Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Add more contrasts for training a multi-contrast model #3

Open
naga-karthik opened this issue Jan 24, 2023 · 5 comments
Open

Add more contrasts for training a multi-contrast model #3

naga-karthik opened this issue Jan 24, 2023 · 5 comments

Comments

@naga-karthik
Copy link
Member

Our experiment based on the preliminary hypothesis used on a single contrast (T2w) from the CanProCo dataset. It is important to gather the remaining contrasts for each subject and train a multi-channel/multi-contrast model (each additional contrast being added as an additional channel). This will be useful for robustness and also contrast-agnostic centerline detection.

@valosekj Would you please be able to preprocess the remaining contrasts as we did for T2w?

@valosekj
Copy link
Member

@valosekj Would you please be able to preprocess the remaining contrasts as we did for T2w?

For remaining contrasts, we do not have ground truth spinal cord segmentation. Thus, we cannot easily obtain the centerline for these contrasts.

But I can co-register all contrasts to T2w. Then, we would be able to use the T2w centerline also for other contrasts.

@jcohenadad
Copy link
Member

 For remaining contrasts, we do not have ground truth spinal cord segmentation. Thus, we cannot easily obtain the centerline for these contrasts.
But I can co-register all contrasts to T2w. Then, we would be able to use the T2w centerline also for other contrasts.

It would be much easier to instead leverage our many other datasets on which we already have the spinal cord segmentations (spine-generic, etc.)

@naga-karthik
Copy link
Member Author

naga-karthik commented Jan 24, 2023

But I can co-register all contrasts to T2w. Then, we would be able to use the T2w centerline also for other contrasts.

Just for my understanding, this should also work right? I understand that it will take time but it should be good if we want to use all contrasts from CanProCo?

@jcohenadad
Copy link
Member

Just for my understanding, this should also work right? I understand that it will take time but it should be good if we want to use all contrasts from CanProCo?

yes, it should work, but registration needs to be QCed, which takes a bit of time

@valosekj
Copy link
Member

valosekj commented Feb 2, 2023

Just for my understanding, this should also work right? I understand that it will take time but it should be good if we want to use all contrasts from CanProCo?

yes, it should work, but registration needs to be QCed, which takes a bit of time

I tried to co-register other contrasts to T2w based on nii headers using -identity 1:

sct_register_multimodal -i ${contrast}.nii.gz -d ${file_t2w}.nii.gz -o ${contrast}2${file_t2w}.nii.gz -identity 1 -x nn

To allow easy QC (we have >1000 coregistered images), I'm projecting the T2w centerline to the coregistered image using:

# Create QC report to assess registration quality
# Note: registration quality is assessed by comparing the ${contrast} image to the T2w centerline
sct_qc -i ${contrast}2${file_t2w}.nii.gz -s ${FILESEG}_centerline.nii.gz -p sct_get_centerline -qc ${PATH_QC} -qc-subject ${SUBJECT}
sct_qc -i ${contrast}2${file_t2w}.nii.gz -s ${FILESEG}_centerline.nii.gz -p sct_label_vertebrae -qc ${PATH_QC} -qc-subject ${SUBJECT}

Example (coregistered PSIR image and T2w_centerline):
image

Based on the initial quick QC assessment, it seems that the registration works well for most subjects, and also, it seems that this type of QC is reasonable!

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

3 participants