syedhaseeburrehman77-hash/BrainTumor_Segmentation ? reverse-engineered prompt

Build me a Python project for brain tumor segmentation with federated learning using Flower.

I want it to work with the FeTS 2022 dataset, with a way to verify the dataset first, then run a small CPU only federated experiment across a few real client partitions. Please support both FedAvg and FedProx, and make sure the client side uses the proximal term correctly when FedProx is selected. I also want local training to hold out part of each institution’s labeled data for validation, and report Dice and HD95 for enhancing tumor, tumor core, and whole tumor each round.

Please include a simple centralized baseline too, just for pooled data comparison, but keep that separate from the federated result. Make it easy to set the dataset paths, and make the default run practical on a normal machine with Python 3.11. If you need to check current Flower or PyTorch docs online for the right API usage, go ahead.