metadata
license: cc-by-nc-4.0
task_categories:
- image-classification
- audio-classification
tags:
- federated-learning
- medical-imaging
- benchmark
- sample
FL-MedClsBench — Representative Sample
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).
Sampling Methodology
- 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) included in full
Included Datasets
| Dataset | Modality | Clients | Task | Full Size |
|---|---|---|---|---|
| FL-Skin | Dermoscopy | 4 | 8-class skin lesion | ~800 MB |
| FL-OCT | Retinal OCT | 5 | 8-class OCT disease | ~1.6 GB |
| FL-COVID-XR | Chest X-ray | 5 | 3-class COVID-19 | ~400 MB |
| FL-COVID-CT | Chest CT | 5 | 3-class COVID-19 | ~500 MB |
| FL-ECG | 12-lead ECG | 3 | 20-class multilabel | ~7 GB |
| FL-BCa | T2W MRI | 4 | Binary bladder cancer | ~1 GB |
| FL-DR | Fundus | 5 | DR grading (5-class) | ~640 MB |
| FL-Glaucoma | Fundus | 4 | Binary glaucoma | ~275 MB |
| FL-Fetal | Ultrasound | 6 | 6-class fetal plane | ~740 MB |
| FL-TB | Chest X-ray | 5 | Binary tuberculosis | ~1.5 GB |
Full Dataset
➡️ https://huggingface.co/datasets/FL-MedClsBench/FL-MedClsBench
Citation
@inproceedings{fl_medclsbench2026,
title={Beyond Synthetic Splits: A Benchmark for Federated Learning on Real-World Medical Data Classification},
year={2026}
}