基于深度学习的超声图像去噪方法研究  

Ultrasound image denoising based on deep learning

作　　者：代哲 杨艳[1] DAI Zhe;YANG Yan(School of Physical Science and Technology,Wuhan University,Wuhan 430072,China)

机构地区：[1]武汉大学物理科学与技术学院,武汉430072

出　　处：《信息技术》2020年第9期57-60,共4页Information Technology

基　　金：国家重点基础研究发展计划(973)(CB707900)。

摘　　要：超声图像以低成本、时效性、无辐射的优点被广泛应用于医疗诊断,但是严重的散斑噪声对图像细节信息造成干扰。为了有效去除超声图像中的散斑噪声,同时保留超声图像的特征信息,文中基于DnCNN改进更匹配于散斑噪声的模型结构,通过对仿真超声图像训练来去噪,并与其他方法做定量对比。通过实验结果表明,改进的深度学习模型运用在超声图像去噪中具有良好的效果。Ultrasound images are widely used in medical diagnosis with the advantages of low cost,timeliness,and no radiation,but severe speckle noise causes interference with image details.In order to effectively remove the speckle noise in the ultrasound image while retaining the characteristic information,the model structure is improved based on DnCNN to match the speckle noise better,and the simulated ultrasound image is trained to denoise,quantitative comparison is made with other methods.The experimental results show that the improved deep learning model has a good effect in denoising ultrasound images.

关 键 词：超声图像 散斑噪声 深度学习 去噪 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]