Deep Learning-based Detection for COVID-19 from Chest CT using Weak Label

Chuansheng Zheng; Xianbo Deng; Qing Fu; Qiang Zhou; Jiapei Feng; Hui Ma; Wenyu Liu; Xinggang Wang

The Proposed DeCoVNet We proposed a 3D deep convolutional neural Network to Detect COVID-19 (DeCoVNet) from CT volumes. As shown in Fig. 1 , DeCoVNet took a CT volume and its 3D lung mask as input. The 3D lung mask was generated by a pre-trained UNet [19] . DeCoVNet was divided into three stages for a clear illustration in Table. 1. The first stage was the network stem, which consisted of a vanilla 3D convolution with a kernel size of 5 × 7 × 7, a batchnorm layer and a pooling layer. The second stage was composed of two 3D residual blocks (ResBlocks). In each ResBlock, a 3D feature map was passed into both a 3D convolution with a batchnorm layer and a shortcut connection containing a 3D convolution that was omitted in Fig. 1 for dimension alignment. The resulted feature maps were added in an element-wise manner. The third stage was a progressive classifier (ProClf), which mainly contained three 3D convolution layers and a fully-connected (FC) layer with the softmax activation function. ProClf progressively abstracts the information in the CT volumes by 3D max-pooling and finally directly output the probabilities of being COVID-positive and COVID-negative. The 3D lung mask of an input chest CT volume helped to reduce background information and better detect COVID-19. Detecting the 3D lung mask was a well-studied issue. In this study, we trained a simple 2D UNet using the CT images in our training set. To obtain the ground-truth lung masks, we 4 . CC-BY-NC-ND 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.