 |
 |
Dear Anderson, Honestly, I can't find words to thank you for your patience and your great kindness. Thank you very much for teaching me how to do it! I am happy that I have learned something new ;-) With all respect Jon Hi Jon, Please, see below: On 7 September 2016 at 00:07, John anderson <[log in to unmask]> wrote: Dear Anderson, I highly appreciate this great guidance. I have in my data set 80 DTI images ( 50 patients and 30 healthy controls). I followed your suggestions and I divided the images depedning on aquisiotn parameters into five groups as follow: TE=110, TR=2750 for 20 images TE=100, TR= 2500 for 25 images TE=106, TR=2600 for 23 images TE=103, TR=2600 for 8 images TE=112, TR=2750 for 4 images Kindly, 1. I will include these TE, TR subgroups as -2,-1,0,1,2 in the model. Is this correct? Not in this case. With 5 protocols, one way of coding is with 4 EVs as nuisance variables, these being: EV1: -1 for protocol 1, 1 for protocol 2, 0 for all others. EV2: -1 for protocol 1, 1 for protocol 3, 0 for all others. EV3: -1 for protocol 1, 1 for protocol 4, 0 for all others. EV4: -1 for protocol 1, 1 for protocol 5, 0 for all others. 2. I will include the groups (patients and controls) as 1 , -1 in the model?. Is this correct? Yes. Include also an intercept: EV5: +1/-1 for group (this is the EV of interest that will be tested with contrasts). EV6: intercept. EV7, EV8, etc: other possible nuisance variables, e.g., age, sex, etc. All the best, Anderson In this model I will give 1 and -1 for two EVs, one for a group, and the other one for a covariate. I am confused a little bit about this point! Is this correct? PS: also note that simply entering TE and TR separately in the design would ignore the possibility of interaction between these. On 6 September 2016 at 07:55, Anderson M. Winkler <[log in to unmask]> wrote: Hi Jon, While it is certainly possible to include the actual TE and/or TR in the GLM, the FA is likely a non-linear function of these parameters (and even so, possibly a quite subtle one). Instead, how many unique combinations of TE and TR you have? That is, how many distinct acquisition protocols you have? In a common scenario you'd have only N=2 protocols, thus adding a single EV that is +1 or -1. If there are N>2 protocols, then there are multiple ways to construct such EVs, and as simple one is to add N-1 EVs to the design, then for the 2nd protocol onwards, create an EV marking as +1 the subjects that were scanned with the 1st protocol, -1 the subjects scanned with the current protocol, and 0 for all other subjects. Another option is, if the model includes an explicit intercept (e.g., a column full of ones), remove such intercept, and replace it with N new EVs, one per protocol, each coded as +1 for that protocol, and 0 for all others. However, if there are many different TEs, many different TRs, and a large number of combinations thereof (i.e., many protocols used), then a more parsimonious model might be to do the same as above, but with EVs entered separately for each TE and each TR, instead of EVs for each protocol. All the best, Anderson On 5 September 2016 at 13:30, John anderson <[log in to unmask]> wrote: Hi Anderson, I highly appreciate your response, and the options that you provided. Kindly, I would like to inquire about the EV method, and be certain that I understood your intention properly. In my database I have 80 DWI images (50 patients, and 30 healthy controls). The images were acquired using a protocol with a minimal echo time (TE) difference between the images. As a results the images were acquired with a range of TE values (100-110), and the range for repetition time TR (2500-2750) I want to run TBSS analysis between these two groups of subjects by including all the images. I ran the preprocessing steps as suggested by FSL and I geenrated all the FA maps using the tool (FDT) Then I followed the steps suggested in TBSS pipeline. Now, I kindly wanted to know that I understood what Steve and you were mentioned regarding EV: In Glm, I will be adding EV for TE, and another one for TR. I will include the actual values for TE and TR for every scan in EV column in Glm. Please is this correct? Do I need to demean these values in Glm? Thank you very much! Jon Hi Jon, Just adding to this: the effect of intensity normalisation will be captured by the protocol-specific EV as Steve indicated. Further, if the variances between your protocols also vary, a second option is, in addition to the extra EV, to also define one exchangeability block and one variance group per protocol, and run the analysis in PALM without the assumption of homogeneous variances. A third option is to run a data transformation per protocol. In this paper (disclaimer: I'm a co-author) a rank-based inverse normal transformation was applied per cohort. This has the additional effect of stabilising the variances and removing outliers, and removes the need for subject-specific EVs and variance groups, but is also likely to be slightly more conservative than the other approaches. A Matlab/Octave function that does the transformation is available in my blog, here. All the best, Anderson On 4 September 2016 at 15:15, Rosalia Dacosta Aguayo <[log in to unmask]> wrote: Hi Jon, Thank you a lot for your explanation. Yours sincerely, Rosalia 2016-09-04 15:40 GMT+02:00 Jon Anderson <[log in to unmask]>: Although some images (e.g. DTI, PET) are more sensitive to the changes in the acquisition parameters. But this is applicable to all imaging data. e.g. if you have T1 images acquired with different acquisition parameters then this will affect the size of the matrix, and the slices thickness. As a result, you will not be able at the end of the analysis to differentiate if the differences between the study groups are related to a pathology, or it is because of the difference in acquisition parameters. The reason why I am looking for a method to normalize the images acquired using different acquisition parameters to include them in the same pipeline. Hi Stephen, I have been reading the link. Is this applicable to MRI data or just only for DTI data?...and..why? Yours sincerely, Rosalia 2016-09-03 12:56 GMT+02:00 Stephen Smith <[log in to unmask]>: Hi - see the FAQ entry: http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/TBSS/Faq Cheers. On 3 Sep 2016, at 11:52, Antoine Nasimian <[log in to unmask]> wrote: Hi Steve, I highly appreciate your response. I have DTI data aquired using different aquisition parameters ( same b value, but different TE and TR). I was thinking of kind of normalization for the images permits of including it, regardless the differences in TE, TR in the same analysis (e.g TBSS) I know aquision parameters like TE can affect FA value. How can I include different FA maps (i.e FA maps generated from DWI aquired using different aquisition parameters) in the same TBSS analysis. Can intensity normalization be of benifits in here? Best, Ant Hi On 2 Sep 2016, at 12:18, Jon Anderson <[log in to unmask]> wrote: Dear FSL experts, I would like to ask the following questions about " intensity normalization in FSL": 1. In which circumstances, intensity normalization (i.e. the flag -inm in the command " fslmaths") is urgent? I'm not sure what you're asking here, sorry - but maybe the below answer gives you what you need. 2. What type of images can be normalized? Are there any benefits of intensity normalization in DTI data? In general this is NOT done for inputs to TBSS, as the various dtifit outputs (eg FA, MD) are supposed to already be directly comparable across subjects as long as all subjects were acquired in the same way. Cheers. 3. Can I use the flag "-inm "in the command "fslmaths" in the script "tbss_3_postreg" to normalize the images "all_FA" then continue with the rest of the pipeline? I am wondering if the image "all_FA" is 2D or 3D? Thanks for any comment and for your help! Jon --------------------------------------------------------------------------- Stephen M. Smith, Professor of Biomedical Engineering Head of Analysis, Oxford University FMRIB Centre FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK +44 (0) 1865 222726 (fax 222717) [log in to unmask] http://www.fmrib.ox.ac.uk/~steve --------------------------------------------------------------------------- Stop the cultural destruction of Tibet -- Rosalía Dacosta-Aguayo, Post-Doctoral Researcher. Biomedicine, PhD Clinical Neuropsychologist Department of Neurosciences. Group of Neurodegenerative Diseases Gaixotasun Neuroendekapenezko Taldea Neurozientziak Arloa [log in to unmask] T. +34 943 006 128 IIS BIODONOSTIA - OSI DONOSTIALDEA Pº Dr. Beguiristain s/n * 20014 Donostia-San Sebastián SPAIN * T.+34 943 006 012 * F. +34 943 006 250 * [log in to unmask] * www.biodonostia.org -- Rosalía Dacosta-Aguayo, Post-Doctoral Researcher. Biomedicine, PhD Clinical Neuropsychologist Department of Neurosciences. Group of Neurodegenerative Diseases Gaixotasun Neuroendekapenezko Taldea Neurozientziak Arloa [log in to unmask] T. +34 943 006 128 IIS BIODONOSTIA - OSI DONOSTIALDEA Pº Dr. Beguiristain s/n * 20014 Donostia-San Sebastián SPAIN * T.+34 943 006 012 * F. +34 943 006 250 * [log in to unmask] * www.biodonostia.org