Improving the Fairness of Artificially Intelligent Skin Disease Detectors Using Stable Diffusion

Overview

This study addresses the ethical implications of using artificially intelligent neural networks for medical diagnosis, with a focus on skin diseases. While such networks have demonstrated high accuracy in diagnostic applications, they have been found to misdiagnose minority populations at a disproportionately high rate due to imbalanced training datasets. To prevent this perpetuation of inequities, our study artificially balances a skin disease training dataset by adding stable diffusion-generated synthetic images depicting skin diseases on underrepresented skin tones. Additionally, prompt engineering and fine-tuning are utilized to optimize the photorealism and diversity of these synthetic images. Notably, our study is the first to achieve a statistically significant increase in the fairness of a skin disease detector. Specifically, we improve fairness by over 50%, as quantified by the average odds difference metric, thereby helping to foster more equitable healthcare outcomes.

Competition history

ISEF 2024 Robotics and Intelligent Machines · Entry ROBO069 · Los Angeles, California, United States

Resources

ISEF abstract page

Related projects

ISEF · 2024

Improving Racial Equity in Skin Cancer Detection: Using Artificial Intelligence Driven Synthetic Image Generation and Cascading Convolutional Neural Networks to Diagnose Cancer in Lesions of Varying Skin Tones

ISEF · 2026

DermEquity: A Universal Safety Framework for Medical AI Using Worst-Case Bias Certification and Influence-Based Attribution Validated in Skin Cancer Screening

ISEF · 2021

Automated Bias Reduction in Deep Learning Based Melanoma Diagnosis Using a Semi-Supervised Algorithm

ISEF · 2023

DermaSkan: Using Convolutional Neural Networks to Detect and Prevent Skin Cancer in Underrepresented Communities by Using an Android App

Closest projects by meaning, across every fair and year in the corpus.