基于边缘检测的乳腺癌细胞核分割预测图融合算法  

Edge detection-based predictive map fusion algorithm for breast cancer cell nuclei segmentation

作　　者：闫伟鑫 吕晓琪[1,2] 谷宇 YAN Weixin;LYU Xiaoqi;GU Yu(Inner Mongolia Key Laboratory of Pattern Recognition and Intelligent Image Progressing,Information Engineering School,Inner Mongolia University of Science and Technology,Baotou 014010,China;School of Information Engineering,Inner Mongolia University of Technology,Hohhot 010051,China)

机构地区：[1]内蒙古科技大学信息工程学院内蒙古自治区模式识别与智能处理重点实验室,内蒙古包头014010 [2]内蒙古工业大学信息工程学院,内蒙古呼和浩特010051

出　　处：《内蒙古科技大学学报》2023年第1期41-45,70,共6页Journal of Inner Mongolia University of Science and Technology

基　　金：国家自然科学基金资助项目(62001255,61771266,61841204);内蒙古自治区科技计划资助项目(2019GG138);中央引导地方科技发展资金资助项目(2021ZY0004);内蒙古自治区自然科学资助项目(2019MS06003,2015MS0604);内蒙古自治区高等学校科学研究基金资助项目(NJZY145);教育部“春晖计划”合作科研资助项目:“基于Hessian选择性增强滤波及粒子群参数优化支持向量机的影像大数据肺癌辅助检测诊断技术研究”(教外司留[2019]1383号).

摘　　要：目前针对病理学图像中的乳腺癌细胞核分割研究中,乳腺癌细胞核分割结果中存在边缘欠分割,过分割等问题.由此提出了一种基于边缘检测的分割预测图融合算法(UDE-M),该算法主要为两个步骤,一是通过U-Net,DeepLab V3对数据集进行训练获取分割预测图,二是通过算法中的逻辑判别规则进行融合获取最终的分割结果.研究结果表明:UDE-M算法Accuracy值达到了92.5%,Precision值达到了77.3%,Recall达到了83.8%,F1分数值达到了80.4%,相比于U-Net和DeepLab V3以及文中组成的FDE-M模型和SDE-M模型,能够更加有效对乳腺癌细胞核进行分割.In the current research of breast cancer cell nuclear segmentation in pathological images,there are problems such as edge under-segmentation and over-segmentation in breast cancer cell nuclei segmentation results.Therefore,a segmentation prediction map fusion algorithm based on edge detection(UDE-M)was proposed.This algorithm mainly consists of two steps:one is to obtain the segmentation prediction map by training the dataset through U-Net and DeepLab V3,and the other is to obtain the final segmentation result by fusing the logical discrimination rules in the algorithm.The research results show that the accuracy value of UDE-M algorithm reaches 92.5%,the precision value reaches 77.3%,the recall value reaches 83.8%,and the F1 score reaches 80.4%.Compared with U-Net and DeepLab V3,as well as the FDE-M model and SDE-M model composed herein,it can segment breast cancer nuclei more effectively.

关 键 词：乳腺癌 边缘检测 细胞核分割 预测图融合 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]