3D dilated and residual convolutional neural network for COVID-19 detection from the chest computed tomography

Chest Computed Tomography (CT) is regarded as one of the most effective tools in diagnosing COVID-19 due to its high sensitivity and ease of use. However, analysis of CT images may be time-consuming for the clinicians, which highly influence their performance. Artificial-intelligence-based methods can help automating the process of interpreting chest CT images and diagnosis of COVID-19 in suspicious patients. In this paper, we propose a 3D deep convolutional neural network for classifying chest CT images into COVID-19-infected and normal classes. Dilated convolution and residual connections are employed to increase the model's performance by enlarging the receptive field of the kernels and direct propagation of the information. The accuracy, precision, sensitivity, specificity, and F1-score achieved by our model are 0.99, 0.98, 1.0, 0.979, and 0.99, respectively. The high sensitivity value of the model demonstrates its efficiency in detecting/identifying all the infected patients correctly, which allows early quarantine and the start of the treatment process.