以人工智能预测人类胚胎整倍性为例讨论算法解释性的意义  

Clinical implications of algorithmic interpretations of artificial intelligence in human embryo ploidy prediction

作　　者：张孝东[1] 王浩[2] 郭小妮 韩伟[1] 韩树标[1] 黄国宁[1] Zhang Xiaodong;Wang Hao;Guo Xiaoni;Han Wei;Han Shubiao;Huang Guoning(IVF Clinical Center,Chongqing Maternal and Child Healthcare Hospital,Affiliated Women's and Children's Hospital of Chongqing Medical University,Chongqing400010,China;Institute for Medical Device Control,National Institutes for Food and Drug Control,Beijing102629,China)

机构地区：[1]重庆市妇幼保健院、重庆医科大学附属妇女儿童医院生殖医学中心,重庆400010 [2]中国食品药品检定研究院医疗器械检定所,北京102629

出　　处：《中华生殖与避孕杂志》2025年第1期31-38,共8页Chinese Journal of Reproduction and Contraception

基　　金：重庆市科技局创新与应用发展专项重点项目(CSTB2022TIAD-KPX0146)。

摘　　要：人工智能(artificial intelligence,AI)技术有望辅助医生提升胚胎评估的准确性和效率。然而,胚胎发育是一个连续的动态过程,什么时段比较有意义?还是整个发育过程都需要考虑?部分研究团队采用静态图像分析丢失诸多重要信息;另一部分团队利用算法驱动的"黑盒"模型计算胚胎发育视频,解释性有限。机器学习及深度学习由于固有的复杂性易被滥用,为了更精准地应用AI,本文以AI预测人类胚胎整倍性为例讨论算法解释性的意义及临床价值。Artificial intelligence(AI)technology is expected to assist physicians in improving the accuracy and efficiency of embryo assessment.However,embryo development is a continuous and dynamic process,when is meaningful or the whole development process need to be considered?Some research teams use static image analysis,which loses much important information,and others utilize algorithm-driven applications of"black-box"models to analyse embryo videos,which have limited their interpretability or explainability.Machine learning or deep learning is prone to abused due to its inherent complexity,and in order to apply AI more accurately,this paper discusses the clinical implications of algorithmic interpretations of AI in human embryo ploidy prediction.

关 键 词：人工智能 胚胎 机器学习 深度学习 黑盒 整倍性 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] R714.8[自动化与计算机技术—控制科学与工程]