[QUESTION] Preprocessing pediatric brain data

[aloyspx]

Hello,

My group would like to evaluate your model on our internal dataset. However, when performing registration from the MRI sequences onto the SRI24 atlas, following the procedure described in your paper and using the brain-lesion-preprocessing repository, we are encountering noticeably choppy results. This may be expected, as pediatric brain volumes are smaller than the adult SRI24 atlas, and the default in the preprocessing pipeline appears to use rigid registration for atlas alignment. We've thought of using affine registration to allow the brain to scale, however, we don't want to distort the brain.

Could you please clarify the exact registration procedure or parameters you used for this step? If possible, would you be open to sharing the script or configuration you used?

Thank you for your time.

[neuronflow]

Thanks for your interest in our work.
Do you have images to illustrate your problems?

We used an older version of the preprocessing package for the manuscript. The API has changed since then; therefore, digging out the old script will not help you much.

I recommend using the atlas correction step. If you are unhappy with ANTs' results, it is worth trying the niftyreg registration backend.

I would not recommend non-rigid registrations if you are interested in volume calculations (as we are and were).

For tumors near the optic nerve (outside the brain) skullstripping/defacing obviously need to be skipped :)

[neuronflow]

Preprocessing is now explained in this PR which should be merged soon.
#23