毒性病理学及人工智能数字组织图像分析质量控制概述
Overview of quality control for digital tissue image analysis in artificial intelligence and toxicologic pathology
滕伊洋 1张亚群 2李一昊 3钱庄 2汪溪洁 1吕建军3
作者信息
- 1. 中国医药工业研究总院,上海 201203;上海益诺思生物技术股份有限公司,上海 200043
- 2. 益诺思生物技术南通有限公司,南通 226133
- 3. 湖北天勤鑫圣生物科技有限公司,武汉 430207
- 折叠
摘要
随着人工智能和机器学习的快速发展,人工智能对全切片图像的诊断几乎可以媲美病理学家,建立人工智能算法需要大量的数字组织图像训练集数据.数字组织图像分析是通过各种算法分析全切片图像,并从其中提取大量复杂的定量数据集.数字组织图像分析的质量控制不仅非常重要,而且是确保建立高质量数据集和AI算法的基础和前提.本文简要概述了全切片图像的质量控制策略、数字组织图像分析的影响因素和质量控制方法、数字组织图像分析结果的质量控制方法、毒性病理学家在数字组织图像分析中的作用、数据解释和报告以及数字组织图像用于毒性病理学诊断及AI的挑战,以期为我国药物非临床安全性评价毒性试验中使用全切片图像进行毒性病理学诊断及建立AI各种算法提供一定参考.
Abstract
With the rapid development of artificial intelligence(AI)and machine learning(ML),the diagnosis of whole slide images(WSIs)by AI is almost comparable to that by pathologists.Establishment of algorithms of AI needs a large number of digital tissue image training set data.Digital tissue image analysis analyzes WSIs through various algorithms and extracts a large number of complex quantitative data sets from WSIs.The quality control(QC)of digital tissue image analysis is not only very important but also the basis and premise to ensure the establishment of high-quality data sets and AI algorithms.The paper briefly overviews the QC strategy for WSIs,factors affecting digital tissue image analysis,QC methods of digital tissue image analysis,QC methods of results of the digital tissue image analysis,roles of toxicologic pathologists in digital tissue image analysis,data interpretation and reporting,as well as the challenges of using digital tissue images in toxicologic pathology diagnosis and AI itself,hoping to provide references for using WSIs in toxicologic pathology diagnosis and establishing various AI algorithms in toxicity studies of non-clinical safety evaluation of drugs in China.
关键词
毒性病理学/人工智能/数字组织图像分析/质量控制/全切片图像Key words
toxicologic pathology/artificial intelligence/digital tissue image analysis/quality control/whole slide images引用本文复制引用
基金项目
江苏省新药一站式高效非临床评价公共服务平台建设项目(BM2021002)
出版年
2024
NETL
NSTL
万方数据