論人工智慧醫療軟體之法規管制— 以美國及歐盟法為中心

近年來「智慧醫療」皆被各國政策列為重要的發展項目，人工智慧技術不但有望解決醫療體系人力短缺及資源分配不均的情形，甚至能幫助人們帶來更精準的治療方案、更低成本的醫療花費及更高品質的醫療照護。然而，人工智慧技術並非完美無缺，尤其不透明運作的本質，現存法規亦難以有效掌握並減低其風險。目前，在醫療領域最常被應用的人工智慧技術，係以「醫療軟體」形式分析輸入之健康資訊，提供醫護人員診斷及治療上之輸出建議。以美國及歐盟法規為例，皆為將醫療軟體納入「醫療器材」範圍管制，惟目前醫療器材法規並無為人工智慧特性所設計之管制措施，學者亦紛紛提出應為人工智慧醫療軟體管制，進行法規上的革新。人工智慧醫療軟體作為醫療產品之一種，本文主要關注其安全性及有效性之管制，檢視近年相關法規管制上有何變革，更進而思考管制上更適切的作法。本文首先觀察人工智慧應用於醫療場域之情形，包含其應用於醫療之目的、種類，以提出其作為一項醫療產品，於法規管制上之正當性、連結性及有效性相應之挑戰。其次，檢視美國及歐盟近年來於醫療器材法規上之增修，以及針對人工智慧醫療軟體所發布之政策文件及法律草案，並以管制主體、管制客體及管制方法三個面向，評析其於管制上的特點及優劣。最後，本文對於美國及歐盟在人工智慧醫療軟體之管制上進行比較分析，並結合法律及科學領域之學者文獻，提出法規對於人工智慧醫療軟體管制上得以加強之處，本文亦提出兩點管制建議：電子健康資料之整合及運用、以及多方參與而行政機關主導之管制，以供管制者參照。

關鍵字

人工智慧；醫療軟體；醫療器材；管制理論；管制架構；美國食品藥物管理署；歐盟醫療器材規則；歐盟人工智慧管制規則草案

並列摘要

In recent years, many countries have prioritized the development of artificial intelligence (AI) technology in healthcare. AI can bring positive changes, though, it is not a perfect solution, especially when it comes to its much-debated black box nature and other unknown risks that are beyond the capabilities of the current legal system. At this stage, AI in healthcare mostly appears in the form of “medical software” that generates advice for medical professionals. In the US and the EU, medical software is regulated as a “medical device,” yet the current legislation lacks mechanisms designed for the regulation of AI. Many scholars have urged to establish a new model for regulating AI-based medical software. By scrutinizing relevant law reforms in recent years, this Thesis aims to explore more appropriate solutions that can ensure the safety and effectiveness of AI-based medical software. First, through examining how AI is used in the medical field, including the purposes and the types of AI-based medical software, this Thesis identifies the regulatory challenges of such software in the aspects of legitimacy, effectiveness, and connection. Furthermore, this Thesis analyzes the pros and cons of current approaches in terms of the regulatory body, object, and method after studying the reforms of medical devices law in the US and the EU, and regulation proposals targeting AI-based software. This Thesis also conducts a comparative study on US and EU regulations of AI-based medical software, and argues that some issues have not been fully addressed. Finally, this thesis proposes two solutions, namely integrating electronic health data and introducing a government-led model that is participated by multiple parties.