The usefulness of the modified deep convolutional neural network model in improving the detection of COVID-19 on chest X-ray images

Abstract

Background: The COVID-19 pandemic has rapidly spread worldwide, leading to a global health crisis. Although the real-time polymerase chain reaction (RT-PCR) test is highly specific and sensitive in detecting COVID-19, chest X-rays have emerged as an optional diagnostic tool for COVID-19-induced lung lesions. Artificial intelligence (AI), particularly deep learning, is a rapidly evolving field with significant potential in medical image analysis, including the quick detection of COVID-19 to improve accuracy. Objectives: This study aims to enhance the accuracy of COVID-19 image detection on chest X-ray images by modifying the deep convolutional neural network. Materials and methods: We conducted lung segmentation and COVID-19 image classification experiments using a dataset of chest X-rays. The U-net algorithm was utilized for lung segmentation of COVID-19 and non-COVID-19 images. We developed a Modified Deep Convolutional Neural Network (MD-CNN) to classify the two image classes. The MD-CNN model was compared with two other models, ResNet and AlexNet, and evaluated for accuracy, sensitivity (recall), specificity, positive predictive value (precision), F1-score, and area under the curve (AUC). Results: Our experimental results demonstrate that the MD-CNN model achieved an accuracy of 97.95%, outperforming ResNet and AlexNet, which achieved 90.25% and 78.95%, respectively. The MD-CNN model also exhibited better sensitivity, F1-score, and AUC than the other models, while its specificity and precision were comparable to those of the ResNet model. Conclusion: The proposed MD-CNN model demonstrates significant potential for high accuracy in COVID-19 image detection compared to ResNet and AlexNet. It can serve as a useful tool for radiologists in the COVID-19 screening process, potentially reducing the workload, and improving the efficiency of COVID-19 diagnosis.