基于多尺度偶数卷积注意力U-Net的医学图像分割  

Medical Image Segmentation Based on Multiscale Even Convolution Attention U-Net

作　　者：杨真真[1,2] 孙雪 邵静 杨永鹏[1,3] YANG Zhenzhen;SUN Xue;SHAO Jing;YANG Yongpeng(Key Laboratory of Ministry of Education in Broadband Wireless Communication and Sensor Network Technology,Nanjing University of Posts and Telecommunications,Nanjing,Jiangsu 210003,China;School of Science,Nanjing University of Posts and Telecommunications,Nanjing,Jiangsu 210023,China;School of Network and Communication,Nanjing Vocational College of Information Technology,Nanjing,Jiangsu 210023,China)

机构地区：[1]南京邮电大学宽带无线通信与传感网技术教育部重点实验室,江苏南京210003 [2]南京邮电大学理学院,江苏南京210023 [3]南京信息职业技术学院网络与通信学院,江苏南京210023

出　　处：《信号处理》2022年第9期1912-1921,共10页Journal of Signal Processing

基　　金：国家自然科学基金(61501251,62171232);南京邮电大学宽带无线通信与传感网技术教育部重点实验室开放研究基金(JZNY202113);南京邮电大学科研项目(NY220207)。

摘　　要：为了提高U-Net网络性能的同时尽可能减少额外计算量,本文提出了一种新的多尺度偶数卷积注意力UNet(Multiscale Even Convolution Attention U-Net,MECAU-Net)网络。该网络在编码端采用2×2偶数卷积代替3×3卷积进行特征提取,并借鉴多尺度思想,采用4×4偶数卷积将得到的信息直接传递给主干部分,以获取更全面的图像信息并减少额外计算开销,同时还采用对称填充解决偶数卷积提取信息过程中产生的偏移问题。此外,在2×2偶数卷积模块后加入卷积注意力模块,结合空间和通道注意力,在提取更丰富的信息的同时几乎不增加额外开销。最后,在两个医学图像数据集上进行仿真实验,实验结果表明提出的MECAU-Net网络相对于U-Net在稍微增加计算成本的情况下,分割性能得到了较大的提升,并比其他对比网络取得更好的分割性能的同时还降低了参数量。In order to improve the performance of U-Net and reduce the additional computational complexity as much as possible,a new multiscale even convolution attention U-Net(MECAU-Net)network is proposed in this paper.This network uses 2×2 even convolution instead of 3×3 convolution for feature extraction at the encoder.And inspired by the idea of the multiscale idea,4×4 even convolution is used to directly transfer the obtained information to the backbone,so as to obtain more comprehensive image information and reduce additional computational cost.At the same time,the symmetric padding is used to solve the shift problem in the process of extracting information from even convolution kernels.In addition,the convolution block attention module is added to combine the spatial and channel attention modules after the 2×2 even convolution module,which can extract richer information without adding additional computational complexity.Finally,simulation experiments are carried out on two medical image datasets.The experimental results show that our proposed MECAU-Net network greatly improves the segmentation performance with slight additional computational cost.In addition,it achieves better segmentation performance than other comparison networks and reduces the amount of parameters.

关 键 词：卷积神经网络 医学图像分割 偶数卷积 多尺度 注意力模块 

分 类 号：TN911.73[电子电信—通信与信息系统]