Input MR resizing #8
Description
Dear developers,
I was going through your code and I've noticed that you in the data loader you resize the MRs to 96x96x96. If I understand how MONAI preprocess data, this does not crop air/neck so it just downsample pixels.
I'm wondering if you reccomend any preprocessing therefore to preserve resolution (e.g.: affine or even just linear registration to some standard bounding box). Related to that, at which "effective spatial resolution" did you train? Given the importance of cortical thickness, if you only used MRs that were resampled to voxels bigger than 2mm isotropic, do you think that this could be an issue?
Thank you in advance for the great idea of a foundation model!