Diagnosis of COVID-19 in X-ray Images using Deep Neural Networks

Abstract

The global COVID-19 pandemic has presented unprecedented challenges, notably the limited availability of test kits, hindering timely and accurate disease diagnosis. Rapid identification of pneumonia, a common COVID-19 consequence, is crucial for effective management. This study focuses on COVID-19 classification from Chest X-ray images, employing an innovative approach: adapting the Xception model into a U-Net architecture via the Segmentation_Models package. Leveraging deep learning and image segmentation, the U-Net architecture, a CNN variant, proves ideal for this task, particularly after tailoring its output layer for classification. By utilizing the Xception model, we aim to enhance COVID-19 classification accuracy and efficiency. The results demonstrate promising autonomous identification of COVID-19 cases, offering valuable support to healthcare professionals. The fusion of medical imaging data with advanced neural network architectures highlights avenues for improving diagnostic accuracy during the pandemic. Notably, precision, recall, and F1 scores for each class are reported: Normal (Precision = 0.98, Recall = 0.9608, F1 Score = 0.9704), Pneumonia (Precision = 0.9579, Recall = 0.9579, F1 Score = 0.9579), and COVID-19 (Precision = 0.96, Recall = 0.9796, F1 Score = 0.9698). These findings underscore the effectiveness of our approach in accurately classifying COVID-19 cases from chest X-ray images, offering promising avenues for enhancing diagnostic capabilities during the pandemic.