data in Y, keeping X fixed, and re-estimate<br>>>> the parameters Beta many times, to find an empirical null distribution. As<br>>>> a test statistic, I use a T-like statistic: a beta value Beta(r,k) divided<br>>>> by its standard deviation SD(Beta(r,k)), where SD(Beta(r,k)) can be<br>>>> estimated by means of the covariance matrix for the beta values in every<br>>>> ROI, i.e. sigma^2 * (X^T*X)^-1, with sigma^2 the variance of the residual<br>>>> in the considered ROI k, as described in e.g. [3]. (I've also tried the<br>>>> Pearson correlation coefficient between any column of X and Y as test<br>>>> statistic.) This whole procedure is thoroughly described in literature and<br>>>> performs adequately.<br>>>><br>>>> 3. Problem statement: design matrix itself is estimated from the data (?)<br>>>><br>>>> The crux in my case is that, contrary to most of the literature on e.g.<br>>>> the General Linear Model, the matrix X is not an a priori fixed design<br>>>> matrix, but rather *X is estimated from the data Y itself*, i.e. it is<br>>>> found by solving a problem that looks like Y = X*S + residual, where both X<br>>>> and S are unknown, and the residual needs to be minimized - think of it as<br>>>> solving an indendent component analysis (ICA) on the data to find relevant<br>>>> time courses and spatial maps (here there is no experimental stimulus, the<br>>>> data is collected from epilepsy patients under resting state). After<br>>>> applying the inference method outlined above, the outcome is then often<br>>>> that "almost all entries in Beta are significant, over nearly the whole<br>>>> brain". At least some of the regressors in X are biomarker time courses for<br>>>> epilepsy, and hence it is nonsensical that the whole brain would be<br>>>> 'active': a diseased, epilepsy-related portion of the brain would comprise<br>>>> a few ROIs, for instance.<br>>>><br>>>> *I think that this is a 'double dipping problem'*: I want to infer<br>>>> whether the parameters Beta, estimated from the unshuffled data Y, are<br>>>> significant, but these parameters correspond to regressors (in X) that were*<br>>>> themselves tuned to explain the data well*. Therefore, it perhaps<br>>>> doesn't come as a great surprise that these values are generally all<br>>>> 'significant'.<br>>>><br>>>> 4. Solution?<br>>>><br>>>> Is this intuition correct, and if so, what can be done to adjust the<br>>>> inference procedure, and perform unbiased statistical tests in such case?<br>>>> In my understanding, the problem is 'performing inference on data-driven<br>>>> factor models', and I've looked for literature on inference on e.g.<br>>>> ICA-derived time courses (the same problem would also occur in that case)<br>>>> but have found nothing relevant... Any advice is welcome!<br>>>><br>>>> Thank you very much for your help!<br>>>><br>>>> PS: Things I tried already:<br>>>><br>>>> 1) Check whether the autocorrelation structure of the residuals is<br>>>> similar before/after reshuffling: ok, so the data are exchangeable under<br>>>> the null hypothesis.<br>>>> 2) Checked with a design matrix that was fixed (not derived from the<br>>>> data): ok, proper proportion of false positives.<br>>>> 3) Checked whether another resampling scheme solves anything: I tried<br>>>> Fourier-based resampling, and the problem was the same.<br>>>> 4) Since several ROIs (~100) at once are tested, I use correction for<br>>>> familywise error (FWE) by setting the significance threshold using the<br>>>> distribution of the 'maximum T-statistic', as described in [4]. I verified<br>>>> that this strongly controlled the FWE, for dummy regressors.<br>>>><br>>>> References:<br>>>><br>>>> [1] Nichols, Thomas E., and Andrew P. Holmes. "Nonparametric permutation<br>>>> tests for functional neuroimaging: a primer with examples." *Human<br>>>> brain mapping* 15.1 (2002): 1-25.<br>>>><br>>>> [2] Bullmore, Ed, et al. "Colored noise and computational inference in<br>>>> neurophysiological (fMRI) time series analysis: resampling methods in time<br>>>> and wavelet domains." *Human brain mapping* 12.2 (2001): 61-78.<br>>>><br>>>> [3] Poline, Jean-Baptiste, and Matthew Brett. "The general linear model<br>>>> and fMRI: does love last forever?." *Neuroimage* 62.2 (2012): 871-880.<br>>>><br>>>> [4] Nichols, Thomas, and Satoru Hayasaka. "Controlling the familywise<br>>>> error rate in functional neuroimaging: a comparative review." *Statistical<br>>>> methods in medical research* 12.5 (2003): 419-446.<br>>>> --<br>>>> Simon Van Eyndhoven<br>>>><br>>>> PhD student<br>>>> Stadius Center for Dynamical Systems, Signal Processing and Data<br>>>> Analytics<br>>>> KU Leuven, Dept. Electrical Engineering (ESAT)<br>>>><br>>>> email: simon.vaneyndhoven@esat.kuleuven.be<br>>>> telephone: +32 16 32 07 70<br>>>><br>>><br>>><br>>> --<br>>> Best,<br>>> Xiaowei Song<br>>><br>>><br>><br>> --<br>> Best,<br>> Xiaowei Song<br>><br>>Giulio Pergola2019-07-10T20:40:12+00:002019-07-10T20:40:12+00:00One PhD and one postdoc fellowship availablehttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;45d20ead.1907The Group of Psychiatric Neuroscience at the University of Bari Aldo Moro offersa fully funded 3-year PhD scholarship in the framework of a partnership withtechnology SME Itel srl. The collaboration is aimed to develop neuroimaginganalysis protocols for longitudinal data analysis in a locally recruited cohortof individuals with at risk mental state for psychosis. The official call canbe found here. The deadline isJuly 25th and the expected starting date is November 2019.<br><br>We will also publish a call for a postdoctoral neuroimaging position (1year renewable up to 4 years, funding already secured), within the same projecton at risk mental state for psychosis. Expressions of interest are welcomebefore September 30th. Candidates who are in the process of writing up their PhDthesis are encouraged to apply. The expected starting date is February 2020. <br><br>The Group of Psychiatric Neuroscience is headedby Prof. Alessandro Bertolino and investigates the association of genetic variability withsystems-level phenotypes, including structural and functional MagneticResonance Imaging, Diffusion Tensor Imaging, Arterial Spin Labeling, andMagnetoencephalography (for recent articles from the lab, see [1], [2], [3]). Prof. Giuseppe Blasi and his team haveaccess to and routinely tests patients, relatives of patients, and individualsat risk for psychosis. The department has an in-house Elekta Triuxmagnetoencephalography installed in October 2016 with a dedicated technician.Two 3T MRI scanners are available in collaborating hospitals. Successfulcandidates will work in the lab of Brain Imaging, Networks, and Data mining (BIND) directed by Dr. Giulio Pergola. The environmentis interdisciplinary, including psychologists, biotechnologists, and MDs, andinternationally oriented, e.g., including Marie Curie awardees Dr. GiulioPergola and Dr. Linda A. Antonucci, along with several other members trained for> 1 year in UK, USA, and Finland. <br><br>The ideal PhD candidate is a young graduate interested in neuroimaging,with a Master’s Degree in Neuroscience, Psychology, Medicine, Biotechnology,Applied Physics, Computational Science, or related fields. Experience withstatistics is a plus. We expect the successful candidate to be motivatedand research-oriented. It is equally important to have a team-workingattitude and a motivation to fit in the group. Proficiency in Italian isnot required. <br><br>The ideal postdoctoral candidate is a neuroimager with internationalexperience and a PhD in Neuroscience, Cognitive Sciences, Neuropsychology, ComputationalScience, or any quantitative research field. Coding experience (Matlab, Python,or R) is required, whereas machine learning and graph theory expertise are aplus. Successful candidates are expected to have at least one published and twosubmitted lead author papers.<br><br>The University of Bari Aldo Moro, with about 50.000students, 1500 permanent professors and researchers, 1500 units of techniciansand administrative staff, is one the largest Italian universities, and thesecond largest in Southern Italy. Bari is a sunny city located by the sea, andthe third greatest Italian city south of Rome. The cost of living is low relativeto Italian standards. According to Italian law, fellowships after taxes areabout 1100 EUR for PhD students and minimum 1450 EUR for early stageresearchers (< 4 years experience); postdoc salaries can be higher.<br><br>Please contact Dr. Giulio Pergola sending a CV and the contact details of atleast two references if you would like to have more information on the positionand on the ongoing projects (giulio.pergola@uniba.it).<br><br>--<br>Giulio Pergola, PhD<br><br>Marie Curie Visiting Scientist<br>Lieber Institute for Brain Development<br>855 N Wolfe St.<br>21205 Baltimore, MD<br><br>Assistant Professor in Biological Psychology<br>Lab Director - Brain Imaging, Networks, and Data mining<br>Department of Basic Medical Science, Neuroscience and Sense Organs<br>University of Bari Aldo Moro<br>Piazza Giulio Cesare, 11<br>70121 Bari, Italy<br><br>https://bit.ly/2EctXJW<br>Tel: +39 080 5478548<br>Fax: +39 080 5593204Simon Van Eyndhoven2019-07-10T22:16:00+02:002019-07-10T22:16:00+02:00Re: Inference on GLM parameters when design matrix is derived from datahttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;2dffa18f.1907Dear Xiaowei,<br><br>Thanks for the clarification. I agree that for the unshuffled data, the<br>beta parameters are taken from S, indeed (I used both the notation with<br>Beta to draw the connection with the GLM, and the notation with S to<br>make the connection with ICA-style methods).<br><br>For the subsequent inference, my approach was to fix the X that was<br>found from this matrix factorization problem, and then solve the GLM<br>(i.e. linear regression) N times, i.e. for N reshufflings of the data,<br>to find the null distribution of the parameters. From that, the<br>percentile of the original parameters indicates their significance.<br>(Alternatively, I could re-estimate both X and S for every reshuffle,<br>but the factorization is quite computationally intensive due to several<br>aspects that I'll omit here for brevity, so this would be prohibitive.)<br><br>However, from what I understand, you have performed inference<br>differently? Did you also aim to determine the significance of the<br>coefficients / scores / loadings / parameters / ... (whichever<br>nomenclature fits the interpretation) in S, or was your objective<br>different (e.g. correlation of the columns in X or the rows of S with a<br>behavioral measure)? I am eager to know more about how you performed it<br>concretely, if you are willing to go in more depth! :)<br><br>All the best, and thanks for your help so far,<br><br>Simon<br><br>On 2019-07-10 19:01, Xiaowei Song wrote:<br><br>> From your "3. Problem statement", the "parameters Beta" you mentioned, is actually S in "Y = X*S + residual" if the design matrix "X" is given in GLM. Do you agree?<br>> In ICA methods, with some abuse of notation, X is the optimization variable instead of being a user-defined design-matrix in GLM. The ICA cost function is usually derived using maximum likelihood estimation, which has many desired properties (For example, Lecture 8: Properties of Maximum Likelihood Estimation (MLE) [1] ). I don't think I am a victim to the fallacy you mentioned.<br>> I guess you have the point our paper tries to make: "these parameters correspond to regressors (in X) that were_ themselves tuned to explain the data well_. ".<br>> By the way, the comparison was done in a journal version.<br>> I hope I answered your question.<br>><br>> On Wed, Jul 10, 2019 at 12:19 PM Simon Van Eyndhoven <Simon.VanEyndhoven@esat.kuleuven.be> wrote:<br>><br>> Dear Xiaowei,<br>><br>> Thanks for your reply! Do you mean that you fell victim to the same fallacy, i.e. that the higher sensitivity was due to the self-fulfilling prophecy of a data-derived design matrix, or do you mean that the higher sensitivity was unbiased and therefore good news?<br>><br>> PS: I consulted the paper, but I don't seem to find the comparison with permutation analysis of linear models (PALM)?<br>><br>> Best regards,<br>><br>> Simon<br>><br>> On 2019-07-10 17:20, Xiaowei Song wrote:<br>><br>> Dear Simon,<br>><br>> Is this intuition correct, and if so, what can be done to adjust the inference procedure, and perform unbiased statistical tests in such case? In my understanding, the problem is 'performing inference on data-driven factor models', and I've looked for literature on inference on e.g. ICA-derived time courses (the same problem would also occur in that case) but have found nothing relevant... Any advice is welcome!<br>><br>> I had similar intuitions:) We had one paper using EBM-ICA to estimate "design-matrix", and compared to PALM with user input design matrix, the estimated "design-matrix" shows more sensitivity.<br>><br>> Reference:<br>> Xiaowei Song, Suchita Bhinge, Raimi Quiton, Tulay Adali (2018) A two-level ICA approach reveals important differences in the female brain responses to thermal pain. In: 2018 IEEE International Symposium on Biomedical Imaging. Omni Shoreham Hotel, Washington, DC, USA. Available at: https://ieeexplore.ieee.org/document/8363828 [2].<br>><br>> Best,<br>> Xiaowei<br>><br>> On Wed, Jul 10, 2019 at 8:48 AM Simon Van Eyndhoven <Simon.VanEyndhoven@esat.kuleuven.be> wrote:<br>><br>> Dear neuroimaging experts,<br>><br>> I would like to solicit your advice on a statistical inference problem.<br>><br>> 1. Goal: perform (non-parametric) inference on betas in linear model<br>><br>> Assume fMRI data in the form of time courses for a large set of regions of interest (ROIs), represented as a matrix Y (scans x ROIs).<br>> Equipped with a design matrix X, I estimate beta parameters for R regressors via the model Y = X*Beta + residual, where Beta = (regressors x ROIs). Then, I wish to perform non-parametric inference: for every entry in Beta, I wish to answer the question "is the Beta(r,k)-th value significantly different from zero?".<br>><br>> 2. Method: based on permutation testing<br>><br>> My attempt is to use a permutation testing procedure, as described in [1], with a wavelet-based data reshuffling scheme as in [2]. Essentially, I thus temporally reshuffle the data in Y, keeping X fixed, and re-estimate the parameters Beta many times, to find an empirical null distribution. As a test statistic, I use a T-like statistic: a beta value Beta(r,k) divided by its standard deviation SD(Beta(r,k)), where SD(Beta(r,k)) can be estimated by means of the covariance matrix for the beta values in every ROI, i.e. sigma^2 * (X^T*X)^-1, with sigma^2 the variance of the residual in the considered ROI k, as described in e.g. [3]. (I've also tried the Pearson correlation coefficient between any column of X and Y as test statistic.) This whole procedure is thoroughly described in literature and performs adequately.<br>><br>> 3. Problem statement: design matrix itself is estimated from the data (?)<br>><br>> The crux in my case is that, contrary to most of the literature on e.g. the General Linear Model, the matrix X is not an a priori fixed design matrix, but rather _X is estimated from the data Y itself_, i.e. it is found by solving a problem that looks like Y = X*S + residual, where both X and S are unknown, and the residual needs to be minimized - think of it as solving an indendent component analysis (ICA) on the data to find relevant time courses and spatial maps (here there is no experimental stimulus, the data is collected from epilepsy patients under resting state). After applying the inference method outlined above, the outcome is then often that "almost all entries in Beta are significant, over nearly the whole brain". At least some of the regressors in X are biomarker time courses for epilepsy, and hence it is nonsensical that the whole brain would be 'active': a diseased, epilepsy-related portion of the brain would comprise a few ROIs, for instance.<br>><br>> _I think that this is a 'double dipping problem'_: I want to infer whether the parameters Beta, estimated from the unshuffled data Y, are significant, but these parameters correspond to regressors (in X) that were_ themselves tuned to explain the data well_. Therefore, it perhaps doesn't come as a great surprise that these values are generally all 'significant'.<br>><br>> 4. Solution?<br>><br>> Is this intuition correct, and if so, what can be done to adjust the inference procedure, and perform unbiased statistical tests in such case? In my understanding, the problem is 'performing inference on data-driven factor models', and I've looked for literature on inference on e.g. ICA-derived time courses (the same problem would also occur in that case) but have found nothing relevant... Any advice is welcome!<br>><br>> Thank you very much for your help!<br>><br>> PS: Things I tried already:<br>><br>> 1) Check whether the autocorrelation structure of the residuals is similar before/after reshuffling: ok, so the data are exchangeable under the null hypothesis.<br>> 2) Checked with a design matrix that was fixed (not derived from the data): ok, proper proportion of false positives.<br>> 3) Checked whether another resampling scheme solves anything: I tried Fourier-based resampling, and the problem was the same.<br>> 4) Since several ROIs (~100) at once are tested, I use correction for familywise error (FWE) by setting the significance threshold using the distribution of the 'maximum T-statistic', as described in [4]. I verified that this strongly controlled the FWE, for dummy regressors.<br>><br>> References:<br>><br>> [1] Nichols, Thomas E., and Andrew P. Holmes. "Nonparametric permutation tests for functional neuroimaging: a primer with examples." _Human brain mapping_ 15.1 (2002): 1-25.<br>><br>> [2] Bullmore, Ed, et al. "Colored noise and computational inference in neurophysiological (fMRI) time series analysis: resampling methods in time and wavelet domains." _Human brain mapping_ 12.2 (2001): 61-78.<br>><br>> [3] Poline, Jean-Baptiste, and Matthew Brett. "The general linear model and fMRI: does love last forever?." _Neuroimage_ 62.2 (2012): 871-880.<br>><br>> [4] Nichols, Thomas, and Satoru Hayasaka. "Controlling the familywise error rate in functional neuroimaging: a comparative review." _Statistical methods in medical research_ 12.5 (2003): 419-446.<br>><br>> --<br>> Simon Van Eyndhoven<br>><br>> PhD student<br>> Stadius Center for Dynamical Systems, Signal Processing and Data Analytics<br>> KU Leuven, Dept. Electrical Engineering (ESAT)<br>><br>> email: simon.vaneyndhoven@esat.kuleuven.be<br>> telephone: +32 16 32 07 70<br>><br>> --<br>><br>> Best,<br>> Xiaowei Song<br><br>--<br><br>Best,<br>Xiaowei Song<br><br>Links:<br>------<br>[1]<br>https://engineering.purdue.edu/ChanGroup/ECE645Notes/StudentLecture08.pdf<br>[2] Ray Arceo2019-07-10T19:47:29+01:002019-07-10T19:47:29+01:00Postdoctoral Training Opportunity at Feinberg School of Medicine - Northwestern Universityhttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;a16d061b.1907The Department of Physical Therapy and Human Movement Sciences (DPTHMS) at Feinberg School of Medicine of Northwestern University is involved in cutting-edge research to 1) understand the neuromechanisms underlying movement deficits following various neuropathological conditions, and in 2) the development of novel rehabilitation interventions and devices based on the latest scientific findings.<br><br>Drs. Julius Dewald and Jun Yao are seeking a highly motivated postdoctoral trainee with expertise in neuroimaging using EEG, fMRI and DTI, and/or using bio- mechanical methods to quantify movement features in human subjects. Enthusiastic candidates with a recent PhD are encouraged to apply. Candidates with no PhD degree but extensive knowledge in the above listed fields may be considered.<br><br>DPTHMS is part of a vibrant educational and cultural community located on the Chicago campus of Northwestern University close to Lake Michigan and the “Magnificent Mile". DPTHMS collaborates with the Departments of Biomedical Engineering, Mechanical Engineering, Radiology, Physiology, Communications Sciences and Disorders, Neurology, Physical Medicine & Rehabilitation, Orthopedic and Neurologic Surgery, Medicine, Preventive Medicine, Emergency Medicine, and Oncology.<br><br>Salary and benefits are competitive. US citizenship is not required.<br><br>Please send your curriculum vitae (CV), contact information, and two to three references to Ray Arceo at ray.arceo@northwestern.edu. Apply by July 31, 2019.<br><br>Northwestern University is an Equal Opportunity, Affirmative Action Employer of all protected classes, including veterans and individuals with disabilities. Women, underrepresented racial and ethnic minorities, individuals with disabilities, and veterans are encouraged to apply. Hiring is contingent upon eligibility to work in the United States.Xiaowei Song2019-07-10T13:01:51-04:002019-07-10T13:01:51-04:00Re: Inference on GLM parameters when design matrix is derived from datahttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;d3947dd6.1907From your "3. Problem statement", the "parameters Beta" you mentioned, is<br>actually S in "Y = X*S + residual" if the design matrix "X" is given in<br>GLM. Do you agree?<br>In ICA methods, with some abuse of notation, X is the optimization variable<br>instead of being a user-defined design-matrix in GLM. The ICA cost function<br>is usually derived using maximum likelihood estimation, which has many<br>desired properties (For example, Lecture 8: Properties of Maximum<br>Likelihood Estimation (MLE)<br><https://engineering.purdue.edu/ChanGroup/ECE645Notes/StudentLecture08.pdf><br>). I don't think I am a victim to the fallacy you mentioned.<br>I guess you have the point our paper tries to make: "these parameters<br>correspond to regressors (in X) that were* themselves tuned to explain the<br>data well*. ".<br>By the way, the comparison was done in a journal version.<br>I hope I answered your question.<br><br>On Wed, Jul 10, 2019 at 12:19 PM Simon Van Eyndhoven <<br>Simon.VanEyndhoven@esat.kuleuven.be> wrote:<br><br>> Dear Xiaowei,<br>><br>> Thanks for your reply! Do you mean that you fell victim to the same<br>> fallacy, i.e. that the higher sensitivity was due to the self-fulfilling<br>> prophecy of a data-derived design matrix, or do you mean that the higher<br>> sensitivity was unbiased and therefore good news?<br>><br>> PS: I consulted the paper, but I don't seem to find the comparison with<br>> permutation analysis of linear models (PALM)?<br>><br>> Best regards,<br>><br>> Simon<br>><br>> On 2019-07-10 17:20, Xiaowei Song wrote:<br>><br>> Dear Simon,<br>><br>> Is this intuition correct, and if so, what can be done to adjust the<br>>> inference procedure, and perform unbiased statistical tests in such case?<br>>> In my understanding, the problem is 'performing inference on data-driven<br>>> factor models', and I've looked for literature on inference on e.g.<br>>> ICA-derived time courses (the same problem would also occur in that case)<br>>> but have found nothing relevant... Any advice is welcome!<br>><br>> I had similar intuitions:) We had one paper using EBM-ICA to estimate<br>> "design-matrix", and compared to PALM with user input design matrix, the<br>> estimated "design-matrix" shows more sensitivity.<br>><br>> Reference:<br>> Xiaowei Song, Suchita Bhinge, Raimi Quiton, Tulay Adali (2018) A two-level<br>> ICA approach reveals important differences in the female brain responses to<br>> thermal pain. In: 2018 IEEE International Symposium on Biomedical Imaging.<br>> Omni Shoreham Hotel, Washington, DC, USA. Available at:<br>> https://ieeexplore.ieee.org/document/8363828<br>> <https://embs.papercept.net/conferences/conferences/ISBI18/program/ISBI18_ProgramAtAGlanceWeb.html#weam21><br>> .<br>><br>><br>> Best,<br>> Xiaowei<br>><br>> On Wed, Jul 10, 2019 at 8:48 AM Simon Van Eyndhoven <<br>> Simon.VanEyndhoven@esat.kuleuven.be> wrote:<br>><br>>> Dear neuroimaging experts,<br>>><br>>> I would like to solicit your advice on a statistical inference problem.<br>>><br>>> 1. Goal: perform (non-parametric) inference on betas in linear model<br>>><br>>> Assume fMRI data in the form of time courses for a large set of regions<br>>> of interest (ROIs), represented as a matrix Y (scans x ROIs).<br>>> Equipped with a design matrix X, I estimate beta parameters for R<br>>> regressors via the model Y = X*Beta + residual, where Beta = (regressors x<br>>> ROIs). Then, I wish to perform non-parametric inference: for every entry in<br>>> Beta, I wish to answer the question "is the Beta(r,k)-th value<br>>> significantly different from zero?".<br>>><br>>> 2. Method: based on permutation testing<br>>><br>>> My attempt is to use a permutation testing procedure, as described in<br>>> [1], with a wavelet-based data reshuffling scheme as in [2]. Essentially, I<br>>> thus temporally reshuffle the data in Y, keeping X fixed, and re-estimate<br>>> the parameters Beta many times, to find an empirical null distribution. As<br>>> a test statistic, I use a T-like statistic: a beta value Beta(r,k) divided<br>>> by its standard deviation SD(Beta(r,k)), where SD(Beta(r,k)) can be<br>>> estimated by means of the covariance matrix for the beta values in every<br>>> ROI, i.e. sigma^2 * (X^T*X)^-1, with sigma^2 the variance of the residual<br>>> in the considered ROI k, as described in e.g. [3]. (I've also tried the<br>>> Pearson correlation coefficient between any column of X and Y as test<br>>> statistic.) This whole procedure is thoroughly described in literature and<br>>> performs adequately.<br>>><br>>> 3. Problem statement: design matrix itself is estimated from the data (?)<br>>><br>>> The crux in my case is that, contrary to most of the literature on e.g.<br>>> the General Linear Model, the matrix X is not an a priori fixed design<br>>> matrix, but rather *X is estimated from the data Y itself*, i.e. it is<br>>> found by solving a problem that looks like Y = X*S + residual, where both X<br>>> and S are unknown, and the residual needs to be minimized - think of it as<br>>> solving an indendent component analysis (ICA) on the data to find relevant<br>>> time courses and spatial maps (here there is no experimental stimulus, the<br>>> data is collected from epilepsy patients under resting state). After<br>>> applying the inference method outlined above, the outcome is then often<br>>> that "almost all entries in Beta are significant, over nearly the whole<br>>> brain". At least some of the regressors in X are biomarker time courses for<br>>> epilepsy, and hence it is nonsensical that the whole brain would be<br>>> 'active': a diseased, epilepsy-related portion of the brain would comprise<br>>> a few ROIs, for instance.<br>>><br>>> *I think that this is a 'double dipping problem'*: I want to infer<br>>> whether the parameters Beta, estimated from the unshuffled data Y, are<br>>> significant, but these parameters correspond to regressors (in X) that were*<br>>> themselves tuned to explain the data well*. Therefore, it perhaps<br>>> doesn't come as a great surprise that these values are generally all<br>>> 'significant'.<br>>><br>>> 4. Solution?<br>>><br>>> Is this intuition correct, and if so, what can be done to adjust the<br>>> inference procedure, and perform unbiased statistical tests in such case?<br>>> In my understanding, the problem is 'performing inference on data-driven<br>>> factor models', and I've looked for literature on inference on e.g.<br>>> ICA-derived time courses (the same problem would also occur in that case)<br>>> but have found nothing relevant... Any advice is welcome!<br>>><br>>> Thank you very much for your help!<br>>><br>>> PS: Things I tried already:<br>>><br>>> 1) Check whether the autocorrelation structure of the residuals is<br>>> similar before/after reshuffling: ok, so the data are exchangeable under<br>>> the null hypothesis.<br>>> 2) Checked with a design matrix that was fixed (not derived from the<br>>> data): ok, proper proportion of false positives.<br>>> 3) Checked whether another resampling scheme solves anything: I tried<br>>> Fourier-based resampling, and the problem was the same.<br>>> 4) Since several ROIs (~100) at once are tested, I use correction for<br>>> familywise error (FWE) by setting the significance threshold using the<br>>> distribution of the 'maximum T-statistic', as described in [4]. I verified<br>>> that this strongly controlled the FWE, for dummy regressors.<br>>><br>>> References:<br>>><br>>> [1] Nichols, Thomas E., and Andrew P. Holmes. "Nonparametric permutation<br>>> tests for functional neuroimaging: a primer with examples." *Human brain<br>>> mapping* 15.1 (2002): 1-25.<br>>><br>>> [2] Bullmore, Ed, et al. "Colored noise and computational inference in<br>>> neurophysiological (fMRI) time series analysis: resampling methods in time<br>>> and wavelet domains." *Human brain mapping* 12.2 (2001): 61-78.<br>>><br>>> [3] Poline, Jean-Baptiste, and Matthew Brett. "The general linear model<br>>> and fMRI: does love last forever?." *Neuroimage* 62.2 (2012): 871-880.<br>>><br>>> [4] Nichols, Thomas, and Satoru Hayasaka. "Controlling the familywise<br>>> error rate in functional neuroimaging: a comparative review." *Statistical<br>>> methods in medical research* 12.5 (2003): 419-446.<br>>> --<br>>> Simon Van Eyndhoven<br>>><br>>> PhD student<br>>> Stadius Center for Dynamical Systems, Signal Processing and Data Analytics<br>>> KU Leuven, Dept. Electrical Engineering (ESAT)<br>>><br>>> email: simon.vaneyndhoven@esat.kuleuven.be<br>>> telephone: +32 16 32 07 70<br>>><br>><br>><br>> --<br>> Best,<br>> Xiaowei Song<br>><br>>Simon Van Eyndhoven2019-07-10T18:19:14+02:002019-07-10T18:19:14+02:00Re: Inference on GLM parameters when design matrix is derived from datahttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;f32261e0.1907Dear Xiaowei,<br><br>Thanks for your reply! Do you mean that you fell victim to the same<br>fallacy, i.e. that the higher sensitivity was due to the self-fulfilling<br>prophecy of a data-derived design matrix, or do you mean that the higher<br>sensitivity was unbiased and therefore good news?<br><br>PS: I consulted the paper, but I don't seem to find the comparison with<br>permutation analysis of linear models (PALM)?<br><br>Best regards,<br><br>Simon<br><br>On 2019-07-10 17:20, Xiaowei Song wrote:<br><br>> Dear Simon,<br>><br>>> Is this intuition correct, and if so, what can be done to adjust the inference procedure, and perform unbiased statistical tests in such case? In my understanding, the problem is 'performing inference on data-driven factor models', and I've looked for literature on inference on e.g. ICA-derived time courses (the same problem would also occur in that case) but have found nothing relevant... Any advice is welcome!<br>><br>> I had similar intuitions:) We had one paper using EBM-ICA to estimate "design-matrix", and compared to PALM with user input design matrix, the estimated "design-matrix" shows more sensitivity.<br>><br>> Reference:<br>> Xiaowei Song, Suchita Bhinge, Raimi Quiton, Tulay Adali (2018) A two-level ICA approach reveals important differences in the female brain responses to thermal pain. In: 2018 IEEE International Symposium on Biomedical Imaging. Omni Shoreham Hotel, Washington, DC, USA. Available at: https://ieeexplore.ieee.org/document/8363828 [1].<br>><br>> Best,<br>> Xiaowei<br>><br>> On Wed, Jul 10, 2019 at 8:48 AM Simon Van Eyndhoven <Simon.VanEyndhoven@esat.kuleuven.be> wrote:<br>><br>>> Dear neuroimaging experts,<br>>><br>>> I would like to solicit your advice on a statistical inference problem.<br>>><br>>> 1. Goal: perform (non-parametric) inference on betas in linear model<br>>><br>>> Assume fMRI data in the form of time courses for a large set of regions of interest (ROIs), represented as a matrix Y (scans x ROIs).<br>>> Equipped with a design matrix X, I estimate beta parameters for R regressors via the model Y = X*Beta + residual, where Beta = (regressors x ROIs). Then, I wish to perform non-parametric inference: for every entry in Beta, I wish to answer the question "is the Beta(r,k)-th value significantly different from zero?".<br>>><br>>> 2. Method: based on permutation testing<br>>><br>>> My attempt is to use a permutation testing procedure, as described in [1], with a wavelet-based data reshuffling scheme as in [2]. Essentially, I thus temporally reshuffle the data in Y, keeping X fixed, and re-estimate the parameters Beta many times, to find an empirical null distribution. As a test statistic, I use a T-like statistic: a beta value Beta(r,k) divided by its standard deviation SD(Beta(r,k)), where SD(Beta(r,k)) can be estimated by means of the covariance matrix for the beta values in every ROI, i.e. sigma^2 * (X^T*X)^-1, with sigma^2 the variance of the residual in the considered ROI k, as described in e.g. [3]. (I've also tried the Pearson correlation coefficient between any column of X and Y as test statistic.) This whole procedure is thoroughly described in literature and performs adequately.<br>>><br>>> 3. Problem statement: design matrix itself is estimated from the data (?)<br>>><br>>> The crux in my case is that, contrary to most of the literature on e.g. the General Linear Model, the matrix X is not an a priori fixed design matrix, but rather _X is estimated from the data Y itself_, i.e. it is found by solving a problem that looks like Y = X*S + residual, where both X and S are unknown, and the residual needs to be minimized - think of it as solving an indendent component analysis (ICA) on the data to find relevant time courses and spatial maps (here there is no experimental stimulus, the data is collected from epilepsy patients under resting state). After applying the inference method outlined above, the outcome is then often that "almost all entries in Beta are significant, over nearly the whole brain". At least some of the regressors in X are biomarker time courses for epilepsy, and hence it is nonsensical that the whole brain would be 'active': a diseased, epilepsy-related portion of the brain would comprise a few ROIs, for instance.<br>>><br>>> _I think that this is a 'double dipping problem'_: I want to infer whether the parameters Beta, estimated from the unshuffled data Y, are significant, but these parameters correspond to regressors (in X) that were_ themselves tuned to explain the data well_. Therefore, it perhaps doesn't come as a great surprise that these values are generally all 'significant'.<br>>><br>>> 4. Solution?<br>>><br>>> Is this intuition correct, and if so, what can be done to adjust the inference procedure, and perform unbiased statistical tests in such case? In my understanding, the problem is 'performing inference on data-driven factor models', and I've looked for literature on inference on e.g. ICA-derived time courses (the same problem would also occur in that case) but have found nothing relevant... Any advice is welcome!<br>>><br>>> Thank you very much for your help!<br>>><br>>> PS: Things I tried already:<br>>><br>>> 1) Check whether the autocorrelation structure of the residuals is similar before/after reshuffling: ok, so the data are exchangeable under the null hypothesis.<br>>> 2) Checked with a design matrix that was fixed (not derived from the data): ok, proper proportion of false positives.<br>>> 3) Checked whether another resampling scheme solves anything: I tried Fourier-based resampling, and the problem was the same.<br>>> 4) Since several ROIs (~100) at once are tested, I use correction for familywise error (FWE) by setting the significance threshold using the distribution of the 'maximum T-statistic', as described in [4]. I verified that this strongly controlled the FWE, for dummy regressors.<br>>><br>>> References:<br>>><br>>> [1] Nichols, Thomas E., and Andrew P. Holmes. "Nonparametric permutation tests for functional neuroimaging: a primer with examples." _Human brain mapping_ 15.1 (2002): 1-25.<br>>><br>>> [2] Bullmore, Ed, et al. "Colored noise and computational inference in neurophysiological (fMRI) time series analysis: resampling methods in time and wavelet domains." _Human brain mapping_ 12.2 (2001): 61-78.<br>>><br>>> [3] Poline, Jean-Baptiste, and Matthew Brett. "The general linear model and fMRI: does love last forever?." _Neuroimage_ 62.2 (2012): 871-880.<br>>><br>>> [4] Nichols, Thomas, and Satoru Hayasaka. "Controlling the familywise error rate in functional neuroimaging: a comparative review." _Statistical methods in medical research_ 12.5 (2003): 419-446.<br>>><br>>> --<br>>> Simon Van Eyndhoven<br>>><br>>> PhD student<br>>> Stadius Center for Dynamical Systems, Signal Processing and Data Analytics<br>>> KU Leuven, Dept. Electrical Engineering (ESAT)<br>>><br>>> email: simon.vaneyndhoven@esat.kuleuven.be<br>>> telephone: +32 16 32 07 70<br>><br>> --<br>><br>> Best,<br>> Xiaowei Song<br><br>Links:<br>------<br>[1] Jacobs, H (NP)2019-07-10T15:47:43+00:002019-07-10T15:47:43+00:00Postdoc position at MGH/Harvard Medical Schoolhttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;32ea2507.1907<< I will attend the AAIC2019 conference in Los Angeles. Send me an email if you like to meet and discuss this position >><br><br>Postdoctoral Opportunity at Mass General Hospital – Harvard Medical School<br><br>The Gordon Center for Medical Imaging (GCMI) in the Department of Radiology at Massachusetts General Hospital (MGH) and Harvard Medical School (HMS) in Boston, Massachusetts, has an opening for highly qualified individuals at the post-doctoral level to work on research related to multi-modal 7T (f)MRI imaging of the brainstem in aging under the mentorship of Dr. Heidi Jacobs. Within this NIA-funded RO1 project, relationships between memory and autonomic tone will be investigated at very high resolution. This project involves dedicated imaging of the brainstem nuclei, high-resolution fMRI (possibility to go to layer-fMRI) with physiological monitoring, HCP sequences for diffusion imaging and PET-imaging. Most of the work in this project will be performed at the Athinoula A. Martinos Center for Biomedical Imaging.<br><br>The Department of Radiology at MGH is equipped with the first mobile PET/CT, the first brain PET/MRI, the first whole-body PET/MRI in the USA and several MRI scanners, including two 7T ultra-high-field scanner. It is also equipped with a substantial large scale shared memory computing facility for parametric image analysis, tomographic reconstruction, Monte Carlo simulation, and other computationally intensive research applications.<br><br>Applicants must have obtained or anticipate soon receiving a Ph.D. in biomedical, computer or electrical engineering, medical physics, cognitive neuroscience or a related field. Strong analytical, quantitative and programming skills are essential (e.g., QSM, fMRI-physiological data analyses). Prior experience in medical imaging, signal/image processing, and physiological investigation is advantageous. The successful candidate will have joint appointments at MGH and HMS and will be closely interacting with the Harvard Aging Brain Study team, the HCP team and the high-field MRI group. If interested, please send your CV, letter describing interests, background, qualifications and 3 references (including contact details).<br><br>MGH & HMS are equal-opportunity, affirmative action employers. Women and minority candidates are encouraged to apply.<br><br>Heidi Jacobs, Ph.D.<br>Instructor, GCMI<br>Email: hjacobs@mgh.harvard.edu<mailto:hjacobs@mgh.harvard.edu><br>[cid:image001.png@01D4F544.4ACC20E0]<br>Gordon Center for Medical Imaging<br>MGH – Nashua 6606; Radiology, MGH<br>125 Nashua Street<br>Boston, MA 02114Xiaowei Song2019-07-10T11:20:41-04:002019-07-10T11:20:41-04:00Re: Inference on GLM parameters when design matrix is derived from datahttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;5b2d1488.1907Dear Simon,<br><br>Is this intuition correct, and if so, what can be done to adjust the<br>> inference procedure, and perform unbiased statistical tests in such case?<br>> In my understanding, the problem is 'performing inference on data-driven<br>> factor models', and I've looked for literature on inference on e.g.<br>> ICA-derived time courses (the same problem would also occur in that case)<br>> but have found nothing relevant... Any advice is welcome!<br><br>I had similar intuitions:) We had one paper using EBM-ICA to estimate<br>"design-matrix", and compared to PALM with user input design matrix, the<br>estimated "design-matrix" shows more sensitivity.<br><br>Reference:<br>Xiaowei Song, Suchita Bhinge, Raimi Quiton, Tulay Adali (2018) A two-level<br>ICA approach reveals important differences in the female brain responses to<br>thermal pain. In: 2018 IEEE International Symposium on Biomedical Imaging.<br>Omni Shoreham Hotel, Washington, DC, USA. Available at:<br>https://ieeexplore.ieee.org/document/8363828<br><https://embs.papercept.net/conferences/conferences/ISBI18/program/ISBI18_ProgramAtAGlanceWeb.html#weam21><br>.<br><br>Best,<br>Xiaowei<br><br>On Wed, Jul 10, 2019 at 8:48 AM Simon Van Eyndhoven <<br>Simon.VanEyndhoven@esat.kuleuven.be> wrote:<br><br>> Dear neuroimaging experts,<br>><br>> I would like to solicit your advice on a statistical inference problem.<br>><br>> 1. Goal: perform (non-parametric) inference on betas in linear model<br>><br>> Assume fMRI data in the form of time courses for a large set of regions of<br>> interest (ROIs), represented as a matrix Y (scans x ROIs).<br>> Equipped with a design matrix X, I estimate beta parameters for R<br>> regressors via the model Y = X*Beta + residual, where Beta = (regressors x<br>> ROIs). Then, I wish to perform non-parametric inference: for every entry in<br>> Beta, I wish to answer the question "is the Beta(r,k)-th value<br>> significantly different from zero?".<br>><br>> 2. Method: based on permutation testing<br>><br>> My attempt is to use a permutation testing procedure, as described in [1],<br>> with a wavelet-based data reshuffling scheme as in [2]. Essentially, I thus<br>> temporally reshuffle the data in Y, keeping X fixed, and re-estimate the<br>> parameters Beta many times, to find an empirical null distribution. As a<br>> test statistic, I use a T-like statistic: a beta value Beta(r,k) divided by<br>> its standard deviation SD(Beta(r,k)), where SD(Beta(r,k)) can be estimated<br>> by means of the covariance matrix for the beta values in every ROI, i.e.<br>> sigma^2 * (X^T*X)^-1, with sigma^2 the variance of the residual in the<br>> considered ROI k, as described in e.g. [3]. (I've also tried the Pearson<br>> correlation coefficient between any column of X and Y as test statistic.)<br>> This whole procedure is thoroughly described in literature and performs<br>> adequately.<br>><br>> 3. Problem statement: design matrix itself is estimated from the data (?)<br>><br>> The crux in my case is that, contrary to most of the literature on e.g.<br>> the General Linear Model, the matrix X is not an a priori fixed design<br>> matrix, but rather *X is estimated from the data Y itself*, i.e. it is<br>> found by solving a problem that looks like Y = X*S + residual, where both X<br>> and S are unknown, and the residual needs to be minimized - think of it as<br>> solving an indendent component analysis (ICA) on the data to find relevant<br>> time courses and spatial maps (here there is no experimental stimulus, the<br>> data is collected from epilepsy patients under resting state). After<br>> applying the inference method outlined above, the outcome is then often<br>> that "almost all entries in Beta are significant, over nearly the whole<br>> brain". At least some of the regressors in X are biomarker time courses for<br>> epilepsy, and hence it is nonsensical that the whole brain would be<br>> 'active': a diseased, epilepsy-related portion of the brain would comprise<br>> a few ROIs, for instance.<br>><br>> *I think that this is a 'double dipping problem'*: I want to infer<br>> whether the parameters Beta, estimated from the unshuffled data Y, are<br>> significant, but these parameters correspond to regressors (in X) that were*<br>> themselves tuned to explain the data well*. Therefore, it perhaps doesn't<br>> come as a great surprise that these values are generally all 'significant'.<br>><br>> 4. Solution?<br>><br>> Is this intuition correct, and if so, what can be done to adjust the<br>> inference procedure, and perform unbiased statistical tests in such case?<br>> In my understanding, the problem is 'performing inference on data-driven<br>> factor models', and I've looked for literature on inference on e.g.<br>> ICA-derived time courses (the same problem would also occur in that case)<br>> but have found nothing relevant... Any advice is welcome!<br>><br>> Thank you very much for your help!<br>><br>> PS: Things I tried already:<br>><br>> 1) Check whether the autocorrelation structure of the residuals is similar<br>> before/after reshuffling: ok, so the data are exchangeable under the null<br>> hypothesis.<br>> 2) Checked with a design matrix that was fixed (not derived from the<br>> data): ok, proper proportion of false positives.<br>> 3) Checked whether another resampling scheme solves anything: I tried<br>> Fourier-based resampling, and the problem was the same.<br>> 4) Since several ROIs (~100) at once are tested, I use correction for<br>> familywise error (FWE) by setting the significance threshold using the<br>> distribution of the 'maximum T-statistic', as described in [4]. I verified<br>> that this strongly controlled the FWE, for dummy regressors.<br>><br>> References:<br>><br>> [1] Nichols, Thomas E., and Andrew P. Holmes. "Nonparametric permutation<br>> tests for functional neuroimaging: a primer with examples." *Human brain<br>> mapping* 15.1 (2002): 1-25.<br>><br>> [2] Bullmore, Ed, et al. "Colored noise and computational inference in<br>> neurophysiological (fMRI) time series analysis: resampling methods in time<br>> and wavelet domains." *Human brain mapping* 12.2 (2001): 61-78.<br>><br>> [3] Poline, Jean-Baptiste, and Matthew Brett. "The general linear model<br>> and fMRI: does love last forever?." *Neuroimage* 62.2 (2012): 871-880.<br>><br>> [4] Nichols, Thomas, and Satoru Hayasaka. "Controlling the familywise<br>> error rate in functional neuroimaging: a comparative review." *Statistical<br>> methods in medical research* 12.5 (2003): 419-446.<br>> --<br>> Simon Van Eyndhoven<br>><br>> PhD student<br>> Stadius Center for Dynamical Systems, Signal Processing and Data Analytics<br>> KU Leuven, Dept. Electrical Engineering (ESAT)<br>><br>> email: simon.vaneyndhoven@esat.kuleuven.be<br>> telephone: +32 16 32 07 70<br>>Amirhossein Jafarian2019-07-10T14:33:18+00:002019-07-10T14:33:18+00:00Re: DCM cross-spectral densities- parameters interpretationhttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;62bbf792.1907Please consider this email:<br>Dear SPM user,<br>Thanks for your email.<br>In DCM for EEG/MEG using CMC model we need to tell the software about the forward and backward connections we would like to establish between regions. The rule for defining these matrices are the same as any other DCM. However in DCM with CMC model the outcome of the estimation are 2 forward and 2 backward extrinsic A matrices which are defined as follows:<br>Element i,j of the matrix DCM.A{1} is the forward connection from superficial pyramidal population (region j) to the spiny stellate population (region i)<br>Element i,j of the matrix DCM.A{2} is the forward connection from superficial pyramidal population (region j) to the deep pyramidal population (region i)Element i,j of the matrix DCM.A{3} is the backward connection from deep pyramidal population (region j) to the superficial pyramidal population (region i)Element i,j of the matrix DCM.A{4} is the backward connection from deep pyramidal population (region j) to the inhibitory population (region i)<br>Otherwise a good explanation can be found in Figure 2 of the " Dynamic causal modelling revisited" paper (https://www.sciencedirect.com/science/article/pii/S1053811917301568?via%3Dihub).<br><br>I hope this would help<br><br>Best regards,<br>Amir.<br>-----Original Message-----<br>From: SPM (Statistical Parametric Mapping) <SPM@JISCMAIL.AC.UK> On Behalf Of Saurabh Sonkusare<br>Sent: 08 July 2019 00:27<br>To: SPM@JISCMAIL.AC.UK<br>Subject: [SPM] DCM cross-spectral densities- parameters interpretation<br><br>Hi DCM experts,<br><br>I have 2 nodes (iEEG data) for which I am employing DCM for cross spectral densities (CSD) with canonical microcircuit model (CMC). The model which is winning consisitently is the fully connected model (with A->B (forward and backward connections and B->A with forward and backward connections). I have not included the lateral connections and there is no condition trials (so no B effect) that I'm currently using. I'm trying to interpret the connectivity parameters (DCM.Ep.A) between 2 nodes for forward and backward connections. <br><br>The four columns in DCM.Ep.A are as this: [-32,-0.2821;0.02172,-32] [-32,0.8138;-0.3227,-32] [-32,-0.4286;-0.1337,-32] [-32,2.04397;-0.04991,-32]<br><br>I had expected only 2 matrices with -32 as constants and the other numbers corresponding to forward and backward connections. Can someone please tell me what all these numbers correspond to?<br><br>Thanks.<br><br>Best,<br>SaurabhMiriam Kampa2019-07-10T16:16:39+02:002019-07-10T16:16:39+02:00open academic positions university Siegen (Germany)https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;9b8a1007.1907The university in Siegen (Germany) currently offers two academic positions.<br>Please see attached announcements.Amirhossein Jafarian2019-07-10T14:14:30+00:002019-07-10T14:14:30+00:00Re: DCM cross-spectral densities- parameters interpretationhttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;79515a0f.1907Dear SPM user,<br><br>Thanks for your email.<br>In DCM for EEG/MEG using CMC model we need to tell the software about the forward and backward connections we would like to establish between regions. The rule for defining these matrices are the same as any other DCM. However in DCM with CMC model the outcome of the estimation are 2 forward and 2 backward extrinsic A matrices which are defined as follows:<br>Element i,j of the matrix DCM.A{1} is the forward connection from superficial pyramidal population (region i) to the spiny stellate population (region j)<br>Element i,j of the matrix DCM.A{2} is the forward connection from superficial pyramidal population (region i) to the deep pyramidal population (region j)Element i,j of the matrix DCM.A{3} is the backward connection from deep pyramidal population (region i) to the superficial pyramidal population (region j)Element i,j of the matrix DCM.A{4} is the backward connection from deep pyramidal population (region i) to the inhibitory population (region j)<br>Otherwise a good explanation can be found in Figure 2 of the " Dynamic causal modelling revisited" paper (https://www.sciencedirect.com/science/article/pii/S1053811917301568?via%3Dihub).<br><br>I hope this would help<br><br>Best regards,<br><br>Amir.<br><br>-----Original Message-----<br>From: SPM (Statistical Parametric Mapping) <SPM@JISCMAIL.AC.UK> On Behalf Of Saurabh Sonkusare<br>Sent: 08 July 2019 00:27<br>To: SPM@JISCMAIL.AC.UK<br>Subject: [SPM] DCM cross-spectral densities- parameters interpretation<br><br>Hi DCM experts,<br><br>I have 2 nodes (iEEG data) for which I am employing DCM for cross spectral densities (CSD) with canonical microcircuit model (CMC). The model which is winning consisitently is the fully connected model (with A->B (forward and backward connections and B->A with forward and backward connections). I have not included the lateral connections and there is no condition trials (so no B effect) that I'm currently using. I'm trying to interpret the connectivity parameters (DCM.Ep.A) between 2 nodes for forward and backward connections. <br><br>The four columns in DCM.Ep.A are as this: [-32,-0.2821;0.02172,-32] [-32,0.8138;-0.3227,-32] [-32,-0.4286;-0.1337,-32] [-32,2.04397;-0.04991,-32]<br><br>I had expected only 2 matrices with -32 as constants and the other numbers corresponding to forward and backward connections. Can someone please tell me what all these numbers correspond to?<br><br>Thanks.<br><br>Best,<br>Saurabh<br><br>-Auer, Tibor2019-07-10T14:42:28+01:002019-07-10T14:42:28+01:00FW: Marie Curie PhD studentship in developmental neurosciencehttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;10b4de11.1907Advertised on behalf of my colleague:<br><br>The Brain Imaging Center at the Research Centre for Natural Sciences of the Hungarian<br><br>Academy of Sciences is offering position for a doctoral students in a project aiming to<br><br>study neurodevelopmental disorders. This project will support<br><br>training in designing evidence-based, individualized treatments of learning, behavioural,<br><br>and social maladjustment, bridging across diagnostic categories. Towards these goals, we<br><br>have assembled a trans-sectoral European network with expertise in cognitive, social,<br><br>educational, clinical, and emotion research.<br><br>The post is a full-time and fixed term for one year with a possible renewal for<br><br>another two years, with an expected start date September 2, 2019. Salary will be €30.372 per<br><br>year, including comprehensive health insurance and social security plans. Salary will be<br><br>supplemented with Mobility Allowance (€600 per month). Qualified applicants based on family<br><br>status may receive an additional Family Allowance of €500 per month. All salaries are taxable<br><br>according to the Hungarian Tax Law.<br><br>Deadline for the applications is July 15, 2019.<br><br>For more information, please contact Dr. Ferenc Honbolygó, at the Brain Imaging Centre,<br><br>Research Centre for Natural Sciences, Hungarian Academy of Sciences (1117 Budapest,<br><br>Magyar tudósok krt. 2, Hungary). Email: honbolygo.ferenc@ttk.mta.hu <mailto:honbolygo.ferenc@ttk.mta.hu> .<br><br>Kind regards,<br><br>Tibor<br><br>Auer, Tibor M.D. Ph.D.<br><br>Research Fellow<br><br>School of Psychology, Faculty of Health and Medical Sciences<br><br>University of Surrey, Guildford GU2 7XH<br><br><mailto:T.Auer@surrey.ac.uk> T.Auer@surrey.ac.uk<br><br><https://eur02.safelinks.protection.outlook.com/?url=https%3A%2F%2Ftwitter.com%2FTiborAuer&data=02%7C01%7Ct.auer%40surrey.ac.uk%7Cd02ac63b3dde437ecf3308d6e8f54d5b%7C6b902693107440aa9e21d89446a2ebb5%7C0%7C0%7C636952537604301848&sdata=nb%2Beqr5a6CQjt2fEtYEvq%2FiAFpT3leQADJ1XU9DyRDQ%3D&reserved=0> @TiborAuerEugenio Abela2019-07-10T14:12:40+01:002019-07-10T14:12:40+01:00Re: SPM fMRI Preprocessing - Realignmenthttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;5599cbff.1907Hi Sophie<br><br>they are just flipped in time - have a close look at the image list at the top of your .ps documents. It seems that for “batch_with-dependencies” you entered volumes from 188,187… down to 1, whereas for “separate_steps” you did it the other way round, 1,2,…188.<br><br>Cheers<br><br>Eu<br><br>On 10 Jul 2019, at 13:57, Sophie-Marie Rostalski <sophie-marie.rostalski@UNI-JENA.DE> wrote:<br><br>Dear SPM experts,<br><br>I’m currently using SPM for the preprocessing of functional fMRI data.<br>During doing the preprocessing in different ways I recognized, that the Realignment Parameter file is given out in different ways, meaning that Realignment Parameters look different when I use dependencies in a full preprocessing batch rather than doing every step separately. I enclosed the two graphics for a better visualization.<br><br>Can you explain what causes these differences? Do I have to account for this issue when using the rp file for the first-level model specification?<br>Also, I was wondering whether the way of doing the preprocessing (separate steps vs batch or script with dependencies) has an effect on the dataset when writing out the realigned files and using them for further steps?<br><br>Thanks a lot in advance for the help!<br><br>Best wishes,<br><br>Sophie<br><br>Sophie-Marie RostalskiSophie-Marie Rostalski2019-07-10T14:57:35+02:002019-07-10T14:57:35+02:00SPM fMRI Preprocessing - Realignmenthttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;92e183ed.1907Dear SPM experts,<br><br>I’m currently using SPM for the preprocessing of functional fMRI data.<br>During doing the preprocessing in different ways I recognized, that the Realignment Parameter file is given out in different ways, meaning that Realignment Parameters look different when I use dependencies in a full preprocessing batch rather than doing every step separately. I enclosed the two graphics for a better visualization.<br><br>Can you explain what causes these differences? Do I have to account for this issue when using the rp file for the first-level model specification?<br>Also, I was wondering whether the way of doing the preprocessing (separate steps vs batch or script with dependencies) has an effect on the dataset when writing out the realigned files and using them for further steps?<br><br>Thanks a lot in advance for the help!<br><br>Best wishes,<br><br>Sophie<br><br>Sophie-Marie RostalskiMarco2019-07-10T13:55:20+01:002019-07-10T13:55:20+01:00Contrasts definition_Flexible Factorialhttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;f02b3dcb.1907Dear all,<br><br>I am computing the analysis for my data. I have 11 subjects divided in 2 groups and for each subject 5 timepoints. One group is the control group.<br>I am performing a Flexible Factorial analyses and I set 3 factors(1=subject, 2=group, 3=timepoint).<br><br>Listed all subject images, one by one (each timepoint) in "Specify Subjects or all Scans & Factors" and for each subject in first group the matrix results as:<br>1 1<br>1 2<br>1 3<br>1 4<br>1 5<br><br>and for subjects in second group:<br>2 1<br>2 2<br>2 3<br>2 4<br>2 5<br><br>I set Main Factor factor 1 (i.e.: subject) and Interaction for factors 2 and 3 (i.e.: group and timepoint).<br><br>Moving forward I am struggling with the contrasts definition. I performed a simple t-test on these data and over there the definition was quite simple (group1-group2, 1 -1).<br>But here, how do you advice me to defining the t-contrasts in this case?<br><br>Thank you all for help!<br>MarcoSimon Van Eyndhoven2019-07-10T14:48:22+02:002019-07-10T14:48:22+02:00Inference on GLM parameters when design matrix is derived from datahttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;6ff250ea.1907Dear neuroimaging experts,<br><br>I would like to solicit your advice on a statistical inference problem.<br><br>1. Goal: perform (non-parametric) inference on betas in linear model<br><br>Assume fMRI data in the form of time courses for a large set of regions<br>of interest (ROIs), represented as a matrix Y (scans x ROIs).<br>Equipped with a design matrix X, I estimate beta parameters for R<br>regressors via the model Y = X*Beta + residual, where Beta = (regressors<br>x ROIs). Then, I wish to perform non-parametric inference: for every<br>entry in Beta, I wish to answer the question "is the Beta(r,k)-th value<br>significantly different from zero?".<br><br>2. Method: based on permutation testing<br><br>My attempt is to use a permutation testing procedure, as described in<br>[1], with a wavelet-based data reshuffling scheme as in [2].<br>Essentially, I thus temporally reshuffle the data in Y, keeping X fixed,<br>and re-estimate the parameters Beta many times, to find an empirical<br>null distribution. As a test statistic, I use a T-like statistic: a beta<br>value Beta(r,k) divided by its standard deviation SD(Beta(r,k)), where<br>SD(Beta(r,k)) can be estimated by means of the covariance matrix for the<br>beta values in every ROI, i.e. sigma^2 * (X^T*X)^-1, with sigma^2 the<br>variance of the residual in the considered ROI k, as described in e.g.<br>[3]. (I've also tried the Pearson correlation coefficient between any<br>column of X and Y as test statistic.) This whole procedure is thoroughly<br>described in literature and performs adequately.<br><br>3. Problem statement: design matrix itself is estimated from the data<br>(?)<br><br>The crux in my case is that, contrary to most of the literature on e.g.<br>the General Linear Model, the matrix X is not an a priori fixed design<br>matrix, but rather _X is estimated from the data Y itself_, i.e. it is<br>found by solving a problem that looks like Y = X*S + residual, where<br>both X and S are unknown, and the residual needs to be minimized - think<br>of it as solving an indendent component analysis (ICA) on the data to<br>find relevant time courses and spatial maps (here there is no<br>experimental stimulus, the data is collected from epilepsy patients<br>under resting state). After applying the inference method outlined<br>above, the outcome is then often that "almost all entries in Beta are<br>significant, over nearly the whole brain". At least some of the<br>regressors in X are biomarker time courses for epilepsy, and hence it is<br>nonsensical that the whole brain would be 'active': a diseased,<br>epilepsy-related portion of the brain would comprise a few ROIs, for<br>instance.<br><br>_I think that this is a 'double dipping problem'_: I want to infer<br>whether the parameters Beta, estimated from the unshuffled data Y, are<br>significant, but these parameters correspond to regressors (in X) that<br>were_ themselves tuned to explain the data well_. Therefore, it perhaps<br>doesn't come as a great surprise that these values are generally all<br>'significant'.<br><br>4. Solution?<br><br>Is this intuition correct, and if so, what can be done to adjust the<br>inference procedure, and perform unbiased statistical tests in such<br>case? In my understanding, the problem is 'performing inference on<br>data-driven factor models', and I've looked for literature on inference<br>on e.g. ICA-derived time courses (the same problem would also occur in<br>that case) but have found nothing relevant... Any advice is welcome!<br><br>Thank you very much for your help!<br><br>PS: Things I tried already:<br><br>1) Check whether the autocorrelation structure of the residuals is<br>similar before/after reshuffling: ok, so the data are exchangeable under<br>the null hypothesis.<br>2) Checked with a design matrix that was fixed (not derived from the<br>data): ok, proper proportion of false positives.<br>3) Checked whether another resampling scheme solves anything: I tried<br>Fourier-based resampling, and the problem was the same.<br>4) Since several ROIs (~100) at once are tested, I use correction for<br>familywise error (FWE) by setting the significance threshold using the<br>distribution of the 'maximum T-statistic', as described in [4]. I<br>verified that this strongly controlled the FWE, for dummy regressors.<br><br>References:<br><br>[1] Nichols, Thomas E., and Andrew P. Holmes. "Nonparametric permutation<br>tests for functional neuroimaging: a primer with examples." _Human brain<br>mapping_ 15.1 (2002): 1-25.<br><br>[2] Bullmore, Ed, et al. "Colored noise and computational inference in<br>neurophysiological (fMRI) time series analysis: resampling methods in<br>time and wavelet domains." _Human brain mapping_ 12.2 (2001): 61-78.<br><br>[3] Poline, Jean-Baptiste, and Matthew Brett. "The general linear model<br>and fMRI: does love last forever?." _Neuroimage_ 62.2 (2012): 871-880.<br><br>[4] Nichols, Thomas, and Satoru Hayasaka. "Controlling the familywise<br>error rate in functional neuroimaging: a comparative review."<br>_Statistical methods in medical research_ 12.5 (2003): 419-446. PERNET Cyril2019-07-10T11:54:30+00:002019-07-10T11:54:30+00:00Re: A respiratory for fmri connectivity datahttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;2cee3f0d.1907this is the place at the moment: https://openneuro.org/<br><br>cyril Ashburner, John2019-07-10T09:45:28+00:002019-07-10T09:45:28+00:00Re: Does SPM handle complex value images?https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;96bf8046.1907There's no support in SPM for complex images. NIfTI can, in theory, encode complex data, but SPM's implementation of reading and writing NIfTI files can't handle it.<br><br>Best regards,<br>-John Didac Vidal-Pineiro2019-07-10T09:00:21+01:002019-07-10T09:00:21+01:00Postdoctoral position, University of Oslohttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;fa8a8162.1907I am delighted to advertise the following postdoctoral research fellowship on behalf of Prof Anders M Fjell:<br><br>At the Center for Lifespan Changes in Brain and Cognition, University of Oslo, Norway, we have an available postdoc position for an excellent candidate with strong skills in functional neuroimaging and linked to the project “The Missing Link of Episodic Memory Decline in Aging: The Role of Inefficient Systems Consolidation”. The position is for a period of 3 years.<br><br>The researcher will be able to work on uni- and multivariate analyses of high-quality task-related fMRI data acquired at our Siemens Prisma 3T MRI scanner using state-of-the-art sequences (multi-band/multi-echo EPI). Topics currently under investigation in our lab cover cognitive functions during episodic memory operations, as well as mechanisms subserving spatial navigation and spatial working memory.<br><br>The application deadline is: 18.08.2019 Please see more info here: https://www.jobbnorge.no/en/available-jobs/job/172827/postdoctoral-research-fellowship-in-functional-neuroimaging<br><br>P.S. Interested candidates are welcome to contact Anders M Fjell a.m.fjell@psykologi.uio.no<br>All the best,<br>DídacFeng Xu2019-07-09T15:26:58-04:002019-07-09T15:26:58-04:00Does SPM handle complex value images?https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;278d5256.1907Dear SPM users,<br>I am wondering if SPM handle images that have values in complex form. I<br>need to do motion correction and registration. I am not sure if how to save<br>complex image in .nii file either.<br><br>Thanks,<br>SusanSara De La Salle2019-07-09T13:17:42-04:002019-07-09T13:17:42-04:00Postdoctoral Position: Examining differences in neural profiles in males & females with EEG+fMRIhttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;790d092b.1907*Postdoctoral Position: Examining differences in neural profiles in males &<br>females with EEG+fMRI*<br><br>*The Lab*<br>Clinical EEG & Cognitive Research Laboratory, Institute of Mental Health<br>Research, affiliated with the University of Ottawa (Ottawa, ON, Canada).<br>https://eeglab-imhr.weebly.com/<br>https://uniweb.uottawa.ca/members/3376<br><br>Our laboratory uses clinical electroencephalography (EEG) and neuroimaging<br>techniques (e.g. functional magnetic resonance imaging), coupled with<br>various clinical and cognitive assessment tools, to better understand<br>mental illness. We also assess the effects of various interventions (e.g.<br>exercise, stimulation therapies, drugs) on neural, clinical and cognitive<br>features. Importantly, we attempt to characterize neural profiles that may<br>be predictive of treatment response. Much of our laboratory’s work has<br>focused on depression. However, we also examine brain features in<br>individuals with schizophrenia, ADHD, dysfunctional anger as well as<br>non-psychiatric populations.<br><br>*The Position*<br>The postdoctoral fellow will be tasked to conduct an NSERC-funded research<br>study examining gender differences in neural profiles between males/females<br>using simultaneous EEG+fMRI. Neural features will be examined in relation<br>to biological measures (estrogen, testosterone and cortisol).<br>The selected postdoctoral fellow will also have an opportunity to assist in<br>the acquisition/data analyses on one of two projects in clinical samples<br>(depressed, schizophrenia). Within this role, he/she will be expected to<br>assist Dr. Jaworska in the mentoring of undergraduate and graduate students<br>in the laboratory, in manuscript preparation and in grant writing.<br><br>*Academic qualifications and requirements:*<br>• PhD in neuroscience, psychology, biomedical science or related<br>disciplines.<br>• Demonstrated experience in human research and/or cognitive or<br>computational neuroscience.<br>Required: Neuroimaging analyses (preferred: SPM, CONN) or EEG analyses.<br>Some computational neuroscience experience preferred.<br>Optimal: Experience in simultaneous EEG+fMRI acquisition & analyses.<br><br>*Timeline and application process:*<br>Application deadline: open until the position is filled<br>Fellowship start date: Late fall 2019<br>Submission process: Documents should be emailed to Dr. Natalia Jaworska (<br>natalia.jaworska@theroyal.ca)<br><br>*Application checklist:*<br>1) A one (1) page research statement demonstrating fit with the program<br>described above.<br>2) Current academic CV demonstrating research excellence and a capacity for<br>leadership in the domain.<br>3) Contact information for two references from academic supervisors<br>/current employers.<br><br>*Value:*<br>The Postdoctoral Fellowship is a stipend valued at a starting salary of<br>$40,000 per year, commensurate on experienceAshburner, John2019-07-09T14:49:54+00:002019-07-09T14:49:54+00:00Re: Manual reorientation transformationhttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;4fa19542.1907Reorienting images changes the voxel to world mappings in the headers. If you just use specify translations and rotations, then the positioning information of the images will be rigidly rotated.<br><br>Best regards,<br>-John Joseph Tracy2019-07-09T14:24:23+00:002019-07-09T14:24:23+00:00Post-Doctoral Neuroimaging Fellowship in Epilepsyhttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;136a89f8.1907POST-DOCTORAL FELLOWSHIP IN THE NEUROIMAGING OF EPILEPSY<br><br>THOMAS JEFFERSON UNIVERSITY/ SIDNEY KIMMEL<br><br>MEDICAL COLLEGE, PHILA., PA.<br>The Department of Neurology at Thomas Jefferson University/Jefferson Medical College has an opening for a Postdoctoral Research Fellowship in Neuroimaging focusing on identification and predictive modeling of both seizure and cognitive/behavioral brain networks. This position is funded by a new 5 year NIH R01 grant aimed at developing multi-modal neuroimaging predictors of cognitive and seizure outcomes following brain surgery. The emphasis in our neuroimaging program is on cognitive and behavioral network organization and plasticity in neurologic disorders such as epilepsy. The lab focuses on the use of advanced network neuroscience measures to understand and characterize seizure networks and cognitive reorganization in epilepsy, and is part of the Jefferson Comprehensive Epilepsy Center, a large internationally-known epilepsy surgery program. Presurgical brain mapping studies are undertaken (MRI volumetrics, task-fMRI, resting state functional connectivity, diffusion imaging, neuropsychological assessment, electrophysiological recordings/electrocortical stimulation), as well as post-surgical neuroimaging studies investigating clinical, cognitive, and behavioral outcomes. The pre-surgical brain mapping studies are conducted on a regular basis, utilized in image-guided surgery, and then made available for research. Studies in brain recovery and the cognitive reorganization of language and memory functions are emphasized, utilizing rich multi-modal datasets for the investigation of both cognitive and pathologic (e.g., seizure) networks. Thomas Jefferson University provides a rich interdisciplinary research environment with grand rounds, seminars, case conferences, and opportunities to collaborate with faculty across departments such as neurosurgery and radiology. Successful applicants must have a strong background in image processing and analysis using programs such as MATLAB, SPM, and FSL with a strong interest in clinical neuroimaging and cognitive neuroscience. Applicants must have a PHD in neuroscience, biomedical engineering, statistics, neuropsychology, or related field. Interviews are being conducted at the upcoming OHBM 2019 conference in Rome, Italy.<br>Interested applicants should contact Joseph I. Tracy, Ph.D., ABPP(CN). Director, Cognitive Neuroscience and Brain Mapping Laboratory, Thomas Jefferson Univ./Sidney Kimmel Medical Coll., Jefferson Hospital for Neuroscience, 901 Walnut Street, Suite #447, Phila.,PA 19107, phone:#215-955-4661, e-mail: joseph.tracy@jefferson.edu<mailto:joseph.tracy@jefferson.edu>. Thomas Jefferson University and Hospitals is an equal Opportunity Employer. Jefferson values diversity and encourages applications from women, members of minority groups, LGBTQ individuals, disabled individuals, and veterans.<br><br>Joseph I. Tracy, Ph.D., ABPP/CN<br>Professor, Departments of Neurology and Radiology<br>Director, Neuropsychology Program<br>Director, Clinical Brain Mapping and Cognitive Neuroscience Laboratory<br>Fellow, American Psychological Association<br>joseph.tracy@jefferson.edu<mailto:joseph.tracy@jefferson.edu><br>Office: 215-955-4661<br>Assistant: 215-955-4676<br>Fax: 215-503-9475<br><br>The information contained in this transmission contains privileged and confidential information. It is intended only for the use of the person named above. If you are not the intended recipient, you are hereby notified that any review, dissemination, distribution or duplication of this communication is strictly prohibited. If you are not the intended recipient, please contact the sender by reply email and destroy all copies of the original message.<br><br>CAUTION: Intended recipients should NOT use email communication for emergent or urgent health care matters. Zeidman, Peter2019-07-09T14:22:22+00:002019-07-09T14:22:22+00:00Re: PEB group analysishttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;de245a34.1907Dear Sadjad<br>I will first clarify a couple of points in my previous answer, about how one would model a 3x3x3 design (thanks to Rik Henson for these this out). If you were to use a (1,0,-1) dummy variable encoding, you would be assuming that the corresponding factor is parametric and linear. If you wanted to relax that assumption – e.g. for the genotype factor – then you would need separate regressors for each level (or pairs of levels) of the factor. Second, if you want to allow for heteroscedasticity – i.e. different levels of between-subject variability in the different groups, then you either need to have separate covariance components per group (not currently implemented) or use the PEB-of-PEB approach described on the SPM Wiki.<br><br>Thank you so much for your reply. Regarding your question, actually not, for example, I have 67 healthy subjects in total with 3 sub-groups categorized according to their SNPs, 33 subjects with intermediate dopamine level, 22 with high and 12 with low dopamine level. Considering only this genotype, how can I group them and design my design matrix? Should I consider them as 3 groups? if yes I would appreciate if you tell me the design matrix I should consider for it. Or should I, for example, mix the subjects with an intermediate and high level of dopamine with each other and then compare them as one group with low level?<br><br>That’s clear. You’ve got a few options for how to model your three groups, e.g.:<br><br>1. A single PEB model (thereby assuming equal between-subject variability for all groups). Three columns the design matrix: for the first regressor, a vector of ones to encode the baseline (low dopamine group), then for the second regressor a dummy variable encoding membership of the intermediate group (1s and 0s), and for the third regressor a dummy variable encoding membership of the high dopamine group (1s and 0s).<br><br>2. As above, but instead of separate regressors for groups 2 and 3, model a linear effect of dopamine by encoding group membership in a parametric regressor (-1, 0, 1). You could also model higher order (non-linear) effects by adding regressors that are the square or cube of this regressor.<br><br>3. The PEB-of-PEBs approach, with one PEB model per group and then a PEB-of-PEBs to model the commonalities and differences between groups. This would enable different levels of between-subject variability per group (heteroscedasticity).<br><br>If you’re not sure which option is best, feel free to try different options and choose the one with the most positive free energy (PEB.F).<br><br>Best<br>Peter<br><br>Best,<br>Sadjad<br><br>On Tue, 9 Jul 2019 at 12:17, Zeidman, Peter <peter.zeidman@ucl.ac.uk<mailto:peter.zeidman@ucl.ac.uk>> wrote:<br>Dear Sadjad<br><br>I already have read the wiki page regarding the PEB analysis, (https://en.wikibooks.org/wiki/SPM/Parametric_Empirical_Bayes_(PEB)<https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fen.wikibooks.org%2Fwiki%2FSPM%2FParametric_Empirical_Bayes_(PEB)&data=02%7C01%7C%7Ccf2fc7cb6f364262bd4908d7045ae4d0%7C1faf88fea9984c5b93c9210a11d9a5c2%7C0%7C0%7C636982660744744838&sdata=xw2oW11xy0uc5z2wxxJUtY2XyN6grjWhpvFT8ZVnGN8%3D&reserved=0>), still, I am not sure which format of design matrix should I use for my analysis. I have 3 healthy groups with different genotypes, for one SNP I have subjects with higher dopamine level, intermediate dopamine level, and lower dopamine level, and for another SNP with 3 groups of high risk, intermediate risk and no risk of schizophrenia and the number of groups is unbalanced. How should I define my design matrix in these cases? Furthermore, Should I consider them as 3 groups or combine two of them as one group and then compare them as two groups?<br>Thanks a lot for your reply in advance.<br><br>To clarify, for the 3 healthy groups, is it the case that within each group you have a mixture of high, intermediate and low dopamine subjects? And the SAME subjects have either high, intermediate or no risk of schizophrenia?<br><br>If that’s right, your design can be described as 3 x 3 x 3, with factors of genotype (groups 1-3), dopamine level (high, intermediate, low) and schizophrenia risk (high, intermediate, low). From this, you know that you’ve got three main effects, three 2-way interactions and one 3-way interaction, plus the average effect over all subjects. If you want to fully represent your design in the PEB model, you’ll therefore need 8 regressors. Your first column would be all ones to represent the mean. You could encode the main effects using 1s, -1s and 0s, which you then mean-centre. The interactions are computed by element-wise multiplying the relevant main effect regressors.<br><br>NB If you want to use the tool for comparing specific pre-define PEB reduced models, then you’ll need to re-order the regressors such that the regressor of interest appears second in the design matrix (i.e. after the column of ones).<br><br>Do let me know if anything remains unclear.<br><br>Best<br>PeterBill Budd2019-07-09T13:35:14+00:002019-07-09T13:35:14+00:00Problem reading EDF+ event markers using spm_eeg_converthttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;82f7300a.1907Hi All<br><br>Just thought I’d summarise workarounds for this issue I reported with EDF+ following off-list discussions with Vladimir in case its helpful.<br><br>Main issue was that EDF+ events/trigger markers could not be converted by SPM/fieldtrip. This is because they are stored in EDF+ in a dedicated ‘annotation channel’ rather than as a continuous time series of binary TTL trigger values as commonly used. So these appear to have a different sampling rate and format (for more info see http://www.fieldtriptoolbox.org/getting_started/edf/).<br><br>In the above link is a solution in fieldtrip that may work to read the EDF+ events into SPM format:<br>event = ft_read_event('testfile.edf', 'detect flank', [])<br>however this didn’t work for me presumably because the exact format of the EDF+ Annotation Channel is not standardised/difficult to predict.<br><br>So as a workaround I used EDFbrowser (https://www.teuniz.net/edfbrowser/) to extract the annotation channel (ie events) from the original EDF+ file as a text file. This data can then be used to create the trl matrix and conditions label list for epoching etc. in SPM12 (See section 12.8.1 in the manual) or alternatively, write a script to create the SPM event structure and then incorporate into the converted SPM data.<br><br>Other EDF+ issue was an error returned by spm_eeg_convert (v7451) :<br><br>Error using ft_scalingfactor (line 178)<br>Error: Expression or statement is incomplete or incorrect.<br>Error in ft_read_data (line 1509)<br>scaling = cellfun(@ft_scalingfactor, hdr.chanunit(chanindx(:)), chanunit(:));<br>Error in spm_eeg_convert (line 427)<br>dat = ft_read_data(S.dataset,'header', hdr, 'begsample', trl(i, 1), 'endsample', trl(i, 2),...<br><br>This seems to occur where data units are outside the scope of ft_read_data.m but can be fixed by setting the ‘scaling’ variable in ft_read_data.m (line 1509) to ‘1’ – which means all data units used in the original data file are retained (unscaled).<br><br>Hope this is useful to someone and very much appreciate if anyone has more advice about converting EDF+ data using SPM/fieldtrip!<br><br>Cheers<br>-Bill<br><br>Cheers<br>-Bill<br><br>From: Bill Budd<br>Sent: Thursday, 4 July 2019 12:32 AM<br>To: Vladimir Litvak <litvak.vladimir@GMAIL.COM>; SPM@JISCMAIL.AC.UK<br>Subject: RE: [SPM] The first MBT workshop (9/30/19-10/1/19) @Belgrade<br><br>Dear Vladimir<br><br>I’d like to use spm_eeg_convert to convert EDF+ format EEG data to SPM format but haven’t been able to get it to read the event data. I have confirmed events are included in the original EDF+ data using an EDF browser.<br><br>Initially I used spm_eeg_convert (v6244) and the data converted into SPM12 format fine – just no event data.<br><br>So I upgraded to the latest version of SPM12 (spm_eeg_convert (v7451)) to see if that would help but now get the following error message:Marco2019-07-09T13:32:34+01:002019-07-09T13:32:34+01:00Manual reorientation transformationhttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;b4442824.1907Dear SPM expert,<br><br>I have a fast qeustion but I think it would be crucial for me to understand.<br><br>I am working with cat brains images from different time points. These images are very not oriented one to each other so I opted for manually reorient them.<br>The question is:<br>-what kind of transformation is applied to the images when manual reorientation is performed? (rigid, non linear, affine..?)<br>Knowing this, I would understand better what is happening to voxels or img dimension.]<br><br>Thank for your help,<br>MarcoChristian N.2019-07-09T12:01:18+01:002019-07-09T12:01:18+01:00Re: CAT12 stopping after preprocessing errorshttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;33cd1450.1907Hello again,<br><br>I commented the line 29 from the 'cat_run_job.m' script (clearvars -global cat_err_res;) and now the ignoreErrors flag works.<br><br>Everything seems fine, but do you think that commenting that line could have any negative consequence?<br><br>Thank you,<br><br>Christian.Sadjad Sadeghi2019-07-09T12:47:39+02:002019-07-09T12:47:39+02:00Re: PEB group analysishttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;84a8973d.1907Dear Peter,<br><br>Thank you so much for your reply. Regarding your question, actually not,<br>for example, I have 67 healthy subjects in total with 3 sub-groups<br>categorized according to their SNPs, 33 subjects with intermediate dopamine<br>level, 22 with high and 12 with low dopamine level. Considering only this<br>genotype, how can I group them and design my design matrix? Should I<br>consider them as 3 groups? if yes I would appreciate if you tell me the<br>design matrix I should consider for it. Or should I, for example, mix the<br>subjects with an intermediate and high level of dopamine with each other<br>and then compare them as one group with low level?<br><br>Best,<br>Sadjad<br><br>On Tue, 9 Jul 2019 at 12:17, Zeidman, Peter <peter.zeidman@ucl.ac.uk> wrote:<br><br>> Dear Sadjad<br>><br>><br>><br>> I already have read the wiki page regarding the PEB analysis, (<br>> https://en.wikibooks.org/wiki/SPM/Parametric_Empirical_Bayes_(PEB)<br>> <https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fen.wikibooks.org%2Fwiki%2FSPM%2FParametric_Empirical_Bayes_(PEB)&data=02%7C01%7C%7Ce456c0e5e737470524a008d704456016%7C1faf88fea9984c5b93c9210a11d9a5c2%7C0%7C0%7C636982568350360851&sdata=r4Fp2ZQnB3HuzaoEGVcer9qD6S9p5xKA%2Bs7thqx%2FN7A%3D&reserved=0>),<br>> still, I am not sure which format of design matrix should I use for my<br>> analysis. I have 3 healthy groups with different genotypes, for one SNP I<br>> have subjects with higher dopamine level, intermediate dopamine level, and<br>> lower dopamine level, and for another SNP with 3 groups of high risk,<br>> intermediate risk and no risk of schizophrenia and the number of groups is<br>> unbalanced. How should I define my design matrix in these cases?<br>> Furthermore, Should I consider them as 3 groups or combine two of them as<br>> one group and then compare them as two groups?<br>><br>> Thanks a lot for your reply in advance.<br>><br>><br>><br>> To clarify, for the 3 healthy groups, is it the case that within each<br>> group you have a mixture of high, intermediate and low dopamine subjects?<br>> And the SAME subjects have either high, intermediate or no risk of<br>> schizophrenia?<br>><br>><br>><br>> If that’s right, your design can be described as 3 x 3 x 3, with factors<br>> of genotype (groups 1-3), dopamine level (high, intermediate, low) and<br>> schizophrenia risk (high, intermediate, low). From this, you know that<br>> you’ve got three main effects, three 2-way interactions and one 3-way<br>> interaction, plus the average effect over all subjects. If you want to<br>> fully represent your design in the PEB model, you’ll therefore need 8<br>> regressors. Your first column would be all ones to represent the mean. You<br>> could encode the main effects using 1s, -1s and 0s, which you then<br>> mean-centre. The interactions are computed by element-wise multiplying the<br>> relevant main effect regressors.<br>><br>><br>><br>> NB If you want to use the tool for comparing specific pre-define PEB<br>> reduced models, then you’ll need to re-order the regressors such that the<br>> regressor of interest appears second in the design matrix (i.e. after the<br>> column of ones).<br>><br>><br>><br>> Do let me know if anything remains unclear.<br>><br>><br>><br>> Best<br>><br>> Peter<br>>Zeidman, Peter2019-07-09T10:17:02+00:002019-07-09T10:17:02+00:00Re: PEB group analysishttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;222f5f2f.1907Dear Sadjad<br><br>I already have read the wiki page regarding the PEB analysis, (https://en.wikibooks.org/wiki/SPM/Parametric_Empirical_Bayes_(PEB)<https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fen.wikibooks.org%2Fwiki%2FSPM%2FParametric_Empirical_Bayes_(PEB)&data=02%7C01%7C%7Ce456c0e5e737470524a008d704456016%7C1faf88fea9984c5b93c9210a11d9a5c2%7C0%7C0%7C636982568350360851&sdata=r4Fp2ZQnB3HuzaoEGVcer9qD6S9p5xKA%2Bs7thqx%2FN7A%3D&reserved=0>), still, I am not sure which format of design matrix should I use for my analysis. I have 3 healthy groups with different genotypes, for one SNP I have subjects with higher dopamine level, intermediate dopamine level, and lower dopamine level, and for another SNP with 3 groups of high risk, intermediate risk and no risk of schizophrenia and the number of groups is unbalanced. How should I define my design matrix in these cases? Furthermore, Should I consider them as 3 groups or combine two of them as one group and then compare them as two groups?<br>Thanks a lot for your reply in advance.<br><br>To clarify, for the 3 healthy groups, is it the case that within each group you have a mixture of high, intermediate and low dopamine subjects? And the SAME subjects have either high, intermediate or no risk of schizophrenia?<br><br>If that’s right, your design can be described as 3 x 3 x 3, with factors of genotype (groups 1-3), dopamine level (high, intermediate, low) and schizophrenia risk (high, intermediate, low). From this, you know that you’ve got three main effects, three 2-way interactions and one 3-way interaction, plus the average effect over all subjects. If you want to fully represent your design in the PEB model, you’ll therefore need 8 regressors. Your first column would be all ones to represent the mean. You could encode the main effects using 1s, -1s and 0s, which you then mean-centre. The interactions are computed by element-wise multiplying the relevant main effect regressors.<br><br>NB If you want to use the tool for comparing specific pre-define PEB reduced models, then you’ll need to re-order the regressors such that the regressor of interest appears second in the design matrix (i.e. after the column of ones).<br><br>Do let me know if anything remains unclear.<br><br>Best<br>PeterAshburner, John2019-07-09T09:25:36+00:002019-07-09T09:25:36+00:00Re: Generate job file for Fieldmap VDM (Voxel Displacement Map)https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;3abbee12.1907If you use the toolbox via the Batch system, you should be able to save things in a job file. I'd suggest setting up a job for a single subject, saving as a .m file, and then basing some sort of script on the saved .m file.<br><br>Best regards,<br>-John Christian Keysers2019-07-09T08:56:59+00:002019-07-09T08:56:59+00:00Why is FDR thresholding with GUI at 0.05 and FDRp different?https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;9fdf3853.1907We are trying to automatise FDR voxelwise thresholding of maps at q=0.05 (to explore its ability to control false positive rates on random data)<br>The way we would normally see if any data survives FDR thresholding would be to set defaults.stats.topoFDR=0, and then use the GUI, click the FDR threshold button on the GUI, and enter 0.05. The resulting results screen then indicates the critical T value as for instance 1.77<br>Then on the bottom of the results screen we also get the filed FDRp. I thought this would be the critical T value to control q=0.05. and that it would be the same as the threshold set by the program if I use FDR, 0.05 in the GUI - but it is not. Its about 3.87 in our example.<br>You can see a screenshot of the results screen at https://www.dropbox.com/s/b3nr5dkz0ba8leh/SPM_FDR.jpg?dl=0<br>Does anyone know why the two are different?<br>In contrast, if I use fwe=0.05 in the GUI, the T threshold used by SPM to display results and the FWEp on the bottom of the results screen is identical…<br>And to automatize the whole proceedure, would someone have an example of a spm_getspm syntax that would take an SPM.mat and give me the xSPM.u value that I could use to threshold the image?<br>Thank you so much<br>ChristianSadjad Sadeghi2019-07-09T10:13:36+02:002019-07-09T10:13:36+02:00PEB group analysishttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;cbdedff0.1907Dear Peter,<br><br>I already have read the wiki page regarding the PEB analysis, (<br>https://en.wikibooks.org/wiki/SPM/Parametric_Empirical_Bayes_(PEB)), still,<br>I am not sure which format of design matrix should I use for my analysis. I<br>have 3 healthy groups with different genotypes, for one SNP I have subjects<br>with higher dopamine level, intermediate dopamine level, and lower dopamine<br>level, and for another SNP with 3 groups of high risk, intermediate risk<br>and no risk of schizophrenia and the number of groups is unbalanced. How<br>should I define my design matrix in these cases? Furthermore, Should I<br>consider them as 3 groups or combine two of them as one group and then<br>compare them as two groups?<br>Thanks a lot for your reply in advance.<br><br>Best,<br>SadjadLeehe Peled-Avron2019-07-09T10:28:36+03:002019-07-09T10:28:36+03:00Field Map recurring errorhttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;94dba37e.1907Dear SPM experts,<br><br>We have this problem when we try to incorporate fieldmap corrections to our<br>preprocessing analyses (Please see script pasted down).<br>We tried using SPM imcalc functionality on our structural volumes with<br>entering the expression max(0,i1) but it didn't work. Any suggestion how to<br>solve this?<br><br>Thank you so much!<br>Leehe.<br><br>Script:<br>Running 'Calculate VDM'<br>Segmenting and extracting brain...<br>Failed 'Calculate VDM'<br>Error using schur<br>Input to SCHUR must not contain NaN or Inf.<br>In file "C:\Program Files\MATLAB\R2014b\toolbox\matlab\matfun\sqrtm.m"<br>(???), function "sqrtm" at line 32.<br>In file "C:\toolbox\spm12\toolbox\FieldMap\pm_segment.m" (v4873), function<br>"get_p" at line 581.<br>In file "C:\toolbox\spm12\toolbox\FieldMap\pm_segment.m" (v4873), function<br>"run_segment" at line 364.<br>In file "C:\toolbox\spm12\toolbox\FieldMap\pm_segment.m" (v4873), function<br>"pm_segment" at line 119.<br>In file "C:\toolbox\spm12\toolbox\FieldMap\pm_brain_mask.m" (v5014),<br>function "pm_brain_mask" at line 48.<br>In file "C:\toolbox\spm12\toolbox\FieldMap\FieldMap.m" (v6416), function<br>"FieldMap" at line 1616.<br>In file "C:\toolbox\spm12\toolbox\FieldMap\FieldMap_create.m" (v6504),<br>function "FieldMap_create" at line 146.<br>In file "C:\toolbox\spm12\toolbox\FieldMap\FieldMap_Run.m" (v6656),<br>function "FieldMap_Run" at line 140.<br>In file "C:\toolbox\spm12\toolbox\FieldMap\tbx_cfg_fieldmap.m" (v6501),<br>function "FieldMap_calculatevdm" at line 766.Fernando Calamante2019-07-09T02:56:45+00:002019-07-09T02:56:45+00:00Senior Lecturer in MRI Methods for Neuroscience (University of Sydney, Australia)https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;bb21f53e.1907Senior Lecturer in Clinical MRIMethods for Neuroscience<br><br>University ofSydney (Australia)<br><br> <br><br>- Full-time, fixed term for 5 years, Academic Level C:<br>- Base Salary: $127,585 - $147,115 p.a. (plus generous employer’s contribution to superannuation)<br>- Closing date: 11:30pm, 11 August 2019 (Sydney Time) <br><br> <br><br>About the opportunity: We are seeking to appoint aSenior Lecturer (Level C academic) to be responsible for a range of taskswithin the lab of Professor Fernando Calamante, including implementation andfurther development of novel MRI technology (MR image acquisition, processing,and/or modelling), especially related to study brain connectivity, structure,function and quantification of tissue properties.<br><br>Please refer to the link below:<br><br>Ref1361/0719F - Senior Lecturer in Clinical MRI Methods for NeuroscienceRan Wang2019-07-08T17:15:21-07:002019-07-08T17:15:21-07:00Generate job file for Fieldmap VDM (Voxel Displacement Map)https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;4d2267fa.1907Hi SPM experts,<br><br>I noticed that in the SPM GUI, every time creating the Fieldmap VDM (Voxel<br>Displacement Map) from the toolbox requires typing parameters by hand. I<br>was wondering if there is a way to make a job file for this toolbox<br>function of FieldMap? Any suggestion would be super helpful. I appreciate<br>your considerations and thanks so much for reading this!<br><br>Best,<br><br>*Ran Wang, M.S.*<br>Junior Technical Specialist<br>UC Davis Imaging Research Center<br>UC Davis Health System<br>4701 X Street<br>Sacramento, CA 95817<br>Phone: (916)840-4529Ramesh Babu2019-07-08T22:23:28+05:302019-07-08T22:23:28+05:30Re: TFCE analysishttps://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;29da63d6.1907Dear Christian,<br>I did TFCE analysis as follows;<br>1. Initial preprocessing segmentation - DARTEL approach<br>3. Quality check<br>2. Smoothed - 8 mm<br>3. Estimated<br>5. Result - Defined contrast.<br>4. TFCE - Estimate<br>- selected SPM.mat<br>- Contrast index - inf<br>- the number of permutation 1000 initially, and 5000 later.<br>5. TFCE - Result showed FWE significant regions.<br>6. Small volume correction<br>I would like to use a mask for small volume correction. For example union<br>of 8 mm (since smoothing kernel was 8 mm, same size mask was created) mask<br>in the frontal region.<br><br>Please give me your suggestions on whether I am in the right direction? If<br>there is any correction please let me know.<br><br>Thanks<br>RB<br><br>On Sat, Jul 6, 2019 at 7:21 AM Ramesh Babu <mgrameshbabu2013@gmail.com><br>wrote:<br><br>> Dear Christian,<br>> Thank you for the clarification. I am doing the analysis I will contact<br>> you if I get any doubt further.<br>><br>> Thanks<br>> Ramesh<br>><br>> On Fri, Jul 5, 2019 at 4:44 AM Christian Gaser <<br>> christian.gaser@uni-jena.de> wrote:<br>><br>>> Dear Ramesh,<br>>><br>>> On Thu, 4 Jul 2019 10:25:26 +0530, Ramesh Babu <<br>>> mgrameshbabu2013@GMAIL.COM> wrote:<br>>><br>>> >Dear Experts,<br>>> ><br>>> >I need a manual explaining the usage of TFCE. A very few details are<br>>> >available in CAT12 manual, but this not sufficient as you proceed for<br>>> TFCE.<br>>> >In the first step for estimation, there is a doubt whether minimally<br>>> >smoothed or no smoothed images should be used.<br>>> Use the same images you have used in your parametric analysis. The<br>>> SPM.mat file contains all the information and after estimating a parametric<br>>> test (model) you can simply use this SPM.mat file for TFCE (after defining<br>>> contrasts).<br>>><br>>> Then as I proceed further in<br>>> >the result section, the result window display to select "TFCE or T" and<br>>> If you are interested in TFCE, select TFCE. The option "T" is just for<br>>> the sake of completeness and is a non-parametric T-test.<br>>> >then "original or inverse". Now which one to select?<br>>> "Original" for your contrast (e.g. sample 1 > sample 2) and "inverse" for<br>>> the opposite (inverse) contrast (e.g. sample 2 > sample 1) without<br>>> estimating the TFCE permutations again.<br>>><br>>> Best,<br>>><br>>> Christian<br>>><br>>> ><br>>> >If you have TFCE manual, could you please share with me?<br>>> ><br>>> >Thank you<br>>> >Ramesh<br>>> ><br>>><br>>><br>>>