應用聯邦學習於醫療人工智慧之倫理分析

Ethical Analysis for Applying Federated Learning in Medical AI

傳統機器學習應用於醫療領域，受制於隱私保護與資料需求間的兩難。機器學習模型依賴大量數據以提升品質與準確性，然而醫療數據因病人隱私限制，導致大量高價值臨床資料滯留於醫療機構，未獲充分利用。聯邦學習（Federated Learning）作為新興機器學習架構，允許參與者在不共享本地數據的前提下協同訓練模型，為突破傳統機器學習困境提供可能，並逐漸成為AI之醫學應用之新典範。然而，聯邦學習的實踐與部署仍面臨技術挑戰，需在病人利益、公平性、隱私等倫理價值間權衡取捨，妥善解決其相應而生之倫理議題。

Traditional medical machine learning faces a dilemma between data demands and privacy concerns: data is crucial for improving accuracy and performance, while privacy, as a necessary ethical standard, can also restrict the full utilization of medical data. Federated Learning (FL) enables collaborative model training without sharing local data, offering a potential solution to this conflict. However, FL also faces technical challenges. This paper first explores the role of FL in addressing the inherent data-privacy conflict in traditional medical machine learning, analyses the technical challenges FL encounters, and argues that FL must manage to balance three critical bioethical values: patient interests, fairness, and privacy.