FairGen: preference-aligned diffusion for demographically equitable medical image synthesis

Zhimin Li, Ruichen Zhang, Zhen Tan, Howard J Aizenstein, Jingtong Hu, Tianlong Chen · 2026

Read FairGen: preference-aligned diffusion for demographically equitable medical image synthesis by Zhimin Li, Ruichen Zhang, Zhen Tan, Howard J Aizenstein, Jingtong Hu, Tianlong Chen (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/PMC13627665.txt.

Description

Zhimin Li and five colleagues, publishing in 2026, present FairGen, an image-making system designed to keep medical AI honest across demographic groups. Training data for diagnostic models skew heavily toward some populations, and diseases can look different across groups, so rare presentations get missed. FairGen is a diffusion generator steered by physician-aligned preferences, synthesizing balanced image sets that keep the visual marks of disease while widening who is represented.

Across skin images, chest X-rays, and brain MRI benchmarks, fairness gains were large, reaching over ninety percent for skin, while diagnostic accuracy held close to models trained on real clinic data. Independent clinician review and outside cohorts backed the gains. Balanced synthetic data, the authors argue, can patch holes that real-world collection keeps leaving.

Catalog record

CREATORZhimin Li, Ruichen Zhang, Zhen Tan, Howard J Aizenstein, Jingtong Hu, Tianlong Chen
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE64346
LICENSEhttps://creativecommons.org/licenses/by/4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/article/MED/PMC13627665

Browse the full library · About the Tower of Babel