基于空洞卷积神经网络的医学超声图像去噪  

Medical ultrasound image denoising based on void convolutional neural network

作　　者：邵党国[1] 朱彧麟 马磊[1] 徐慧[2] SHAO Dangguo;ZHU Yulin;MA Lei;XU Hui(Faculty of Information Engineering and Automation,Kunming University of Science and Technology,Kunming 650500,China;First Affiliated Hospital of Kunming Medical University,Kunming 650000,China)

机构地区：[1]昆明理工大学信息工程与自动化学院,云南昆明650500 [2]昆明医科大学第一附属医院,云南昆明650000

出　　处：《现代电子技术》2023年第13期55-61,共7页Modern Electronics Technique

基　　金：国家自然科学基金项目(62266025)资助。

摘　　要：散斑严重影响医学超声图像的质量,从而导致临床诊断和图像处理困难。为解决上述问题,提出一种基于空洞卷积神经网络的散斑减小方法。不同于其他的散斑抑制方法,文中通过建立超声图像和散斑图像之间的非线性映射,同时结合跳跃连接使网络不会发生梯度爆炸或消失,从而具有较好的反向传播能力。该方法可以通过超声图像预测散斑,再由超声图像与预测的散斑图像相减从而得到干净的超声图像。该方法相较于散斑减少各向异性扩散(SRAD)和细节保持各向异性扩散(DPAD),具有更好的散斑减少和结构保存性能,在对比的量化数据上提升巨大。实验结果表明,该方法在超声医学图像上具有较好的性能和鲁棒性。Speckle seriously affects the quality of medical ultrasound images,resulting in difficulties in clinical diagnosis and image processing.In order to solve the above problem,a speckle reduction method based on void convolutional neural network is proposed.Different from other speckle reduction methods,the nonlinear mapping between ultrasound images and speckle images is established,and the jump connection is combined to prevent gradient explosion or disappearance of the network,so that the method has good backpropagation ability.This method can predict speckle by ultrasound image,and then subtract the ultrasound images from the predicted speckle images to obtain the clean ultrasound images.In comparison with speckle reduction anisotropic diffusion(SRAD)and detail preservation anisotropic diffusion(DPAD),this method has better speckle reduction and structure preservation performance,and can greatly improve the comparative quantitative data.The experimental results show this method has good performance and robustness in ultrasound medical images.

关 键 词：医学超声图像 图像去噪 空洞卷积 散斑减少 卷积神经网络 网络模型 跳跃连接 

分 类 号：TN911.73-34[电子电信—通信与信息系统] TP751.1[电子电信—信息与通信工程]