Dataset Card: FairGenMed

Dataset Summary

FairGenMed is the first dataset for studying fairness in medical generative models. It provides detailed quantitative clinical measurements alongside demographic annotations to investigate the semantic correlation between text prompts and anatomical regions across demographic subgroups. The dataset supports both generative model evaluation and downstream classification tasks for glaucoma detection.

This dataset accompanies the FairDiffusion framework — an equity-aware latent diffusion model that enhances fairness in medical image generation via Fair Bayesian Perturbation — published in Science Advances (2025).

Dataset Details

Dataset Description

Field	Value
Institution	Department of Ophthalmology, Harvard Medical School
Task	Glaucoma detection; fairness evaluation of generative models
Modality	Scanning Laser Ophthalmoscopy (SLO) fundus images, OCT B-scans
Scale	10,000 subjects
Image size	512 × 664 (SLO fundus)
License	CC BY-NC-ND 4.0

• Curated by: Yan Luo, Muhammad Osama Khan, Congcong Wen, Muhammad Muneeb Afzal, Titus Fidelis Wuermeling, Min Shi, Yu Tian, Yi Fang, Mengyu Wang
• License: CC BY-NC-ND 4.0 — non-commercial research only
• Paper: Science Advances, Vol. 11, No. 14 (2025)
• Contact: harvardophai@gmail.com, harvardairobotics@gmail.com

Data Fields

Each subject includes one SLO fundus image and one .npz file. The NPZ files contain:

Field	Description
glaucoma	Disease label: 0 = non-glaucoma, 1 = glaucoma
oct_bscans	OCT B-scan images
race	0 = Asian, 1 = Black, 2 = White
male	0 = Female, 1 = Male
hispanic	0 = Non-Hispanic, 1 = Hispanic
maritalstatus	0 = Married/Partnered, 1 = Single, 2 = Divorced, 3 = Widowed, 4 = Legally Separated, -1 = Unknown
language	0 = English, 1 = Spanish, 2 = Other

Clinical Metadata

All clinical measurements for the 10,000 samples are provided in data_summary.csv:

Column	Description
cdr_status	Cup-disc ratio status
md_severity	Severity of vision loss
se_status	Spherical equivalent status

Demographics

6 demographic attributes are annotated per subject: age, gender, race, ethnicity, preferred language, and marital status.

Uses

Direct Use

• Fairness evaluation of medical generative models (text-to-image diffusion)
• Glaucoma detection with demographic fairness analysis
• Studying semantic correlations between text prompts and anatomy across subgroups

Out-of-Scope Use

Clinical decisions, patient care, or any commercial application. This dataset shall not be used for clinical decisions at any time.

Associated Method: FairDiffusion

FairDiffusion is an equity-aware latent diffusion model built on Stable Diffusion 2.1, trained with Fair Bayesian Perturbation to reduce demographic bias in generated medical images. It is evaluated on FairGenMed (ophthalmology), HAM10000 (dermatology), and CheXpert (chest radiology).

Citation

BibTeX:

@article{FairDiffusion_Science_Advances_2025,
  author = {Yan Luo and Muhammad Osama Khan and Congcong Wen and Muhammad Muneeb Afzal and Titus Fidelis Wuermeling and Min Shi and Yu Tian and Yi Fang and Mengyu Wang},
  title = {FairDiffusion: Enhancing equity in latent diffusion models via fair Bayesian perturbation},
  journal = {Science Advances},
  volume = {11},
  number = {14},
  pages = {eads4593},
  year = {2025},
  doi = {10.1126/sciadv.ads4593}
}

APA:

Luo, Y., Khan, M. O., Wen, C., Afzal, M. M., Wuermeling, T. F., Shi, M., Tian, Y., Fang, Y., & Wang, M. (2025). FairDiffusion: Enhancing equity in latent diffusion models via fair Bayesian perturbation. Science Advances, 11(14), eads4593. https://doi.org/10.1126/sciadv.ads4593