Automated Detection of COVID-19 with Chest CT Scans using 3D Deep Learning model

ABSTRACT

COVID-19 is breaking out and spreading globally, posing a severe threat to public health and economies worldwide due to its highly transmissible and pathogenic nature. Early, accurate and rapid diagnosis of COVID-19 can effectively stop the spread of the COVID-19 virus. Automatic diagnostic models based on deep learning can detect COVID-19 quickly and accurately. This paper uses a three-dimensional Convolutional Neural Network (3D CNN) to build a COVID-19 diagnostic prediction model for COVID-19 detection. All 192 sets of chest Computed Tomography(CT) data collected are used for this study, including 96 sets of confirmed COVID-19 patients and 96 sets of CT scans of normal human lungs. 5-fold cross-validation is used to train and validate the model. 154 data sets are used to train the model, and 38 sets are used for testing. All experimental data are segmented using a pre-trained SP-V-Net to obtain 3D lung masks fed into 3D CNN for training and validation of the prediction model. In addition, to verify the accuracy of the model predictions and provide interpretability for medical diagnosis, we visualize the experimental results using Class Activation Maps(CAM) to localize the predicted disease regions. The results from several experiments show that the accuracy of our prediction model is 0.911, the Area Under Curve (AUC) 0.976, for no-COVID-19(Precision, 0.902, Recall 0.911, F1-Score 0.900), COVID-19 (Precision, 0.932, Recall 0.911, F1-Score 0.902). The experimental results show that our established diagnostic model can help physicians make a rapid and accurate diagnosis of COVID-19 in response to the spread of COVID-19. © 2021 IEEE.