harvardairobotics/FairFedMed

---
license: mit
task_categories:
- image-classification
modality:
- image
language:
- en
tags:
- medical
- ophthalmology
- radiology
- fairness
- federated-learning
- fundus
- glaucoma
- chest-xray
- OCT
pretty_name: FairFedMed
size_categories:
- 10K<n<100K
---

# Dataset Card: FairFedMed

## Dataset Summary

FairFedMed is the first federated learning (FL) benchmark dataset for medical imaging with demographic annotations, designed to study **group fairness across institutions** in a federated setting. It comprises two subsets spanning ophthalmology and chest radiology, enabling research on fairness-aware federated learning under realistic cross-institutional data heterogeneity.

This dataset was introduced in the IEEE Transactions on Medical Imaging 2025 paper: [FairFedMed: Benchmarking Group Fairness in Federated Medical Imaging with FairLoRA](https://ieeexplore.ieee.org/document/11205878).

## Dataset Details

### Dataset Description

- **Curated by:** Minghan Li, Congcong Wen, Yu Tian, Min Shi, Yan Luo, Hao Huang, Yi Fang, Mengyu Wang
- **Institution:** Harvard Medical School / Harvard AI and Robotics Lab
- **License:** See individual subset licenses (CheXpert and MIMIC-CXR have their own terms)
- **Repository:** [Harvard-AI-and-Robotics-Lab/FairFedMed](https://github.com/Harvard-AI-and-Robotics-Lab/FairFedMed)
- **Paper:** [IEEE TMI 2025](https://ieeexplore.ieee.org/document/11205878) / [arXiv:2508.00873](https://arxiv.org/abs/2508.00873)

### Subsets

#### FairFedMed-Oph (Ophthalmology)

| Field | Value |
|------------------|-------|
| **Task** | Glaucoma detection (binary classification) |
| **Modalities** | 2D SLO fundus images, 3D OCT B-scans |
| **Scale** | 15,165 patients |
| **Demographics** | Age, gender, race, ethnicity, preferred language, marital status (6 attributes) |
| **FL Setup** | Multi-site federated (3 sites) |

#### FairFedMed-Chest (Chest Radiology)

| Field | Value |
|------------------|-------|
| **Task** | Chest pathology classification |
| **Sources** | [CheXpert](https://stanfordmlgroup.github.io/competitions/chexpert/) + [MIMIC-CXR](https://physionet.org/content/mimic-cxr/2.1.0/) |
| **Demographics** | Age, gender, race (3 attributes) |
| **FL Setup** | 2 clients simulating cross-institutional FL |

## Uses

### Direct Use

Research on group fairness in federated medical image classification, including studies of demographic disparity across institutions and evaluation of fairness-aware FL methods.

### Out-of-Scope Use

Clinical diagnosis, commercial applications. Note that FairFedMed-Chest inherits the usage restrictions of CheXpert and MIMIC-CXR — consult those datasets' licenses before use.

## Evaluation

| Metric | Description |
|-------------|-------------------------------------|
| **AUC** | Area Under ROC Curve |
| **ESAUC** | Equalized Selection AUC |
| **EOD** | Equalized Odds Difference |
| **SPD** | Statistical Parity Difference |
| **Group AUC** | Per-demographic-group AUC |

## Associated Method: FairLoRA

The paper introduces **FairLoRA**, a fairness-aware FL framework using SVD-based low-rank adaptation. It customizes singular values per demographic group while sharing singular vectors across clients for communication efficiency.

Supported backbones: ViT-B/16, ResNet-50.

## Citation

**BibTeX:**

```bibtex
@ARTICLE{11205878,
author={Li, Minghan and Wen, Congcong and Tian, Yu and Shi, Min and Luo, Yan and Huang, Hao and Fang, Yi and Wang, Mengyu},
journal={IEEE Transactions on Medical Imaging},
title={FairFedMed: Benchmarking Group Fairness in Federated Medical Imaging with FairLoRA},
year={2025},
pages={1-1},
doi={10.1109/TMI.2025.3622522}
}
```

**APA:**

Li, M., Wen, C., Tian, Y., Shi, M., Luo, Y., Huang, H., Fang, Y., & Wang, M. (2025). FairFedMed: Benchmarking Group Fairness in Federated Medical Imaging with FairLoRA. *IEEE Transactions on Medical Imaging*. https://doi.org/10.1109/TMI.2025.3622522