人工智慧與精準醫學

Artificial Intelligence and Precision Medicine

精準醫學目標是尋找個人潛在生理和臨床特徵的差異，用以預估疾病風險和治療反應的不同，引導個人化治療達成病患預後改善。由於電子病歷的廣泛使用以及人類基因的解碼完成，產生了多維的醫療大數據供精準醫學發展。這些大數據可以運用人工智慧（artificial intelligence）的機器學習（machine learning）演算法，進行多維的數據分析，找到隱式臨床表現型或基因型結構，供臨床醫生精準治療或預防措施之用。當代醫學是以隨機臨床試驗為基礎，尋求群體平均最佳而非個人最佳的治療效果，與精準醫學決策過程形成鮮明對比。然而，利用人工智慧發展精準醫學的過程會遭遇許多的挑戰，包含：（1）機器學習演算法過程的不透明，無法形成具體有效的醫學知識，（2）機器學習的演算法是架構在相關推論並不是因果推論，以及（3）沒有一個完善的架構去規範機器學習醫療軟體，落地融入臨床決策過程。本文將就這幾個重點，回顧當前文獻，討論利用人工智慧，發展精準醫學的機會和挑戰。

The goal of precision medicine is to find potential phenotypical and genotypical differences of individuals that determine disease risk and treatment response, so as to guide personalized treatment with improved outcomes. Due to the widely use of electronic medical records and the decoding of human genome, multi-dimensional medical big data has been generated for the development of precision medicine. These big data can use artificial intelligence (machine learning) algorithms to find implicit phenotypical or genotypical structures to guide precise treatment or preventive measures. Contemporary medicine is based on randomized clinical trials, seeking the population average rather than individual best treatment results, which is in sharp contrast with the precision medicine decision-making process. However, the process of using artificial intelligence to develop precision medicine will encounter many challenges, including: (1) The process of machine learning algorithms is a black box and cannot form specific and effective medical knowledge. (2) Machine learning algorithms are based on associational inferences rather than causal inference, and (3) there is no legal framework to regulate the implementation of machine learning software into the clinical decision-making process. This article will review the current literatures on these key points and discuss the opportunities and challenges of using artificial intelligence to develop precision medicine.