 |
 |
Moran, try the following command with your image. More info if you type
simply 'fslmaths' with no arguments, at the command line.
fslmaths <input_image.nii> -fillh <output_image.nii>
- Eli
On 05/22/2016 06:00 PM, FSL automatic digest system wrote:
> There are 10 messages totaling 1403 lines in this issue.
>
> Topics of the day:
>
> 1. zero filling (3)
> 2. No results with VBM (3)
> 3. BEDPOSTX on data with replicated b-vectors in one dataset (3)
> 4. probabilistic tractography
>
> ----------------------------------------------------------------------
>
> Date: Sun, 22 May 2016 06:42:17 +0100
> From: Moran Artzi <[log in to unmask]>
> Subject: zero filling
>
> Hi,
> Is there any way to fill the holes in a binary image (as the brain_mask img - the output of bet) using FSL command line?
> Thanks
> Moran
>
> ------------------------------
>
> Date: Sun, 22 May 2016 09:20:06 +0100
> From: "Anderson M. Winkler" <[log in to unmask]>
> Subject: Re: No results with VBM
>
> Hi Rosalia,
>
> Please see below:
>
> On 21 May 2016 at 19:30, Rosalia Dacosta Aguayo <[log in to unmask]> wrote:
>
>> Dear Anderson,
>>
>> The data has been demeaned, and I am working with just one group of 45
>> subjects, I am not sure about including an intercept in this model. Just
>> partial correlations between this group and a cog variable, regressing out
>> for age, gender and TIV....
>>
> It's necessary to demean both design and data, or else, include an
> intercept. Mean-centering the data only isn't sufficient.
>
>
>> And about my question regarding FIRST? I do not understand why I can get
>> .vtk and .bvars files for my 4. ROIS but when I give a look into firstseg
>> files and I check for registeting....I only find two of the deep gm
>> structures but not the 4 previosly selected.
>>
> This refers to a different thread. I'm not sure what is going on.
>
>
>> What is more, I am not quite sure about surface and vertex or shape
>> analysis and how to interpret results...
>>
> The shape analysis indicates the places where there is a significant shape
> difference (i.e., a significant effect on the positions of the vertices
> that delineate the structure).
>
> All the best,
>
> Anderson
>
>
>> I would greatly appreciate your helping hear.
>>
>> Kind regards,
>>
>> Rosalia
>> El 20/05/2016 11:44, "Rosalia Dacosta Aguayo" <[log in to unmask]>
>> escribió:
>>
>>> Thank you Anderson!!
>>>
>>> Rosalia
>>>
>>> 2016-05-20 10:07 GMT+02:00 Anderson M. Winkler <[log in to unmask]>:
>>>
>>>> Hi Rosalia,
>>>>
>>>> Just adding to this: need to include an intercept in the model, or use
>>>> the option -D.
>>>>
>>>> All the best,
>>>>
>>>> Anderson
>>>>
>>>>
>>>> On 19 May 2016 at 16:10, Colin Hawco <[log in to unmask]> wrote:
>>>>
>>>>> While I am afraid I can offer no specific advice for VBM in FSL, I do
>>>>> have a general piece of advice.
>>>>>
>>>>>
>>>>>
>>>>> When I see a null result I think is suspicious, I immediately drop
>>>>> threshold and take a look. As instead, instead of p < 0.05 corrected, try
>>>>> viewing the t-map without a threshold. . See if there looks like an
>>>>> appropriate level of variability in the t-stats, or if it looks off (zero
>>>>> at many data points, really speckled).
>>>>>
>>>>>
>>>>>
>>>>> This is a way to QC check the output and see if maybe something went
>>>>> wrong which caused the stats to mess up.
>>>>>
>>>>>
>>>>>
>>>>> Good luck
>>>>>
>>>>>
>>>>>
>>>>> Colin Hawco, PhD
>>>>>
>>>>> Neuranalysis Consulting
>>>>>
>>>>> Neuroimaging analysis and consultation
>>>>>
>>>>> https://urldefense.proofpoint.com/v2/url?u=http-3A__www.neuranalysis.com&d=DQIFaQ&c=QmPtDiFixEjkMvDKaP3E2Vb9C2z4M0PdarxyAHQ2iDQ&r=xbyrBxm81l6mG_XEX66jgCNTXfK3eVo30T9sVQVTz1Q&m=Jc1PNVoYrsBQ4MCSXUR-BueqHdpSGdf2yPHWfNLLkVg&s=0ZyXhJLLLeGsN14Cl3JI-1_QU62MaC37ikInn-vP2yQ&e=
>>>>>
>>>>> [log in to unmask]
>>>>>
>>>>>
>>>>>
>>>>> ,
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> *From:* FSL - FMRIB's Software Library [mailto:[log in to unmask]] *On
>>>>> Behalf Of *Rosalia Dacosta Aguayo
>>>>> *Sent:* May-19-16 4:35 AM
>>>>> *To:* [log in to unmask]
>>>>> *Subject:* [FSL] No results with VBM
>>>>>
>>>>>
>>>>>
>>>>> Dear FSL experts,
>>>>>
>>>>> I have been working with 46 MRI images in order to find some
>>>>> correlations between on cognitive function and gray matter structures (deep
>>>>> gray matter structures).
>>>>>
>>>>> Pre-processing steps were as follow:
>>>>>
>>>>> 1. Cutting long necks
>>>>>
>>>>> 2. Noise correction
>>>>>
>>>>> 3. Bias field correction
>>>>>
>>>>> 4. Skull-stripping (using CAT12, (Tool for SPM12) because with bet I
>>>>> got that many images were not fine...as I had to decide wether improve this
>>>>> step manually...(methodology is not as strong and homogeneous as one could
>>>>> expect...or using another tool that gives good results..). With this Tool I
>>>>> got excellent results for all the images...and I after doing an homogeneity
>>>>> analysis, I discarded one of them...so n = 45
>>>>>
>>>>> 5. I used the skull stripped images in the first step of FSLVBM (that
>>>>> is: after doing first step, I reemplaced the skull stripped images created
>>>>> by the ones I got from CAT12).
>>>>>
>>>>> 6. Following fslvbm processes were fine, I visually inspect the GM
>>>>> template as well as the 4D_GM template and it was very good. Images were
>>>>> well aligned...
>>>>>
>>>>> The problem:
>>>>>
>>>>> I run correlation analysis between my template (with a gaussian kernel
>>>>> of 3 sigma) and my cognitive variable of interest regressing out for TIV,
>>>>> Age and Gender. It seems that my cog variable is not related to any
>>>>> structure...what I think it is not possible and I was thinking about if I
>>>>> missed something in the way or if my design.mat is not fine (I attach you
>>>>> the design files...)....
>>>>>
>>>>> Thank you a lot for your helping,
>>>>>
>>>>> Kind regards,
>>>>>
>>>>> Rosalia
>>>>>
>>>>
> ------------------------------
>
> Date: Sun, 22 May 2016 10:38:47 +0200
> From: Rosalia Dacosta Aguayo <[log in to unmask]>
> Subject: Re: No results with VBM
>
> Dear Anderson,
>
> Thank you for your reply.
>
> If I have understood you well, although my data has been demeaned, I have
> to use -D flag in order to demean my design?
>
> Kind regards,
>
> Rosalia
> El 22/05/2016 10:30, "Anderson M. Winkler" <[log in to unmask]>
> escribió:
>
>> Hi Rosalia,
>>
>> Please see below:
>>
>> On 21 May 2016 at 19:30, Rosalia Dacosta Aguayo <[log in to unmask]>
>> wrote:
>>
>>> Dear Anderson,
>>>
>>> The data has been demeaned, and I am working with just one group of 45
>>> subjects, I am not sure about including an intercept in this model. Just
>>> partial correlations between this group and a cog variable, regressing out
>>> for age, gender and TIV....
>>>
>> It's necessary to demean both design and data, or else, include an
>> intercept. Mean-centering the data only isn't sufficient.
>>
>>
>>> And about my question regarding FIRST? I do not understand why I can get
>>> .vtk and .bvars files for my 4. ROIS but when I give a look into firstseg
>>> files and I check for registeting....I only find two of the deep gm
>>> structures but not the 4 previosly selected.
>>>
>> This refers to a different thread. I'm not sure what is going on.
>>
>>
>>> What is more, I am not quite sure about surface and vertex or shape
>>> analysis and how to interpret results...
>>>
>> The shape analysis indicates the places where there is a significant shape
>> difference (i.e., a significant effect on the positions of the vertices
>> that delineate the structure).
>>
>> All the best,
>>
>> Anderson
>>
>>
>>> I would greatly appreciate your helping hear.
>>>
>>> Kind regards,
>>>
>>> Rosalia
>>> El 20/05/2016 11:44, "Rosalia Dacosta Aguayo" <[log in to unmask]>
>>> escribió:
>>>
>>>> Thank you Anderson!!
>>>>
>>>> Rosalia
>>>>
>>>> 2016-05-20 10:07 GMT+02:00 Anderson M. Winkler <[log in to unmask]>:
>>>>
>>>>> Hi Rosalia,
>>>>>
>>>>> Just adding to this: need to include an intercept in the model, or use
>>>>> the option -D.
>>>>>
>>>>> All the best,
>>>>>
>>>>> Anderson
>>>>>
>>>>>
>>>>> On 19 May 2016 at 16:10, Colin Hawco <[log in to unmask]> wrote:
>>>>>
>>>>>> While I am afraid I can offer no specific advice for VBM in FSL, I do
>>>>>> have a general piece of advice.
>>>>>>
>>>>>>
>>>>>>
>>>>>> When I see a null result I think is suspicious, I immediately drop
>>>>>> threshold and take a look. As instead, instead of p < 0.05 corrected, try
>>>>>> viewing the t-map without a threshold. . See if there looks like an
>>>>>> appropriate level of variability in the t-stats, or if it looks off (zero
>>>>>> at many data points, really speckled).
>>>>>>
>>>>>>
>>>>>>
>>>>>> This is a way to QC check the output and see if maybe something went
>>>>>> wrong which caused the stats to mess up.
>>>>>>
>>>>>>
>>>>>>
>>>>>> Good luck
>>>>>>
>>>>>>
>>>>>>
>>>>>> Colin Hawco, PhD
>>>>>>
>>>>>> Neuranalysis Consulting
>>>>>>
>>>>>> Neuroimaging analysis and consultation
>>>>>>
>>>>>> https://urldefense.proofpoint.com/v2/url?u=http-3A__www.neuranalysis.com&d=DQIFaQ&c=QmPtDiFixEjkMvDKaP3E2Vb9C2z4M0PdarxyAHQ2iDQ&r=xbyrBxm81l6mG_XEX66jgCNTXfK3eVo30T9sVQVTz1Q&m=Jc1PNVoYrsBQ4MCSXUR-BueqHdpSGdf2yPHWfNLLkVg&s=0ZyXhJLLLeGsN14Cl3JI-1_QU62MaC37ikInn-vP2yQ&e=
>>>>>>
>>>>>> [log in to unmask]
>>>>>>
>>>>>>
>>>>>>
>>>>>> ,
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>> *From:* FSL - FMRIB's Software Library [mailto:[log in to unmask]] *On
>>>>>> Behalf Of *Rosalia Dacosta Aguayo
>>>>>> *Sent:* May-19-16 4:35 AM
>>>>>> *To:* [log in to unmask]
>>>>>> *Subject:* [FSL] No results with VBM
>>>>>>
>>>>>>
>>>>>>
>>>>>> Dear FSL experts,
>>>>>>
>>>>>> I have been working with 46 MRI images in order to find some
>>>>>> correlations between on cognitive function and gray matter structures (deep
>>>>>> gray matter structures).
>>>>>>
>>>>>> Pre-processing steps were as follow:
>>>>>>
>>>>>> 1. Cutting long necks
>>>>>>
>>>>>> 2. Noise correction
>>>>>>
>>>>>> 3. Bias field correction
>>>>>>
>>>>>> 4. Skull-stripping (using CAT12, (Tool for SPM12) because with bet I
>>>>>> got that many images were not fine...as I had to decide wether improve this
>>>>>> step manually...(methodology is not as strong and homogeneous as one could
>>>>>> expect...or using another tool that gives good results..). With this Tool I
>>>>>> got excellent results for all the images...and I after doing an homogeneity
>>>>>> analysis, I discarded one of them...so n = 45
>>>>>>
>>>>>> 5. I used the skull stripped images in the first step of FSLVBM (that
>>>>>> is: after doing first step, I reemplaced the skull stripped images created
>>>>>> by the ones I got from CAT12).
>>>>>>
>>>>>> 6. Following fslvbm processes were fine, I visually inspect the GM
>>>>>> template as well as the 4D_GM template and it was very good. Images were
>>>>>> well aligned...
>>>>>>
>>>>>> The problem:
>>>>>>
>>>>>> I run correlation analysis between my template (with a gaussian kernel
>>>>>> of 3 sigma) and my cognitive variable of interest regressing out for TIV,
>>>>>> Age and Gender. It seems that my cog variable is not related to any
>>>>>> structure...what I think it is not possible and I was thinking about if I
>>>>>> missed something in the way or if my design.mat is not fine (I attach you
>>>>>> the design files...)....
>>>>>>
>>>>>> Thank you a lot for your helping,
>>>>>>
>>>>>> Kind regards,
>>>>>>
>>>>>> Rosalia
>>>>>>
>>>>>
> ------------------------------
>
> Date: Sun, 22 May 2016 12:02:50 +0200
> From: Francesca Zidda <[log in to unmask]>
> Subject: Re: No results with VBM
>
> Look at this!!! They talk about correlations, one group and t-stats!!!!!
>
> The mistery is gettino shape ;)
>
> :****
>
> Frà
>
> Inviato da iPhone
>
>> Il giorno 19 mag 2016, alle ore 17:10, Colin Hawco <[log in to unmask]> ha scritto:
>>
>> While I am afraid I can offer no specific advice for VBM in FSL, I do have a general piece of advice.
>>
>> When I see a null result I think is suspicious, I immediately drop threshold and take a look. As instead, instead of p < 0.05 corrected, try viewing the t-map without a threshold. . See if there looks like an appropriate level of variability in the t-stats, or if it looks off (zero at many data points, really speckled).
>>
>> This is a way to QC check the output and see if maybe something went wrong which caused the stats to mess up.
>>
>> Good luck
>>
>> Colin Hawco, PhD
>> Neuranalysis Consulting
>> Neuroimaging analysis and consultation
>> https://urldefense.proofpoint.com/v2/url?u=http-3A__www.neuranalysis.com&d=DQIFaQ&c=QmPtDiFixEjkMvDKaP3E2Vb9C2z4M0PdarxyAHQ2iDQ&r=xbyrBxm81l6mG_XEX66jgCNTXfK3eVo30T9sVQVTz1Q&m=Jc1PNVoYrsBQ4MCSXUR-BueqHdpSGdf2yPHWfNLLkVg&s=0ZyXhJLLLeGsN14Cl3JI-1_QU62MaC37ikInn-vP2yQ&e=
>> [log in to unmask]
>>
>> ,
>>
>>
>>
>> From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf Of Rosalia Dacosta Aguayo
>> Sent: May-19-16 4:35 AM
>> To: [log in to unmask]
>> Subject: [FSL] No results with VBM
>>
>> Dear FSL experts,
>>
>> I have been working with 46 MRI images in order to find some correlations between on cognitive function and gray matter structures (deep gray matter structures).
>>
>> Pre-processing steps were as follow:
>>
>> 1. Cutting long necks
>> 2. Noise correction
>> 3. Bias field correction
>> 4. Skull-stripping (using CAT12, (Tool for SPM12) because with bet I got that many images were not fine...as I had to decide wether improve this step manually...(methodology is not as strong and homogeneous as one could expect...or using another tool that gives good results..). With this Tool I got excellent results for all the images...and I after doing an homogeneity analysis, I discarded one of them...so n = 45
>> 5. I used the skull stripped images in the first step of FSLVBM (that is: after doing first step, I reemplaced the skull stripped images created by the ones I got from CAT12).
>> 6. Following fslvbm processes were fine, I visually inspect the GM template as well as the 4D_GM template and it was very good. Images were well aligned...
>>
>> The problem:
>>
>> I run correlation analysis between my template (with a gaussian kernel of 3 sigma) and my cognitive variable of interest regressing out for TIV, Age and Gender. It seems that my cog variable is not related to any structure...what I think it is not possible and I was thinking about if I missed something in the way or if my design.mat is not fine (I attach you the design files...)....
>>
>> Thank you a lot for your helping,
>>
>> Kind regards,
>>
>> Rosalia
> ------------------------------
>
> Date: Sun, 22 May 2016 07:56:30 -0400
> From: Mahmoud <[log in to unmask]>
> Subject: Re: zero filling
>
> Fslmaths -fillh
> On May 22, 2016 1:43 AM, "Moran Artzi" <[log in to unmask]> wrote:
>
>> Hi,
>> Is there any way to fill the holes in a binary image (as the brain_mask
>> img - the output of bet) using FSL command line?
>> Thanks
>> Moran
>>
> ------------------------------
>
> Date: Sun, 22 May 2016 12:48:44 +0000
> From: Saad Jbabdi <[log in to unmask]>
> Subject: Re: BEDPOSTX on data with replicated b-vectors in one dataset
>
> Hi
>
> With such low number of directions there should not be a significant difference between averaging and stacking wrt model selection performance. I would worry *much* more about having such low #directions in the first place, as Matt said.
>
> The noise issue is also a non-issue in most cases (e.g. if bvalues are not too high) - in practice if you see that bedpostx overfits due to noise floor, you can try the —f0 option which adds a noise floor parameter to the model.
>
> Cheers,
> Saad
>
>
>
>
>
> On 21 May 2016, at 21:54, Antonin Skoch <[log in to unmask]<mailto:[log in to unmask]>> wrote:
>
> Dear Matt, Michael,
>
> thank you for your contribution.
>
> Here is the relevant post from 3/2010 by Saad Jbabdi suggesting NOT to stack/concatenate the data for BEDPOSTX.
>
> https://urldefense.proofpoint.com/v2/url?u=https-3A__www.jiscmail.ac.uk_cgi-2Dbin_webadmin-3FA2-3Dind1003-26L-3DFSL-26D-3D0-261-3DFSL-269-3DA-26J-3Don-26K-3D2-26X-3DCBB60236BC562DF50E-26Y-3Dansk-2540ikem.cz-26d-3DNo-2BMatch-253BMatch-253BMatches-26z-3D4-26P-3D165079&d=DQIFaQ&c=QmPtDiFixEjkMvDKaP3E2Vb9C2z4M0PdarxyAHQ2iDQ&r=xbyrBxm81l6mG_XEX66jgCNTXfK3eVo30T9sVQVTz1Q&m=Jc1PNVoYrsBQ4MCSXUR-BueqHdpSGdf2yPHWfNLLkVg&s=xd8B6xO_DclrEoNOSsUYAkpJzyxrlYmIlZpfCGqp6ls&e=
>
> Conversely, averaging of the data effectively changes distribution from Rician to non-central chi as we discussed here:
>
> https://urldefense.proofpoint.com/v2/url?u=http-3A__community.mrtrix.org_t_anatomically-2Dconstrained-2Dtractography-2Dusing-2Dodf-2Dfrom-2Dbedpostx_189_6&d=DQIFaQ&c=QmPtDiFixEjkMvDKaP3E2Vb9C2z4M0PdarxyAHQ2iDQ&r=xbyrBxm81l6mG_XEX66jgCNTXfK3eVo30T9sVQVTz1Q&m=Jc1PNVoYrsBQ4MCSXUR-BueqHdpSGdf2yPHWfNLLkVg&s=3bQvyP3vRbeIrtwqktHhXJQSww693j0DoVyNIoHR0Oc&e=
>
> Would not this alteration of noise distribution have any spurious consequence on validity of Bayesian inference on diffusion orientations in BEDPOSTX?
>
> With recent protocols we acquire data with 64 unique directions, but we have quite large amount of data from previous studies acquired by 2x30 directions ("half-sphere" diffusion encoding). I would like to learn what method of processing is optimal for these data.
>
> Regards,
>
> Antonin Skoch
>
> ------------------------------
>
> Date: Sun, 22 May 2016 12:54:47 +0000
> From: Saad Jbabdi <[log in to unmask]>
> Subject: Re: probabilistic tractography
>
> Hi Irina
>
>> Dear experts,
>> I have a couple of questions regarding the results of probabilistic tractography:
>> 1. Too low intencity values in the fdt_paths.
>> After normalization of the paths (I divided the fdt_paths by (#voxels in the seed-mask * #permutations per voxel) and multiplied by 100 to get the resulting paths in %) I get the intencity values between 0.02 and 2% (when the paths are clearly seen). This is far too low, considering that the seed is practically within the corpus callosum. What are the usual values for the transcallosal paths?
> Those numbers don’t surprise me given that streamlines can travel in 3D (and therefore, e.g. under isotropic conditions, probabilities will vanish with 1/r^2). If you normalise by waytotal the numbers will be higher because you will be excluding sample streamlines that do not contribute to the fdt_paths distribution (having been rejected by the inclusion/exclusion criteria)
>
>
>> 2. FA vs. waytotal values between a seed and a target.
>> I compared the waytotal values (to waypoint mask, normalized) between the 2 groups (15 subjects in each). I found that in the group of patients both the #paths between a seed and a target (normalized waytotal values) and the FA values within the same tract in the skeleton were significantly lower, which was expected. But for another tract I've got significantly lower FA values in patients and higher #paths between a seed and a target (and also higher intencity values in the resulting path). Is this possible at all and how to explain this paradox?
> It’s not necessarily a paradox. The fdt_paths values depend on the uncertainty in fibre orientations, which relate to but are not the same as FA. For example, FA can drop considerably in regions of crossing fibres, but uncertainty in fitting the crossing orientation can still be low if you have sufficient sensitivity to crossings.
>
> Cheers,
> Saad
>
>
>> Thank you very much.
> ------------------------------
>
> Date: Sun, 22 May 2016 16:25:48 +0300
> From: Moran Artzi <[log in to unmask]>
> Subject: Re: zero filling
> ------------------------------>
> Thx!
>
> On Sun, May 22, 2016 at 2:56 PM, Mahmoud <[log in to unmask]> wrote:
>
>> Fslmaths -fillh
>> On May 22, 2016 1:43 AM, "Moran Artzi" <[log in to unmask]> wrote:
>>
>>> Hi,
>>> Is there any way to fill the holes in a binary image (as the brain_mask
>>> img - the output of bet) using FSL command line?
>>> Thanks
>>> Moran
>>>
>
> Date: Sun, 22 May 2016 14:58:23 +0000
> From: "Harms, Michael" <[log in to unmask]>
> Subject: Re: BEDPOSTX on data with replicated b-vectors in one dataset
>
> Perhaps Saad can elaborate on that earlier (2010) post, and whether it is
> still relevant.
>
> That’s the first I’ve heard that simply concatenating the data would be
> sub-optimal in terms of bedpostx performance (other than the run-time
> issue).
>
> cheers,
> -MH
>
> --
> Michael Harms, Ph.D.
>
> -----------------------------------------------------------
> Conte Center for the Neuroscience of Mental Disorders
> Washington University School of Medicine
> Department of Psychiatry, Box 8134
> 660 South Euclid Ave.Tel: 314-747-6173
> St. Louis, MO 63110Email: [log in to unmask]
>
>
>
>
> On 5/21/16, 3:54 PM, "FSL - FMRIB's Software Library on behalf of Antonin
> Skoch" <[log in to unmask] on behalf of [log in to unmask]> wrote:
>
> Dear Matt, Michael,
>
> thank you for your contribution.
>
> Here is the relevant post from 3/2010 by Saad Jbabdi suggesting NOT to
> stack/concatenate the data for BEDPOSTX.
>
> https://urldefense.proofpoint.com/v2/url?u=https-3A__www.jiscmail.ac.uk_cgi-2Dbin_webadmin-3FA2-3Dind1003-26L-3DFSL-26D-3D0-261-3DFSL-269-3DA-26&d=DQIFaQ&c=QmPtDiFixEjkMvDKaP3E2Vb9C2z4M0PdarxyAHQ2iDQ&r=xbyrBxm81l6mG_XEX66jgCNTXfK3eVo30T9sVQVTz1Q&m=Jc1PNVoYrsBQ4MCSXUR-BueqHdpSGdf2yPHWfNLLkVg&s=D8oku-jBRGd8I509CCSQqJcNpu_kzeB5-5VXYCQkjYA&e=
> J=on&K=2&X=CBB60236BC562DF50E&Y=ansk%40ikem.cz&d=No+Match%3BMatch%3BMatches
> &z=4&P=165079
>
> Conversely, averaging of the data effectively changes distribution from
> Rician to non-central chi as we discussed here:
>
> https://urldefense.proofpoint.com/v2/url?u=http-3A__community.mrtrix.org_t_anatomically-2Dconstrained-2Dtractography-2Dusing-2Do&d=DQIFaQ&c=QmPtDiFixEjkMvDKaP3E2Vb9C2z4M0PdarxyAHQ2iDQ&r=xbyrBxm81l6mG_XEX66jgCNTXfK3eVo30T9sVQVTz1Q&m=Jc1PNVoYrsBQ4MCSXUR-BueqHdpSGdf2yPHWfNLLkVg&s=MlmUQ7zYGqxZaCYoCWEWchirBBXwZtz0hAc-g7CHpKA&e=
> df-from-bedpostx/189/6
>
> Would not this alteration of noise distribution have any spurious
> consequence on validity of Bayesian inference on diffusion orientations in
> BEDPOSTX?
>
> With recent protocols we acquire data with 64 unique directions, but we
> have quite large amount of data from previous studies acquired by 2x30
> directions ("half-sphere" diffusion encoding). I would like to learn what
> method of processing is optimal for these data.
>
> Regards,
>
> Antonin Skoch
>
>
>
> ________________________________
> The materials in this message are private and may contain Protected Healthcare Information or other information of a sensitive nature. If you are not the intended recipient, be advised that any unauthorized use, disclosure, copying or the taking of any action in reliance on the contents of this information is strictly prohibited. If you have received this email in error, please immediately notify the sender via telephone or return mail.
>
> ------------------------------
>
> Date: Sun, 22 May 2016 16:19:17 +0000
> From: Saad Jbabdi <[log in to unmask]>
> Subject: Re: BEDPOSTX on data with replicated b-vectors in one dataset
>
> Hi Michael,
>
> I haven’t done any theory on that, my comment was based on both intuition and my own experience. The ARD prior competes with the data likelihood, which gets bigger when you have more data points.
> Note that when the likelihood is Gaussian, it should not make a difference whether to stack or average: the effect of stacking (linear with n) is balanced by the effect of averaging (sigma^2 linear with n). But in bedpostx the sigma^2 is not estimated but integrated out.
>
> Cheers
> Saad
>
>
>
>
>
> On 22 May 2016, at 15:58, Harms, Michael <[log in to unmask]<mailto:[log in to unmask]>> wrote:
>
> Perhaps Saad can elaborate on that earlier (2010) post, and whether it is
> still relevant.
>
> That’s the first I’ve heard that simply concatenating the data would be
> sub-optimal in terms of bedpostx performance (other than the run-time
> issue).
>
> cheers,
> -MH
>
> --
> Michael Harms, Ph.D.
>
> -----------------------------------------------------------
> Conte Center for the Neuroscience of Mental Disorders
> Washington University School of Medicine
> Department of Psychiatry, Box 8134
> 660 South Euclid Ave.Tel: 314-747-6173
> St. Louis, MO 63110Email: [log in to unmask]<mailto:[log in to unmask]>
>
>
>
>
> On 5/21/16, 3:54 PM, "FSL - FMRIB's Software Library on behalf of Antonin
> Skoch" <[log in to unmask]<mailto:[log in to unmask]> on behalf of [log in to unmask]<mailto:[log in to unmask]>> wrote:
>
> Dear Matt, Michael,
>
> thank you for your contribution.
>
> Here is the relevant post from 3/2010 by Saad Jbabdi suggesting NOT to
> stack/concatenate the data for BEDPOSTX.
>
> https://urldefense.proofpoint.com/v2/url?u=https-3A__www.jiscmail.ac.uk_cgi-2Dbin_webadmin-3FA2-3Dind1003-26L-3DFSL-26D-3D0-261-3DFSL-269-3DA-26&d=DQIFaQ&c=QmPtDiFixEjkMvDKaP3E2Vb9C2z4M0PdarxyAHQ2iDQ&r=xbyrBxm81l6mG_XEX66jgCNTXfK3eVo30T9sVQVTz1Q&m=Jc1PNVoYrsBQ4MCSXUR-BueqHdpSGdf2yPHWfNLLkVg&s=D8oku-jBRGd8I509CCSQqJcNpu_kzeB5-5VXYCQkjYA&e=
> J=on&K=2&X=CBB60236BC562DF50E&Y=ansk%40ikem.cz&d=No+Match%3BMatch%3BMatches
> &z=4&P=165079
>
> Conversely, averaging of the data effectively changes distribution from
> Rician to non-central chi as we discussed here:
>
> https://urldefense.proofpoint.com/v2/url?u=http-3A__community.mrtrix.org_t_anatomically-2Dconstrained-2Dtractography-2Dusing-2Do&d=DQIFaQ&c=QmPtDiFixEjkMvDKaP3E2Vb9C2z4M0PdarxyAHQ2iDQ&r=xbyrBxm81l6mG_XEX66jgCNTXfK3eVo30T9sVQVTz1Q&m=Jc1PNVoYrsBQ4MCSXUR-BueqHdpSGdf2yPHWfNLLkVg&s=MlmUQ7zYGqxZaCYoCWEWchirBBXwZtz0hAc-g7CHpKA&e=
> df-from-bedpostx/189/6
>
> Would not this alteration of noise distribution have any spurious
> consequence on validity of Bayesian inference on diffusion orientations in
> BEDPOSTX?
>
> With recent protocols we acquire data with 64 unique directions, but we
> have quite large amount of data from previous studies acquired by 2x30
> directions ("half-sphere" diffusion encoding). I would like to learn what
> method of processing is optimal for these data.
>
> Regards,
>
> Antonin Skoch
>
>
>
> ________________________________
> The materials in this message are private and may contain Protected Healthcare Information or other information of a sensitive nature. If you are not the intended recipient, be advised that any unauthorized use, disclosure, copying or the taking of any action in reliance on the contents of this information is strictly prohibited. If you have received this email in error, please immediately notify the sender via telephone or return mail.
>
> ------------------------------
>
> End of FSL Digest - 21 May 2016 to 22 May 2016 (#2016-145)
> **********************************************************
Houston Methodist. Leading Medicine.
Ranked by U.S.News & World Report as one of America's "Best Hospitals" in 11 specialties. Named to FORTUNE® Magazine's "100 Best Companies to Work For®" list 10 years in a row. Designated as a Magnet hospital for excellence in nursing. Visit us at houstonmethodist.org. Follow us at twitter.com/MethodistHosp and www.facebook.com/HoustonMethodist.
***CONFIDENTIALITY NOTICE*** This e-mail is the property of Houston Methodist Hospital and/or its relevant affiliates and may contain restricted and privileged material for the sole use of the intended recipient(s). Any review, use, distribution or disclosure by others is strictly prohibited. If you are not the intended recipient (or authorized to receive for the recipient), please contact the sender and delete all copies of the message. Thank you.