基于多尺度空洞残差的U-Net模型用于乳腺肿块分割  

U-Net Model Based on Multi-scale Dilated Residual for Breast Mass Segmentation

作　　者：李净阳 宁春玉[1] LI Jingyang;NING Chunyu(School of Life Science and Technology,Changchun University of Science and Technology,Changchun 130022)

机构地区：[1]长春理工大学生命科学技术学院,长春130022

出　　处：《长春理工大学学报（自然科学版）》2023年第1期108-113,共6页Journal of Changchun University of Science and Technology(Natural Science Edition)

基　　金：吉林省科技发展计划项目(20200404219YY)。

摘　　要：乳腺肿块是乳腺癌常见征象,实现乳腺肿块的自动分割可极大减轻医生工作负担,有利于乳腺癌的早诊断早治疗。为进一步提升分割性能,提出了一种基于多尺度空洞卷积残差的网络模型。在U-Net经典结构基础上,将多尺度空洞卷积特征提取模块嵌入到残差模块,并通过添加批量归一化层、引入乘以权重的二元交叉熵损失函数的方式,在保证空间信息的前提下,增大感知范围,增强模型对乳腺肿块区域特征的提取能力。提出的分割模型在CBIS-DDSM数据库上进行验证,得到的Dice系数为82.93%,灵敏度为84.72%,较U-Net模型分别提高了0.75%、1.36%。Breast mass is a common sign of breast cancer.The automatic segmentation of breast mass can reduce the workload of doctors,which is conducive to the early diagnosis and treatment of breast cancer.In order to further improve the segmentation performance,a network model based on multi-scale dilated convolution residual was proposed.Based on the classical structure of U-Net,the feature extraction module based on multi-scale dilated convolution was embedded into the residual module,the batch normalization layer was added,and the binary cross entropy loss function multiplied by weight was introduced into the proposed network model.Therefore,the model increased the sensing range and enhanced the ability to extract the regional features of breast mass on the premise of ensuring spatial information.The proposed segmentation model was verified on CBIS-DDSM database,and the Dice coefficient and sensitivity obtained were 82.93% and 84.72%,which were increased by 0.75% and 1.36% compared with U-Net model,respectively.

关 键 词：乳腺肿块 分割 U-Net 空洞卷积 残差模块 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]