计算机工程

• •    

一种辅助新冠肺炎检测的肺实质分割算法

  

  • 发布日期:2021-01-05

A lung parenchyma segmentation algorithm for the COVID-19 detection

  • Published:2021-01-05

摘要: 新冠肺炎给人民生命健康及社会经济带来了巨大冲击,X 光胸片中的肺实质部分对肺炎诊断具有重要的指导意义, 因此肺实质分割成为临床诊断肺炎的迫切需求。本文改进了一种编解码模式的图像分割算法,该算法主要针对组织结构复杂 的人体肺部结构分割,采用了 A 形特征融合模块(A-block),有效减少了数据传输过程中的信息丢失。针对分割目标形变较 大问题,向网络中加入了改进的可变形卷积。同时,提出了一种类距惩罚分割损失函数,使二分类预测结果更倾向于两极分 化。通过实验表明,本文算法的准确率、相似度系数、敏感度、Jaccard 指数分别为 98.16%、98.32%、98.13%、98.54%,均 优于其他对比算法,能有效实现肺实质分割任务。

Abstract: The COVID-19 has brought great impact on people's health and social economy. The lung parenchyma in X-ray chest film has important guiding significance for the diagnosis of the COVID-19. In this paper, an image segmentation algorithm based on encoding and decoding mode is improved. The algorithm is mainly aimed at the segmentation of human lung structure with complex tissue structure. A-block is used to reduce the loss of information in the process of data transmission. To solve the problem of large deformation of segmented objects, an improved deformable convolution is added to the network. At the same time, a class distance penalty partition loss function is proposed to make the binary classification prediction results more polarized. The experimental results show that the accuracy, similarity coefficient, sensitivity and Jaccard index of this algorithm are 98.16%, 98.32%, 98.13% and 98.54% respectively, which are better than other comparative algorithms, and can effectively achieve the task of lung parenchyma segmentation.