Dear Dr. Manjon and Dr. Coupe,

I read your paper on volBrain pipeline as well as the user area and instructions. I do still have some questions and I will appreciate if you can help with that.

We used volBrain to calculate WM, GM (cortex) and CSF (ventricles and sulci separately). I downloaded nifti files and I use ITK-Snap to do the manual corrections. We have brain MRI of 33 patients, and we aim to have a consistent measurement among patients, not necessarily the true-exact volumes (we will compare the brain atrophy).


1 - ITK-Snap works with 16-bit data and gives the warning below (attached file 1). I am still able to delineate and manually correct the segmentation. Does that still affect the volumetric calculations? Should we use another software instead of ITK-Snap?

2 - On the CSV file, we do not have the explanation for the abbreviations; what do they correspond to and how they are calculated. For instance, Scale factor, SNR, mSNR and QC; and how tissue percentage is calculated, e.g. is WM% = WM / WM+GM+CSF or WM/Intracranial volume (intracranial volume will include vessels, dura and other non-brain structures as well). Is IC, intracranial volume corresponds to intradural space? We will be glad if you could provide an explanation about that.

3 - How asymmetry is calculated?

4 - One of our patient MRI could not be segmented ("Finished with an error", 146036) and another one is segmented sub-optimally ("Registration is suboptimal", 146032). Since our sample size is not large, we do not want to exclude any patients. Do you have a suggestion for that? Should I do completely manual segmentation on ITK-Snap for these two patients?

5 - On crisp segmentation file, cortex GM and cerebellum GM have the same label. Therefore, while doing the manual correction, I only change cortex so that I can subtract/add the difference to the cortex GM volume on the CSV file. However, I am not sure if the dura between the cerebrum and cerebellum (tentorium cerebelli) is included to cerebellum or cerebrum on the volBrain results file. Is it included to cortex GM or cerebellum?


Thank you in advance.

Best regards,
Merve

Dear Merve

Thank you very much for your response. I run the pipeline for the sub-optimally registered case, and I used n_mni_fjob file as the input per your suggestion. It worked well.

Good news.

I would like to briefly describe how I do the manual corrections and I will appreciate if you can tell if you find that approach useful, or have any better ideas. I will also some questions regarding three cases within our dataset. The questions below will constitute the backbone of the discussion part of our project, and since you are the expert on this subject, I wanted you ask for your opinion.

Manual Correction

Our aim is to understand how shrinked the brain is within the skull from cross-sectional MRI image. Therefore, extra-ventricular CSF volume (indicating how sulci are distanced away from each other), ventricular sizes, and their proportions to cortex GM, cerebral WM are significant.

On crisp_mni_fjob file, you see some parts of the dura and sagittal sinus are labeled as GM (green) or CSF (red) (see Image Image). I delete such voxels in all orthogonal planes. Then, I will check volumes and statistics on ITK-SNAP to see how each volume is changed. Since, I am only deleting cerebral cortical GM and sulci CSF, I will subtract/add this difference to the volume we have on the volBrain report. I think using manually corrected volumetric result may give a better correlation, if there is. What do you think about this approach? Is there a path you follow while doing manual corrections on volBrain files with ITK-SNAP?

Sounds correct. To obtain final volume you have to divide volume in Mni space with the scale factor provided in the report.

Here, this question also emerges: are volumetric results on the final volBrain report calculated from native crisp or mni registered crisp file? Since I am doing the manual corrections on n_mmni crisp file, if the final volumes on the report are calculated from native space, I need to apply inverse transformation to the manually corrected mni crisp file and then add/subtract the volume difference to results on the volBrain report (or I need to re-do all of the manual corrections on native_crisp files).

It is better to use scale factor directly on the mni volume to avoid interpolation artifact. However you can to both to check that both give you similar results.

I do not delete the voxels of tentorium cerebelli, since as we talked, I do not know whether that part is included to the cerebrum or cerebellum. If it is included to the cerebellum, I should not subtract it from the GM volume result on volBrain report. But, I am guessing it is included into the

You can check that using the cerebellum segmentation provided in the package.

Questions about 4 Cases

Patient 1 (job146032, Report Date June 10) (Re-do: job151027, Report Date July 11)
This case was the one that is sub-optimally registered. Using ITK-SNAP, you can appreciate that the segmentation is affected. I used n_mni_fjob as input file and it gave a much better segmentation result and different volumetric results (job151027, Report Date July 11) (see Image).

I do not have access to job. I cannot check.

Patient 2 (job145969, Report Date June 9) (Re-do: job151308, Report Date July 12, Input: n_mmni_fjob)
This case was registered successfully. I wanted to re-do it with n_mmni_fjob input to see the effect of using normalized, filtered image as the input in an optimally registered case.

In volBrain results file, the volumes are pretty different, thought the percentages are almost the same. Is it because the volumetric calculation is done on the native image? (see Image) Since in your paper, the pipeline includes MNI registration before the tissue segmentation, I am guessing that the segmentation and volumetric calculations are done on MNI space, then with inverse transformation, calculated for native space and the final result on volBrain report is for the native space. Is that correct?

% are normalized volume by ICC. We use these volumes since there are more stable that absolute ones. As explainypu have to divide volume in mni by scale factor ( structure + ICC) then divide structure_native by ICC_native to get final normalized volume.

Patient 3 (job145972, Report Date June 9) (Re-do: job151387, Report Date July 12, Input: n_mmni_fjob)
Though we had no error during the processing, the resulting segmentation file of this case is problematic. As you can see on Image , left frontal lobe is missing and the region supposed to be labeled as WM is labeled as CSF. It resulted with the similar problem even after re-running the pipeline with the input n_mmni. The original image is Image . What is your suggestion for this case? Should we do it manually from scratch?

I guess Manuel seg is necessary. Sorry for that.

Patient 4 (job146036, Upload Date June 10) (Re-do: job151028, Upload Date July 11)
This case could not be segmented at all, and there is no file nor report available. You can see the error on Image. Is there an explanation why this case could not be processed and the previous one had a problem? Should we do this case manually as well?

I do not know.

Best

Pierrick

Dear Dr. Coupe,

Thank you very much for your response. I run the pipeline for the sub-optimally registered case, and I used n_mni_fjob file as the input per your suggestion. It worked well.

I would like to briefly describe how I do the manual corrections and I will appreciate if you can tell if you find that approach useful, or have any better ideas. I will also some questions regarding three cases within our dataset. The questions below will constitute the backbone of the discussion part of our project, and since you are the expert on this subject, I wanted you ask for your opinion.

Manual Correction

Our aim is to understand how shrinked the brain is within the skull from cross-sectional MRI image. Therefore, extra-ventricular CSF volume (indicating how sulci are distanced away from each other), ventricular sizes, and their proportions to cortex GM, cerebral WM are significant.

On crisp_mni_fjob file, you see some parts of the dura and sagittal sinus are labeled as GM (green) or CSF (red) (see  ). I delete such voxels in all orthogonal planes. Then, I will check volumes and statistics on ITK-SNAP to see how each volume is changed. Since, I am only deleting cerebral cortical GM and sulci CSF, I will subtract/add this difference to the volume we have on the volBrain report. I think using manually corrected volumetric result may give a better correlation, if there is. What do you think about this approach? Is there a path you follow while doing manual corrections on volBrain files with ITK-SNAP?

Here, this question also emerges: are volumetric results on the final volBrain report calculated from native crisp or mni registered crisp file? Since I am doing the manual corrections on n_mmni crisp file, if the final volumes on the report are calculated from native space, I need to apply inverse transformation to the manually corrected mni crisp file and then add/subtract the volume difference to results on the volBrain report (or I need to re-do all of the manual corrections on native_crisp files).

I do not delete the voxels of tentorium cerebelli, since as we talked, I do not know whether that part is included to the cerebrum or cerebellum. If it is included to the cerebellum, I should not subtract it from the GM volume result on volBrain report. But, I am guessing it is included into the cerebrum.

Questions about 4 Cases

Patient 1 (job146032, Report Date June 10) (Re-do: job151027, Report Date July 11)
This case was the one that is sub-optimally registered. Using ITK-SNAP, you can appreciate that the segmentation is affected. I used n_mni_fjob as input file and it gave a much better segmentation result and different volumetric results (job151027, Report Date July 11) (see ).

Patient 2 (job145969, Report Date June 9) (Re-do: job151308, Report Date July 12, Input: n_mmni_fjob)
This case was registered successfully. I wanted to re-do it with n_mmni_fjob input to see the effect of using normalized, filtered image as the input in an optimally registered case.

In volBrain results file, the volumes are pretty different, thought the percentages are almost the same. Is it because the volumetric calculation is done on the native image? (see ) Since in your paper, the pipeline includes MNI registration before the tissue segmentation, I am guessing that the segmentation and volumetric calculations are done on MNI space, then with inverse transformation, calculated for native space and the final result on volBrain report is for the native space. Is that correct?

Patient 3 (job145972, Report Date June 9) (Re-do: job151387, Report Date July 12, Input: n_mmni_fjob)
Though we had no error during the processing, the resulting segmentation file of this case is problematic. As you can see on  , left frontal lobe is missing and the region supposed to be labeled as WM is labeled as CSF. It resulted with the similar problem even after re-running the pipeline with the input n_mmni. The original image is  . What is your suggestion for this case? Should we do it manually from scratch?

Patient 4 (job146036, Upload Date June 10) (Re-do: job151028, Upload Date July 11)
This case could not be segmented at all, and there is no file nor report available. You can see the error on . Is there an explanation why this case could not be processed and the previous one had a problem? Should we do this case manually as well?

I appreciate any contribution you may have. Thank you very much for allocation your time.

Best regards,
Merve

Dear Merve

1 - ITK-Snap works with 16-bit data and gives the warning below (attached file 1). I am still able to delineate and manually correct the segmentation. Does that still affect the volumetric calculations? Should we use another software instead of ITK-Snap?

No it is ok. We are using 64bit image but the conversion to 32 is not an issue. Itksnap is an excellent choice.

2 - On the CSV file, we do not have the explanation for the abbreviations; what do they correspond to and how they are calculated. For instance, Scale factor, SNR, mSNR and QC; and how tissue percentage is calculated, e.g. is WM% = WM / WM+GM+CSF or WM/Intracranial volume (intracranial volume will include vessels, dura and other non-brain structures as well). Is IC, intracranial volume corresponds to intradural space? We will be glad if you could
provide an explanation about that.

Scale factor is the obtained during registration between native space and MNI space.
SNR is the signal noise ration. You can ignore this part.
QC is the correlation between your image and the MNI template, it just give an estimation of the registration quality.
WM% is the WM/ICV (see NICE paper for the definition: https://hal.archives-ouvertes.fr/hal-01060348/en/ )


3 - How asymmetry is calculated?

As explained in the pdf repot:
The Asymmetry Index is calculated as the difference between right and left volumes divided by their mean (in percent).

4 - One of our patient MRI could not be segmented ("Finished with an error", 146036) and another one is segmented sub-optimally ("Registration is suboptimal", 146032). Since our sample size is not large, we do not want to exclude any patients. Do you have a suggestion for that? Should I do completely manual segmentation on ITK-Snap for these two patients?

You can tried by submitting the n_mmni* file Maybe the registration will be easier.

5 - On crisp segmentation file, cortex GM and cerebellum GM have the same label. Therefore, while doing the manual correction, I only change cortex so that I can subtract/add the difference to the cortex GM volume on the CSV file. However, I am not sure if the dura between the cerebrum and cerebellum (tentorium cerebelli) is included to cerebellum or cerebrum on the volBrain results file. Is it included to cortex GM or cerebellum?

I will ask to José.

Best,

Pierrick