MRAU-net网络下的X光胸片肺野分割算法

为了解决U-net网络进行X光胸片肺野分割时，受限于特征提取能力不足导致分割结果不精确的问题，提出一种多尺度残差注意力U型网络(MRAU-net)模型.利用多尺度信息融合(MIF)模块，改善网络结构，增加对多尺度信息的获取；利用通道和空间双注意力(CSDA)模块，解决网络在有限算力下的信息过载问题.同时，对残差模块进行改进，并与U-net网络进行深度结合，提升网络的学习稳定性，缓解梯度消失和过拟合现象.实验结果表明：文中方法具有优秀的X光胸片肺野分割能力，能获得更精确的分割结果.

Lung Field Segmentation Algorithm of X-Ray Chest Film Based on MRAU-Net Network

In order to solve the problem of imprecise segmentation results caused by insufficient feature extraction ability when U-net network is used to segment lung fields in X-ray chest films, a multi-scale residual attention U-net(MRAU-net)model is proposed. The multi-scale information fusion(MIF)module is used to improve the network structure and increase the acquisition of multi-scale information. Using channel and space dual attention(CSDA)module, the problem of information overload in the network under limited computing power is solved. At the same time, the residual module is improved and deeply combined with the U-net network to improve the learning stability of the network and alleviate the phenomenon of gradient disappearance and over fitting. The experimental results show that the proposed method has excellent segmentation ability of lung field in X-ray chest film, and can obtain more accurate segmentation results.