Beyond Synthetic Splits: A Benchmark for Federated Learning on Real-World Medical Data Classification
This is a representative sample of the full FL-MedClsBench dataset (>14 GB).
Images: 5 samples randomly selected per client (seed=42)
FL-BCa (3D NIfTI MRI): 2 volumes per center
FL-ECG (HDF5 ECG signals): full dataset included (~7 GB, 3 files)
Metadata: All CSV split files (train/val/test × 3 seeds)… See the full description on the dataset page:
https://huggingface.co/datasets/ccbi/FL-MedClsBench.
