FedCIAL: Federated Color-Invariant Adversarial Learning for Enhancing Fairness and Performance in Skin Lesion Classification

Abstract

Bias in skin lesion classification, particularly in federated learning (FL) environments, poses a significant challenge due to the diversity in skin tone representation. In this work, we propose Federated Color-Invariant Adversarial Learning (FedCIAL), a novel approach that leverages known color distribution shifts to generate target samples for training a color-invariant feature extractor without requiring shared data. Experimental results on the Fitzpatrick17k dataset show that FedCIAL outperforms the state-of-the-art model FeSViBS, achieving an average accuracy of 0.7754, compared to 0.7666 for the baseline, with a statistically significant improvement (p = 0.044). Additionally, FedCIAL improves model fairness, reducing the standard deviation across clients to 0.044, compared to 0.053 for the baseline. These findings demonstrate that FedCIAL not only enhances performance but also offers a promising solution for fairer federated learning models in medical imaging.