Some images in BRACS & BACH are fairly big. Therefore resizing to 224x224 as most FMs require could lead to significant information loss and resulting in non-common MPPs. We should apply patching to these datasets and then use a multi instance learning head such as ABMIL for classification.