人工智能技术在计算机辅助诊断领域的发展新趋势  

The trend of artificial intelligence method in computer aided diagnosis

作　　者：赵愉[1] 王得旭 顾力栩[1] ZHAO Yu;WANG DeXu;GU LiXu(Institute of Medical Robotics,Department of Biomedical Engineering,Shanghai Jiao Tong University,Shanghai 200240,China)

机构地区：[1]上海交通大学医疗机器人研究院生物医学工程学院,上海200240

出　　处：《中国科学：生命科学》2020年第11期1321-1334,共14页Scientia Sinica(Vitae)

基　　金：北京市自然科学基金(批准号:L192006);上海交通大学医疗机器人研究院孵化基金资助。

摘　　要：计算机辅助诊断是医学影像分析发展的重要方向之一.该技术具有精确定量分析、可重复处理,降低影像医师工作负担等优点.本文对近年来国内外基于人工智能技术的计算机辅助诊断技术的研究进展进行了综述和讨论,介绍了当前针对医学影像的计算机辅助诊断系统所涉及的人工智能算法、流程及相关技术,分析了目前研究所存在的不足和挑战,并提出了有可能弥补这些不足的新思路.通过深入的分析研究发现,相比传统有监督学习方法,半监督学习方法更贴近医学任务的实际需求,同时也能大幅降低系统的实现成本,因此本文认为,对半监督方法的研究与发展是医学影像分析领域的研究新趋势,对医学影像的计算机辅助诊断系统的研究有重要意义.Computer aided diagnosis(CAD)has become an important aspect of medical imagology development.This technology has advantages in precise quantitative analysis,reproducibility and physician workload reduction.In this paper,the research literature of computer aided diagnosis based on artificial intelligence has been reviewed,where the typical artificial intelligence algorithm procedure and related technologies in medical image-based CAD system have also been introduced.We also analysed the weaknesses and challenges of the current CAD system,and proposed the solutions and suggestions to counter these deficiencies.The recent study indicates that compared with the traditional supervised learning method,the semi-supervised active learning is more suitable to the practical requirements of clinical tasks.It will significantly reduce the implementation cost of the CAD system.Therefore,we believe that the research and development semi-supervised active learning method are of great significance to the medical image-based CAD system.

关 键 词：医学影像分析 计算机辅助诊断 人工智能 主动学习 样本标注成本 

分 类 号：TP391.7[自动化与计算机技术—计算机应用技术] TP18[自动化与计算机技术—计算机科学与技术] R445[医药卫生—影像医学与核医学]