Pulmonary parenchymal segmentation based on CT images of lungs in a highland population

Abstract

Abstract In response to the more complex characterization of lung CT image data in the highland population, the problem of left and right lung adhesions, gross contours and blurred borders that are very likely to arise during lung parenchymal segmentation, we propose a new network structure based on U-Net for lung parenchyma segmentation in highland population. First, we introduce residual block, which replaced ordinary convolution with residual convolution in the encoding stage to improve the speed of network convergence and accuracy; Then, the Augmented Attention Module (AAM) is introduced in the skip connection part to enhance the feature of the lung parenchyma contour information; Finally, considering the processing efficiency of the network, we reduce the depth of the network model to 4 layers. After that, we verify the effectiveness of our method on the public dataset LUNA16, and the segmentation results show that the selected evaluation indicators are improved to varying degrees. This shows that our proposed method has a good effect on the segmentation of lung parenchyma. Finally, applying our method to lung CT image segmentation in a highland population gives some advantages over other existing methods.