Application of 3D-CNN Model in the Diagnosis of COVID-19

Abstract

In recent years, COVID-19 has become the largest epidemic in the world and has had a great impact on the world. The diagnosis of COVID-19 also heavily relies on lung CT. This study presents a COVID-19 CT scan categorization computer-aided diagnostic (CAD) system. The performances of two 3D-CNNmodels with the similar structures are compared. The data used from MosMedData is the Chest CT Scans with COVID-19 Related Findings Dataset. The best segmentation technique for separating the chest tissue from the rest of the CT picture is threshold segmentation. If a volume has several slice groups, the characteristics from each group are extracted and added together to create the eigenvector for the whole CT scanning volume. The combined eigenvectors are then further categorized using a straightforward multi-layer perceptron (MLP) network. One model generated a test set with an accuracy of 78.8%, while the other model generated a test set with an accuracy of 79.8%. This study demonstrates the particular potential of 3D-CNN for COVID-19 diagnosis.