LesionTABE: Equitable AI for Skin Lesion Detection

TL;DR

LesionTABE tackles bias against darker skin tones in dermatology AI by uniting adversarial debiasing with dermatology-focused foundation-model embeddings. Evaluated on malignant and inflammatory skin conditions using Fitzpatrick17k, PAD-UFES, and SCIN, the approach consistently improves fairness (notably Equality of Opportunity) while maintaining or modestly improving diagnostic accuracy. The best configuration—LesionTABE with LesionCLIP—delivers substantial fairness gains (≈34% relative) and competitive accuracy, underscoring the potential of foundation-model debiasing for equitable clinical AI. The study also introduces a standardized external-validation framework and highlights the critical role of external tests in assessing bias mitigation for real-world deployment.

Abstract

Bias remains a major barrier to the clinical adoption of AI in dermatology, as diagnostic models underperform on darker skin tones. We present LesionTABE, a fairness-centric framework that couples adversarial debiasing with dermatology-specific foundation model embeddings. Evaluated across multiple datasets covering both malignant and inflammatory conditions, LesionTABE achieves over a 25\% improvement in fairness metrics compared to a ResNet-152 baseline, outperforming existing debiasing methods while simultaneously enhancing overall diagnostic accuracy. These results highlight the potential of foundation model debiasing as a step towards equitable clinical AI adoption.

Figures (2)

• Figure 1: EOM fairness scores vs Overall Balanced Accuracy for malignant lesion detection. TABE with LesionCLIP exhibits optimal trade-off compared to all other combinations.
• Figure 2: PQD fairness scores vs Overall Balanced Accuracy for eczema vs psoriasis classification. TABE with LesionCLIP exhibits optimal trade-off compared to other combinations.