XcepCovidNet: deep neural networks-based COVID-19 diagnosis

Abstract

Coronavirus (also known as COVID-19) is an extremely contagious disease spreading around the globe. An efficient and fast diagnostic method must be designed to identify COVID-19 patients. There are several methods for identification and monitoring of the disease, namely radiological imaging of the patient's chest and polymerase chain reaction (RT-PCR) test. Recent investigations have shown that radiological images are used to observe the effect of COVID on the lungs. Deep Learning is proven effective for image detection and classification in many applications. The majority of existing COVID-19 architectures detect irrelevant features for decision-making. In this paper, a novel network called XcepCovidNet is proposed for feature detection of chest X-rays. It employs transfer learning using hyperparameter-tuning to account for the inadequacies of the training dataset. The proposed model is found superior to pre-trained models such as VGG-19, ResNet-50, DenseNet-201, Xception, and DarkNet-19, in terms of different performance metrics. It is an automated, fast, reliable, and precise COVID-19 detection system for initial screening and diagnosing infected individuals. The obtained results indicate that XcepCovidNet yielded an accuracy of 98.67% and 93.66% for binary and four-class classification, respectively. A two-step verification is performed to validate the proposed model using different models of explainable artificial intelligence, i.e., LIME and occlusion sensitivity. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2024.