Thank you Andy, so much, for these extremely clear and helpful tutorials! I would like to ask- how can I derive an atlas-based anatomically defined ROI, but only for part of an atlas region? Specifically, I want to derive a ROI only of the anterior part of the anterior cingulate cortex. Can I do that?
This tutorial was aimed at doing ROI analysis through the featquery GUI, which I assume is what most people use. I may include some additional information later about how to do it from the command line; thanks for pointing it out. -Andy
Hi Andrew, thank you for your helpful videos! I'm trying to extract data from the BA9 region using SPM and the WFUpickatlas but I'm finding problems with the dimension of the my MRI data and BA9. Both MRI scan and BA9 mask are in MNI space but the MRI scan has a dimension of 256x256x256 while the BA9 mask has a dimension of 91x109x91. I'm not sure if this is trivial but how would I create the BA9 mask with the same dimensions as my MRI scan?
Hi Pei, You will need to resample the data you are extracting from so that it has the same dimensions as the mask you are using as the ROI; for details about how to do it, see this link: andysbrainbook.readthedocs.io/en/latest/FrequentlyAskedQuestions/FrequentlyAskedQuestions.html Make sure that you are using your preprocessed functional images for data extraction, not the anatomical image. It may be that if you use the preprocessed functional data (which includes warping), there might not be a need to do further resampling of the data. Best, -Andy
Hi Andrew, I am trying to use randomise for my data I thought it could help as my N is so small, 20 or more! but I can not get an activity cluster! you see, it seems that all my data are aligned on each other! I created a design matrix according to my control and patient subjects, ran the randomise command on filtered_func_data that was registered to MNI previously for all subjects! I can't find out what the problem could be! would you please help me? BR, Mehrdad
thanks for your soon reply Andy, actually I have got two different groups, say healthy and patient. but because I have just acquired not much data, 15 in total, 8 healthy and 7 patient, the result is not satisfying! Do you have any recommendations so I can get the best possible result? I mean which test do you think is better to use? for group analysis, again because of the number of data I have, I used fixed effect analysis, how can I compensate for this lack of data to get reliable results? BR, Mehrdad
Hi! Thanks for your sharing. I am wondering if there is a good way to draw a ROI mask according to the melodic IC maps resulted from a group-level ICA? For example, the ACC region in the DMN.
You could highlight the image in fslview by clicking on its name in the lower right window, and then saving it as an image. Then use fslmaths to threshold (at whichever t-statistic yo used for correction) and binarize it. -Andy
thanks for your sharing! i have a question about functional connectivity! i do not know how to do rsFC based on feed ROI using FSL(not ICA). i heard it could make out through some way similiar to PPI analysis, but i dont know how to deal with it? thanks if u would like share to me!!
Hey there, You can find some steps here: www.ncbi.nlm.nih.gov/pmc/articles/PMC4448548/ My understanding is that you use the res4d.nii.gz residuals after modeling out nuisance regressors such as motion and white matter signal. You then normalize the residuals using these commands: fslmaths res4d.nii.gz -Tmean resMean.nii.gz fslmaths res4d.nii.gz -Tstd resSD.nii.gz fslmaths res4d.nii.gz -sub resMean.nii.gz -div resSD.nii.gz -add 100 res4d_normalized.nii.gz These normalized residuals can then be entered as a regressor into the GLM (as is, not convolved with any basis function) and then the parameter estimates submitted to a second-level analysis. I prefer to work with correlations that are converted to z-scores, but I don't know how to do that in FSL. Good luck! -Andy
There is also a mysterious command called "fsl_sbca" which I've heard can generate correlation maps, but it's unsupported. Many of the questions people ask about it on the message boards go unanswered, so I have no idea whether the FSL developers recommend using it or not. -Andy
Thank you for your great videos. They are absolutely helpful. How can I compare activity in my ROI between to group, using Featquery report? Should I use paired t-test?
Hi Mehrdad, It depends on your experimental design. If you have two different groups, then use a 2-sample t-test; if the data from different conditions come from the same subject, then use a paired-samples t-test. Once you've extracted the data from each ROI, you can test it however you want. In either case, I would recommend extracting the data for each condition individually, instead of only doing a paired-sample (or 2-sample) t-test and reporting the ROI result. That way you can see what is driving the effect. Best, -Andy
thanks for your soon reply Andy, actually I have got two different groups, say healthy and patient. but because I have just acquired not much data, 15 in total, 8 healthy and 7 patient, the result is not satisfying! Do you have any recommendations so I can get the best possible result? I mean which test do you think is better to use? for group analysis, again because of the number of data I have, I used fixed effect analysis, how can I compensate for this lack of data to get reliable results? BR, Mehrdad
Hey Mehrdad, Unfortunately, that is a small N for a between-groups analysis; it's not surprising that you're getting a null result. Even though fixed-effect analysis may boost your power, it also makes it impossible to generalize your results to the population. The only solutions that I can see are: 1) Increase your sample size; or 2) Restrict your analysis to an ROI or small volume (although it already appears that you've done that). -Andy
Hi Andy! Your videos are fantastic and have helped me a great deal! I just wanted to ask if you could offer some advice on my problem: Currently, I have a FLAIR image (nifti format) with a region of abnormal hyper-intensity on one side of the brain. Using MRICron, I manually drew and saved a volume of interest (VOI) as a nifti file. Then, I created another VOI in native space in the contralateral side (using flirt, fnirt, and applywarp). Do you know a way to extract the average intensity of each VOI I made, from the FLAIR image? My current guess is to use something like MATLAB/SPM, but I'm not sure. Thanks!
So helpful! You're a lifesaver!
Thank you! This is very helpful!
Thank you Andy, so much, for these extremely clear and helpful tutorials!
I would like to ask- how can I derive an atlas-based anatomically defined ROI, but only for part of an atlas region? Specifically, I want to derive a ROI only of the anterior part of the anterior cingulate cortex.
Can I do that?
i have another question. why don't u use the command line "fslmaths to extract the ROI and fslmeants to calculate the mean of ROI "?
This tutorial was aimed at doing ROI analysis through the featquery GUI, which I assume is what most people use. I may include some additional information later about how to do it from the command line; thanks for pointing it out.
-Andy
Hi Andrew, thank you for your helpful videos! I'm trying to extract data from the BA9 region using SPM and the WFUpickatlas but I'm finding problems with the dimension of the my MRI data and BA9. Both MRI scan and BA9 mask are in MNI space but the MRI scan has a dimension of 256x256x256 while the BA9 mask has a dimension of 91x109x91. I'm not sure if this is trivial but how would I create the BA9 mask with the same dimensions as my MRI scan?
Hi Pei,
You will need to resample the data you are extracting from so that it has the same dimensions as the mask you are using as the ROI; for details about how to do it, see this link: andysbrainbook.readthedocs.io/en/latest/FrequentlyAskedQuestions/FrequentlyAskedQuestions.html
Make sure that you are using your preprocessed functional images for data extraction, not the anatomical image. It may be that if you use the preprocessed functional data (which includes warping), there might not be a need to do further resampling of the data.
Best,
-Andy
Hi Andrew,
I am trying to use randomise for my data I thought it could help as my N is so small, 20 or more!
but I can not get an activity cluster! you see, it seems that all my data are aligned on each other!
I created a design matrix according to my control and patient subjects, ran the randomise command on filtered_func_data that was registered to MNI previously for all subjects!
I can't find out what the problem could be! would you please help me?
BR,
Mehrdad
thanks for your soon reply Andy,
actually I have got two different groups, say healthy and patient.
but because I have just acquired not much data, 15 in total, 8 healthy and 7 patient,
the result is not satisfying!
Do you have any recommendations so I can get the best possible result?
I mean which test do you think is better to use?
for group analysis, again because of the number of data I have, I used fixed effect analysis, how can I compensate for this lack of data to get reliable results?
BR,
Mehrdad
Hi! Thanks for your sharing. I am wondering if there is a good way to draw a ROI mask according to the melodic IC maps resulted from a group-level ICA? For example, the ACC region in the DMN.
You could highlight the image in fslview by clicking on its name in the lower right window, and then saving it as an image. Then use fslmaths to threshold (at whichever t-statistic yo used for correction) and binarize it.
-Andy
thanks for your sharing!
i have a question about functional connectivity! i do not know how to do rsFC based on feed ROI using FSL(not ICA).
i heard it could make out through some way similiar to PPI analysis, but i dont know how to deal with it?
thanks if u would like share to me!!
Hey there,
You can find some steps here: www.ncbi.nlm.nih.gov/pmc/articles/PMC4448548/
My understanding is that you use the res4d.nii.gz residuals after modeling out nuisance regressors such as motion and white matter signal. You then normalize the residuals using these commands:
fslmaths res4d.nii.gz -Tmean resMean.nii.gz
fslmaths res4d.nii.gz -Tstd resSD.nii.gz
fslmaths res4d.nii.gz -sub resMean.nii.gz -div resSD.nii.gz -add 100 res4d_normalized.nii.gz
These normalized residuals can then be entered as a regressor into the GLM (as is, not convolved with any basis function) and then the parameter estimates submitted to a second-level analysis. I prefer to work with correlations that are converted to z-scores, but I don't know how to do that in FSL.
Good luck!
-Andy
There is also a mysterious command called "fsl_sbca" which I've heard can generate correlation maps, but it's unsupported. Many of the questions people ask about it on the message boards go unanswered, so I have no idea whether the FSL developers recommend using it or not.
-Andy
Thanks, it is very helpful! i have figured out how to do seed-based FC analysis through the article u provided !
Thanks Andrew!
Thank you for your great videos. They are absolutely helpful.
How can I compare activity in my ROI between to group, using Featquery report?
Should I use paired t-test?
Hi Mehrdad,
It depends on your experimental design. If you have two different groups, then use a 2-sample t-test; if the data from different conditions come from the same subject, then use a paired-samples t-test. Once you've extracted the data from each ROI, you can test it however you want.
In either case, I would recommend extracting the data for each condition individually, instead of only doing a paired-sample (or 2-sample) t-test and reporting the ROI result. That way you can see what is driving the effect.
Best,
-Andy
thanks for your soon reply Andy,
actually I have got two different groups, say healthy and patient.
but because I have just acquired not much data, 15 in total, 8 healthy and 7 patient,
the result is not satisfying!
Do you have any recommendations so I can get the best possible result?
I mean which test do you think is better to use?
for group analysis, again because of the number of data I have, I used fixed effect analysis, how can I compensate for this lack of data to get reliable results?
BR,
Mehrdad
Hey Mehrdad,
Unfortunately, that is a small N for a between-groups analysis; it's not surprising that you're getting a null result. Even though fixed-effect analysis may boost your power, it also makes it impossible to generalize your results to the population.
The only solutions that I can see are:
1) Increase your sample size; or
2) Restrict your analysis to an ROI or small volume (although it already appears that you've done that).
-Andy
Hi Andrew,
Thanks, yup I look for activity in an ROI using featquery.
Thank you very much again for your help.
BR,
Mehrdad
Hi Andy!
Your videos are fantastic and have helped me a great deal!
I just wanted to ask if you could offer some advice on my problem:
Currently, I have a FLAIR image (nifti format) with a region of abnormal hyper-intensity on one side of the brain. Using MRICron, I manually drew and saved a volume of interest (VOI) as a nifti file. Then, I created another VOI in native space in the contralateral side (using flirt, fnirt, and applywarp). Do you know a way to extract the average intensity of each VOI I made, from the FLAIR image? My current guess is to use something like MATLAB/SPM, but I'm not sure.
Thanks!