中北大学学报（自然科学版）2025，Vol.46Issue(4)：411-421,11.DOI:10.62756/jnuc.issn.1673-3193.2024.07.0016
基于边缘融合和多尺度特征增强的结肠癌腺体分割网络
Colon Cancer Gland Segmentation Network Based on Edge Fusion and Multi Scale Feature Enhancement
摘要
Abstract
In histopathological sections,the morphological characteristics of glands are an important basis for diagnosing colon cancer,and accurate gland segmentation can help doctors classify cancer.However,due to the diverse shapes and blurred edges of malignant glands,existing segmentation methods rarely directly focus on edge features,resulting in unsatisfactory segmentation results in malignant cases.In response to the limitations of existing methods in malignant gland edge segmentation,this study proposed an enhanced network(BFMSE Net)that integrated edge information and multi-scale features,aiming to improve the segmentation accuracy of colon cancer glands.This network added edge fusion module(BFM)and multi-scale enhancement module(MSEM)on the basis of U-net.The edge fusion module not only extracted and enhanced edge features,but also significantly improved the segmentation ability of edge blur and adhesive glands through advanced feature fusion strategies;The multi-scale enhancement module adaptively captured the information of the best receptive field in the region of interest to handle complex situations with significant changes in glandular size and shape.The experimental results indicate that the network can detect clear glandular contours.On the GlaS Challenge dataset and CRAG dataset,the shape similarity metrics are 74.586 and 61.572,respectively,which improves by 10.17%and 85.19%compared to state-of-the-art methods.In addition,the F1 scores of BFMSE Net reaches 92.3%and 83.9%,and the Dice coefficients reaches 91.4%and 85.3%,both of which are better than existing models.The proposed BFMSE Net performs stably in two publicly available datasets and has good segmentation performance for malignant glands with blurred edges and glands with edge adhesions.Its shape similarity is much higher than that of existing models,which can assist doctors in better diagnos-ing colon cancer.关键词
多尺度特征/结肠癌/腺体分割/医学图像处理Key words
multi-scale feature/colon cancer/gland segmentation/medical image processing分类
计算机与自动化引用本文复制引用
柴锐,罗钰冰,秦品乐,翟双姣,闫俊义..基于边缘融合和多尺度特征增强的结肠癌腺体分割网络[J].中北大学学报（自然科学版）,2025,46(4):411-421,11.基金项目
山西省科技重大专项计划"揭榜挂帅"项目(2021010:0101018) （2021010:0101018）
山西省基础研究计划(自由探索类)项目(20210302123033) （自由探索类）
