Chest X-ray Image Segmentation Based on Improved Unet

In this paper, we use the Unet network to achieve automatic segmentation of chest X-ray images and develop a chest X-ray image segmentation application. To further improve segmentation performance, we have improved the original network structure by changing the network depth, adding attention mechanisms and changing the encoder convolution module. Our improved methods achieved a Dice coefficient of 0.973 and an accuracy of 0.983 on the Kaggle Chest X-ray Masks and Labels dataset.