一种辅助新型冠状病毒肺炎检测的肺实质分割算法

新型冠状病毒肺炎给人类健康及社会经济造成了巨大的负面影响，而X光胸片中的肺实质提取成为新型冠状病毒肺炎诊断过程中的关键环节。在U-Net的基础上，提出一种结合编解码模式的肺实质分割算法。应用特征融合思想，构建A形特征融合模块，充分学习深层特征的语义信息。引入注意力机制，在深层卷积神经网络中加入密集空洞卷积模块和残差多核池化模块，扩大卷积感受野并提取上下文特征信息。通过改进可变形卷积和分割损失函数，提升网络模型的泛化能力和鲁棒性。实验结果表明，该算法的分割准确度、Dice系数、敏感度、Jaccard指数分别为98.16%、98.32%、98.13%、98.54%，能够实现X光胸片中肺实质部位的有效分割。

A Pulmonary Parenchyma Segmentation Algorithm Aiding in COVID-19 Detection

As a rapidly evolving pandemic,COVID-19 has caused severe health and economic impact.In the diagnosis of COVID-19,the extraction of pulmonary parenchyma in chest X-ray images plays an important role.A U-Net-based pulmonary parenchyma segmentation algorithm using the encoding and decoding mode is proposed.The algorithm applies the idea of feature fusion to the construction of an A-Block to fully learn the semantic information of deep features.The attention mechanism is introduced into the deep convolutional neural network by adding a Dense Atrous Convolution(DAC) module and a Residual Multi-kernel Pooling(RMP) module in order to extend the receptive field of the convolution and to extract the contextual feature information.By improving the deformable convolution and the segmentation loss function,the generalization ability and the robustness of the network model are enhanced.Experimental results show that the segmentation accuracy,Dice coefficient,sensitivity and Jaccard index of this algorithm are 98.16%,98.32%,98.13% and 98.54% respectively.The algorithm can effectively implement pulmonary parenchyma segmentation.