基于图特征的组织病理学图像分析方法的最新发展情况与展望  

The Latest Development and Prospects of Histopathological Image Analysis Methods Based on Graph Features

作　　者：何睿琳 杨欣怡 孙洪赞[2] 李晨[1] HE Ruilin;YANG Xinyi;SUN Hongzan;LI Chen(College of Medicine and Biological Information Engineering,Northeastern University,Shenyang,Liaoning 110819,China;Shengjing Hospital of China Medical University,Shenyang,Liaoning 110004,China)

机构地区：[1]东北大学,医学与生物信息工程学院,辽宁沈阳110819 [2]中国医科大学附属盛京医院,辽宁沈阳110004

出　　处：《数据与计算发展前沿》2024年第2期101-116,共16页Frontiers of Data & Computing

基　　金：国家自然科学基金重点项目(82220108007)。

摘　　要：【目的】本文旨在综述最近五年人工智能在辅助组织病理学分析方面的研究进展,主要是图特征方法的应用、当前面临的问题以及未来的挑战。【方法】文章回顾了图论在组织病理学图像分析中的应用,包括图像分割、检测和分类。探讨了图像拓扑结构特征提取的各种图构建算法,例如经典的最小生成树算法及其衍生创新算法等,并分析了图卷积神经网络等网络结构的性能。【结果】通过结构图提取的图特征能够有效表示组织病理学图像中的拓扑信息,有助于实现精确的肿瘤分割、检测以及分类、分级等任务。此外,图特征方法综合全局与局部特征,提供了一种系统化的分析方式,促进了对复杂病理学图像的理解。【结论】图特征与先进的机器学习技术相结合在组织病理学图像分析中展现出强大的潜力,未来这些方法将被优化以提高临床诊断的准确性和效率。[Objective]This article aims to review the research progress of artificial intelligence in assisting histopathology analysis in the past five years,mainly focusing on the application of graph feature methods,current problems,and future challenges.[Methods]The article reviews the application of graph theory in histopathological image analysis,including image segmentation,detection,and classification,explores various graph construction algorithms for feature extraction of image topological structures,such as the classic minimum spanning tree algorithm and its derivative innovative algorithms,and analyzes the performance of network structures such as graph convolutional neural networks.[Results]The graph features extracted through structural maps can effectively represent topological information in histopathological images,which helps to achieve accurate tumor segmentation,detection,classification,and cancer grading tasks.In addition,the graph feature method provides a systematic analysis approach by considering global and local features,promoting the understanding of complex tissue pathology images.[Conclusions]The combination of graph features and advanced machine learning technologies has shown strong potential in histopathological image analysis.In the future,these methods will be optimized to improve the accuracy and efficiency of clinical diagnosis.

关 键 词：组织病理学图像 图特征 人工智能 机器学习 肿瘤辅助诊断 

分 类 号：R361.2[医药卫生—病理学] TP391.41[医药卫生—基础医学]