A COVID-19 medical image Segmentation method based on U-NET

ABSTRACT

COVID-19 covers many countries around the world, Chest X-ray is the mainstream method for identifying COVID-19 infection. Traditional Chest X-ray detection requires professional medical personnel, which is time-consuming and laborious. Accurate medical segmentation can be used as an auxiliary means to detect COVID-19, which not only greatly reduces the cost and time, but also greatly improves the applicability. With the rapid development of deep learning, a network model based on U-NET has been proposed and widely used in medical image segmentation in recent years. However, in U-NET network, multiple convolutional pooling operations cause the loss of image spatial information features, and each channel of output features is treated equally, thus lacking flexibility in processing different information. Therefore, in this paper, we add gray bars to the samples to avoid the distortion and feature reduction caused by clipping and resize. the U-NET model architecture is taken as the main body to improve the weight of each channel in the U-NET encoding layer to increase the semantic information of the feature map and improve the segmentation accuracy of the network. In the decoding channel, feature information is restored by up-sampling. Finally, convolution and Softmax function are used to obtain the predictive segmentation image with the same size as the original image. The results show that the improved model has better performance than the traditional U-NET network.