Recommended architecture for localization and classification based on medical images [MRI]

Hi all,
We are using fast.ai for brain tumors localization (by bounding box) and classification, based on MRI images [may be single / multiple lesions per image]
Using resnet34, and single lesion per image we got fair results
We’re still working on improving results and adapting the system for multiple lesions detection and classification
Does anyone have any experience / recommendations for architecture that is most suitable for MRI data based localization and classification?

Thanks a lot
Moran

i would expect u-net to be a good choice

also there are some great articles on researchgate