基于双路径网络和注意力机制的胰腺图像分割  

Research on pancreatic image segmentation based on dual path network and attention mechanism

作　　者：王嘉瑶 吕晓琪 谷宇 张明 WANG Jiayao;LÜ Xiaoqi;GU Yu;ZHANG Ming(Key Laboratory of Pattern Recognition and Intelligent Image Processing,School of Information Engineering,Inner Mongolia University of Science and Technology,Baotou 014010,China;College of Information Engineering,Inner Mongolia University of Technology,Hohhot 010051,China;School of Information Science and Technology,Dalian Maritime University,Dalian 116026,China)

机构地区：[1]内蒙古科技大学信息工程学院模式识别与智能图像处理重点实验室,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 [3]大连海事大学信息科学技术学院,辽宁大连116026

出　　处：《现代电子技术》2022年第7期47-52,共6页Modern Electronics Technique

基　　金：国家自然科学基金项目(61771266);国家自然科学基金项目(62001255);国家自然科学基金项目(61841204);内蒙古自治区科技计划项目(2019GG138);内蒙古自治区自然科学基金项目(2019MS06003,2015MS0604);内蒙古自治区高等学校科学研究项目(NJZY145,NJZY18150);教育部“春晖计划”合作科研项目: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：针对胰腺CT图像因类别不平衡、背景分散和非刚性几何特征的特性而分割精度不高的问题,提出一种基于Unet网络,采用改进的双路径网络和通道域注意力机制的图像分割模型。首先采用空洞卷积(DilatedConv)代替传统卷积优化双路径网络,扩大感受野,让每个卷积输出都包含较大范围的信息;再将双路径网络的编码器结构引入Unet网络,加强特征的重复利用和新特征的不断探索,获取更多图像细节信息;在此改进网络基础上引入通道域注意力机制,聚焦分割重要部位,提高分割准确率。实验数据表明,设计的网络最高Dice相似系数(DSC)达到了89.81%,最低DSC为72.33%,平均DSC为(85.82±4.73)%。结果表明该模型具有较高的准确率,准确分割胰腺是计算机辅助诊断的重要前提,其研究意义重大。In view of the low segmentation accuracy of pancreatic CT images because of the imbalance of pancreatic categories,scattered background and non-rigid geometric features,a new image segmentation model based on Unet is proposed.The model combines the improved dual path network(DPN)and the channel domain attention mechanism.DilatedConv(Dilated Convolution)is used to replace the traditional convolution to optimize DPN,so as to expand the receptive field and make each convolution output contain a large range of information.And then,the encoder structure of the DPN is introduced into the Unet to enhance the feature reuse and the continuous exploration of new features,so as to obtain more detailed image information.On the basis of the improved network,the channel domain attention mechanism is introduced to focus on the segmentation of important parts,so as to improve the segmentation accuracy.The experimental data show that the highest dice similarity coefficient(DSC)of the proposed network achieves 89.81%,its lowest DSC is 72.33%and its average DSC is(85.82±4.73)%.Therefore,the proposed model has high accuracy rate.Accurate segmentation of the pancreas is an important prerequisite for computer-aided(CAD)diagnosis,so this research is of great significance.

关 键 词：图像处理 胰腺CT图像 空洞卷积 Unet网络 双路径网络 通道域注意力机制 计算机辅助诊断 

分 类 号：TN911.73-34[电子电信—通信与信息系统] TP391.4[电子电信—信息与通信工程]