基于双路径特征融合的结肠组织病理腺体分割方法  

Colonic tissue pathological gland segmentation method based on dual-path feature fusion

作　　者：王红玉 张墺琦 卜起荣[3] 崔磊[3] 冯筠[3] WANG Hongyu;ZHANG Aoqi;BU Qirong;CUI Lei;FENG Jun(School of Computer Science and Technology,Xi′an University of Posts and Telecommunications,Xi′an 710121,China;Shaanxi Key Laboratory of Network Data Analysis and Intelligent Processing,Xi′an University of Posts and Telecommunications,Xi′an 710121,China;School of Information Science and Technology,Northwest University,Xi′an 710127,China)

机构地区：[1]西安邮电大学计算机学院,陕西西安710121 [2]西安邮电大学陕西省网络数据分析与智能处理重点实验室,陕西西安710121 [3]西北大学信息科学与技术学院,陕西西安710127

出　　处：《西北大学学报（自然科学版）》2021年第4期577-586,共10页Journal of Northwest University（Natural Science Edition）

基　　金：国家自然科学基金青年项目(62001380)。

摘　　要：结肠腺癌是一种严重危害人们生命健康的常见癌症,作为癌症检测与诊断的关键环节,腺体分割在结肠腺癌计算机辅助诊断研究中至关重要。针对结肠组织病理图像中腺体分割存在不同癌变水平腺体外观差异大,单一模型难以同时实现对良性、恶性腺体的高精度和高形状相似性分割的问题,设计了一种基于双路径特征融合结肠组织病理图像腺体分割网络。该网络利用带注意力的上下文特征提取路径和空间特征提取路径,获得较大的感受野和空间信息,增强网络对腺体形态学特征的学习能力,最终提升了腺体自动分割的性能。在Warwick-QU数据集上进行实验,与目前流行的分割算法对比,该文算法在不同类型测试集上的Dice系数、F1得分和Hausdorff距离均取得较好的性能,模型泛化性较强,具有重要的应用前景。Adenocarcinoma of colon is a kind of common cancer which seriously endangers people′s life and health.As the key link of cancer detection and diagnosis,gland segmentation is very important in computer-aided diagnosis of colonic adenocarcinoma.In view of the large differences in the appearance of different cancerous levels of glands in colon tissue pathological images,it is difficult for a single model to segment benign and malignant glands with high accuracy and high shape similarity.A dual path feature based fusion algorithm for colon pathological image segmentation is designed.The network uses context feature extraction path and spatial feature extraction path with attention to obtain larger receptive field and spatial information,enhance the network′s ability to learn gland morphological features,and finally improve the performance of gland automatic segmentation.Experiments are carried out on Warwick-Qu data sets.Compared with the current popular segmentation algorithms,the proposed algorithm achieves better performance in Dice coefficient,F1 score and Hausdorff distance on different types of test sets.The model has strong generalization and has important application prospects.

关 键 词：组织病理图像 腺体分割 注意力机制 双路径特征 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]