CPNet: Covid19 and Viral Pneumonia Automated Detection Based on Convolutional Neural Network

Novel coronavirus (covid19) has affected the entire world since its outbreak in late 2019. Millions of people have died, and hundreds of millions of people got infected with the covid19 around the globe. For early diagnosis of covid19, an automated detection system is much needed that is accurate in detecting the virus. This research proposes a CNN based solution for distinguishing covid19 from viral pneumonia through chest X-ray images. The dataset used for this research has binary classes with 1281 covid19 images and 1655 viral pneumonia images. The proposed CNN method is named CPNet, which is based on the VGG19 backbone network for the considered problem. We replaced the FC layers of the model with our custom convolution, max pooling, fully connected and classification layers for discriminating covid19 and viral pneumonia. We introduced small filter of 3× 3 convolution to learn the features more efficiently without missing the important information. The proposed CPNet performance is compared to existing CNN models in terms of various considered performance metrics. The results demonstrate that the proposed CPNet outperforms other CNN models with 99.15% prediction accuracy, 98.44% recall, 99.60% precision and 99.02% Fl-score for the considered task of covid19 and viral pneumonia classification.