I'm french radiologist resident in France, doing a study using DTI on patient for my work.
I would like asking some help for the final part on FSL using TBSS.
Let me explain my work :
I have a cohort of 24 patients, and i performed 2 IRM on each patient, the first <1 week after a mild brain traumatism, and the second after 1 month.
I would like to evaluate if there is FA variation and MD variation, on the group, and between the 2 irm for each patient.
Things are that i don't have a group control unfortunately.
So I worked in the first time on a software called Olea, using ROI method, but i found anarchic results, not very pertinent.
So I'm trying running TBSS. I already did the pre-processing on my data : denoising, degibbsing, and remove eddy current.
Now for the final step running stats, i feel lost.
I would like to know what is the best way to have my result :
- Looking for significant diffrences on FA and MD on all of the First IRM for the 24 patients.
- Looking for significant variation from IRM 1 to IRM 2 for every patient.
Do i need to create 2 groups "IRM1" and "IRM2" and run TBSS on every groups ?
Thank you,
Regards
Ali
Hi Ali, There are many ways to achieve this. You can compare the patients and the healthy controls at your baseline (I assume you have the controls cuz you said 'variations'?), and then see how the regions with significant between-group differences changes longitudinally. Or you can just look at the changes of whole brain between baseline and follow-up. What's important is, do make sure you register the image at IRM2 to IRM1 (see the FSLwiki "FLIRT" / "FNIRT" for more information). What I usually do is: 1. Compare the between-group differences at IRM1 with TBSS. (for now don't run the IRM2 data) 2. Extract the regions with significant between-group differences. (Usually done with 'cluster'.) For eg. I'll call it 'PT_low_c95.nii.gz'. 3. Extract the ROI masks from'PT_low_c95.nii.gz' based on the standard JHU atlas. This could be done by fslmaths. Now you'll get some specific regions contained in the PT_low_c95.nii.gz like Superior_longitudinal_fasciculus_R_mask.nii.gz, Anterior_corona_radiata_L_mask.nii.gz etc 4. Deproject those ROI masks into native space by tbss_deproject. This will generate individual masks for each subject and each ROI: sub001_Anterior_corona_radiata_L_mask.nii.gz sub002_Anterior_corona_radiata_L_mask.nii.gz sub001_Superior_longitudinal_fasciculus_R_mask.nii.gz sub002_Superior_longitudinal_fasciculus_R_mask.nii.gz ... 5. Extract IRM1 FA values from those individual masks by fslmeants 6. Register the FA images at IRM2 to IRM1 by flirt and create new IRM2 FA images. 7. Run tbss_1_preporc on the new IRM2 images. 8. Extract IRM2 FA values by the baseline individual masks. Then it will be all set to test the statistical differences between IRM1 and IRM2. Best, Wenjun
Thanks for your response.
Unfortunately, i don't have a group control, that is the main limit on my work.
Can you tell me how to do the 2 and 3rd step ? Or do you know where i can get explanations ?
This is my first contact with FSL, all i see for TBSS is based on the TBSS page.
Best regards,
Originally posted by Wenjun Su:
There are many ways to achieve this. You can compare the patients and the healthy controls at your baseline (I assume you have the controls cuz you said 'variations'?), and then see how the regions with significant between-group differences changes longitudinally. Or you can just look at the changes of whole brain between baseline and follow-up. What's important is, do make sure you register the image at IRM2 to IRM1 (see the FSLwiki "FLIRT" / "FNIRT" for more information).
What I usually do is:
1. Compare the between-group differences at IRM1 with TBSS. (for now don't run the IRM2 data)
2. Extract the regions with significant between-group differences. (Usually done with 'cluster'.) For eg. I'll call it 'PT_low_c95.nii.gz'.
3. Extract the ROI masks from'PT_low_c95.nii.gz' based on the standard JHU atlas. This could be done by fslmaths. Now you'll get some specific regions contained in the PT_low_c95.nii.gz like Superior_longitudinal_fasciculus_R_mask.nii.gz, Anterior_corona_radiata_L_mask.nii.gz etc
4. Deproject those ROI masks into native space by tbss_deproject. This will generate individual masks for each subject and each ROI: sub001_Anterior_corona_radiata_L_mask.nii.gz
sub002_Anterior_corona_radiata_L_mask.nii.gz
sub001_Superior_longitudinal_fasciculus_R_mask.nii.gz
sub002_Superior_longitudinal_fasciculus_R_mask.nii.gz
...
5. Extract IRM1 FA values from those individual masks by fslmeants
6. Register the FA images at IRM2 to IRM1 by flirt and create new IRM2 FA images.
7. Run tbss_1_preporc on the new IRM2 images.
8. Extract IRM2 FA values by the baseline individual masks.
Then it will be all set to test the statistical differences between IRM1 and IRM2.
Best,
Wenjun
Hi Ali, If you don't have a control group then I think you can run just by setting the two-sample paired t test. But you still need to register the IRM2 images to IRM1 images before you run the data. Setting the contrasts: https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/G... https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/F... Etracting the results with significant differences: https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/C... Extract the ROI masks from the regions that showed significant differences: https://fsl.fmrib.ox.ac.uk/fslcourse/lec... You can also check this paper for reference as they articulate the stats throughly in their methods: Krakauer K, Nordentoft M, Glenthøj BY, Raghava JM, Nordholm D, Randers L, Glenthøj LB, Ebdrup BH, Rostrup E. White matter maturation during 12 months in individuals at ultra-high-risk for psychosis. Acta Psychiatr Scand. 2018 Jan;137(1):65-78. Hope it helps! Originally posted by Ali EL AMEEN:
Krakauer K, Nordentoft M, Glenthøj BY, Raghava JM, Nordholm D, Randers L, Glenthøj LB, Ebdrup BH, Rostrup E. White matter maturation during 12 months in individuals at ultra-high-risk for psychosis. Acta Psychiatr Scand. 2018 Jan;137(1):65-78.