Towards Fair Medical AI: Adversarial Debiasing of 3D CT Foundation Embeddings

2025-02-05

AI Summary

This study reveals implicit demographic biases—specifically age, sex, and race—in embeddings from 3D CT self-supervised foundation models, posing significant threats to clinical fairness. To address this, we propose the first adversarial debiasing framework tailored for 3D medical imaging foundation embeddings: a variational autoencoder (VAE)-based feature disentanglement architecture that jointly optimizes sensitive-attribute removal and downstream task preservation via adversarial training. Evaluated on the NLST dataset, our method completely eliminates encoding of all three sensitive attributes while maintaining robust lung cancer risk prediction performance—achieving stable AUCs exceeding 0.82 for both 1-year and 2-year risk forecasting. Moreover, it improves robustness against bias injection attacks by over 90%. Crucially, our approach achieves synergistic optimization of multi-dimensional fairness enhancement and embedding robustness without compromising diagnostic accuracy, establishing a novel paradigm for trustworthy medical AI.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Adversarial Learning & Robustness
• Natural Language Processing: Safety and Robustness

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs

Abstract

Self-supervised learning has revolutionized medical imaging by enabling efficient and generalizable feature extraction from large-scale unlabeled datasets. Recently, self-supervised foundation models have been extended to three-dimensional (3D) computed tomography (CT) data, generating compact, information-rich embeddings with 1408 features that achieve state-of-the-art performance on downstream tasks such as intracranial hemorrhage detection and lung cancer risk forecasting. However, these embeddings have been shown to encode demographic information, such as age, sex, and race, which poses a significant risk to the fairness of clinical applications. In this work, we propose a Variation Autoencoder (VAE) based adversarial debiasing framework to transform these embeddings into a new latent space where demographic information is no longer encoded, while maintaining the performance of critical downstream tasks. We validated our approach on the NLST lung cancer screening dataset, demonstrating that the debiased embeddings effectively eliminate multiple encoded demographic information and improve fairness without compromising predictive accuracy for lung cancer risk at 1-year and 2-year intervals. Additionally, our approach ensures the embeddings are robust against adversarial bias attacks. These results highlight the potential of adversarial debiasing techniques to ensure fairness and equity in clinical applications of self-supervised 3D CT embeddings, paving the way for their broader adoption in unbiased medical decision-making.

Problem

• Address demographic bias in 3D CT embeddings
• Maintain performance in critical medical tasks
• Ensure fairness in clinical AI applications

Innovation

• Adversarial debiasing with VAE
• Eliminates demographic information encoding
• Maintains predictive accuracy in tasks