Use the attMIL approach for MRI which Marta used for the GBM for our breast dataset. At this point, we only have one sequence (T2w images). it's an adapted version of marugoto.