AM-RESNET50 Method for CT Image Diagnosis of COVID-19

Abstract

At the beginning of 2020, coronavirus disease (covid-19) spread all over the world, making the world face a survival and health crisis. Automatic detection of pulmonary infection through computed tomography (CT) images provides great potential for strengthening the traditional health care strategy to deal with covid-19. At present, the use of artificial intelligence technology for image classification and lesion segmentation of COVID-19CT image has become a widely concerned content in medical image analysis. Segmenting the infected area from CT image faces several challenges, including high variation of infection characteristics, low-intensity comparison between infection and normal tissue and so on. Based on the in-depth analysis of covid-19 CT image features, this paper adds a mixed attention mechanism module to the RESNETneural network model, including channel attention mechanism and spatial attention mechanism. The combination of channel attention mechanism and spatial attention mechanism makes the backbone network have the ability to pay attention to more important local features from global features, making the model more sensitive to covid CT images. In terms of implementation efficiency, the convolution layer of the model is improved with smaller convolution kernel, and the loss function is modified to adjust the data training model, so as to realize the more accurate and efficient automatic recognition of covid-19 CT image.