文章摘要

深度学习在肿瘤组织病理图像分析中的应用

作者: 1,2徐贵璇, 1,2王阳, 1,2张杨杨, 1,2李春森, 1,2刘春霞, 1,3李锋
1 石河子大学医学院病理系,新疆 石河子 832002
2 石河子大学医学院第一附属医院病理科,新疆 石河子 832002
3 首都医科大学附属北京朝阳医院病理科与医学研究中心,北京 100020
通讯: 刘春霞 Email: liuliu2239@sina.com
李锋 Email: lifeng7855@126.com
DOI: 10.3978/j.issn.2095-6959.2021.06.035
基金: 新疆兵团科技发展专项(2018AB033)。

摘要

全载玻片数字扫描技术自1999年首次提出以来,在许多病理领域得到应用和验证。近年来,将先进的人工智能(artificial intelligence,AI)技术应用于医学诊断领域,不仅为改善医疗保健提供了新思路,也掀起了在肿瘤病理学领域研究的新浪潮。在大数据及数字显微技术背景下,深度学习(deep learning,DL)作为实现AI的一种新兴手段,在肿瘤检测、分类、转移和预后预测等组织病理图像分析中显示出巨大潜力。传统病理诊断结果受病理医师个人知识储备、临床经验以及逻辑思维方式的影响,主观性强且重复率低。AI作为一种新技术,在辅助病理医师进行病理诊断时,可以在一定程度上规避上述人为因素,减少人工失误,提高病理诊断的准确率和重复率,支持实时诊断决策。这不仅能够缓解医疗卫生资源的压力,而且能够为精准医疗助力。本文就DL在肺癌、乳腺癌、前列腺癌组织病理图像分析中的应用现状、机遇及挑战作一综述,并从病理医师角度讨论模型开发和应用监管中存在的问题。
关键词: 深度学习;组织病理图像;肺癌;乳腺癌;前列腺癌

Application of deep learning in tumor histopathological image analysis

Authors: 1,2XU Guixuan, 1,2WANG Yang, 1,2ZHANG Yangyang, 1,2LI Chunsen, 1,2LIU Chunxia, 1,3LI Feng
1 Department of Pathology, Medical College of Shihezi University, Shihezi Xinjiang 832002, China
2 Department of Pathology, First Affiliated Hospital, Medical College of Shihezi University, Shihezi Xinjiang 832002, China
3 Research Center of Pathology and Medicine, Beijing Chaoyang Hospital, Capital Medical University, Beijing 100020, China

CorrespondingAuthor: LIU Chunxia Email: liuliu2239@sina.com

DOI: 10.3978/j.issn.2095-6959.2021.06.035

Foundation: This work was supported by the Science and Technology Development Project of Xinjiang Production and Construction Corps, China (2018AB033).

Abstract

Many applications of whole slide imaging technology in the field of pathology have been verified since it was proposed in 1999. In recent years, the application of artificial intelligence (AI) technology in the field of medical diagnosis has not only provided new ideas for improving healthcare, but also set off a new wave of research in the field of tumor pathology. As an emerging means of AI implementation, deep learning (DL) shows great potential in the analysis of histopathological images, such as the detection, classification, metastasis, and prognosis prediction of tumors, in the context of big data and digital microscopy technology. Traditional pathological diagnosis results are affected by the pathologists’ personal knowledge, clinical experience, and logical way of thinking; such results are subjective and have low repetition rate. As a new technology, AI can help pathologists in making pathological diagnosis, avoid the above mentioned issues to a certain extent, reduce artificial errors, improve the accuracy and repetition rate of pathological diagnosis, and support real-time diagnosis decision. These advantages can not only ease the pressure on medical and health resources, but also support precision medicine. This article reviews the status quo, opportunities and challenges of DL application in pathological image analysis of lung cancer, breast cancer and prostate cancer, and discusses the problems in model development and application supervision from the perspective of pathologists.
Keywords: deep learning; histopathological image; lung cancer; breast cancer; prostate cancer