Closed Jbartlett6 closed 2 years ago
Currently subject 101410 has been removed from the training set - the only alternative is to just mask out sufficiently many pixels from the edge of the image, then pass this into the training. Due to the patch-wise training of the network this shouldn't make a huge difference.
This has now been properly fixed, the images are padded when they are loaded into RAM in pytorch, this prevents having to process the images in mrtrix. A quick and easy fix.
The subject mentioned causes an error when extracting 9 by 9 by 9 patches since the brain mask is too close to the edge of the image.