SF-CXRnet: A Lightweight Dual-Channel CNN for Enhanced Detection of COVID-19 and Pneumonia from Chest X-Ray Images

Pneumonia and COVID-19 rank among the most severe and prevalent respiratory diseases of our time. Recently, there has been a resurgence in cases, with patients often testing negative for one condition while being positive for the other, creating challenges in accurate diagnosis. However, limited research has been conducted for joint diagnosis. This research presents a highly efficient and lightweight Convolutional Neural Network, named SF-CXRnet, meticulously designed with minimal parameters to enhance computational efficiency, offering an effective method for identifying COVID-19 and pneumonia cases through chest X-ray images. Additionally, The proposed model effectively classifies lung X-rays into three categories: Normal, COVID-19, and Pneumonia, offering a powerful tool for accurate diagnosis. In this study, the proposed model crafted and trained using the COVID-19 radiography dataset offered remarkable outcomes. Despite utilizing just 1.54 million parameters, the proposed SF-CXRnet model achieved an outstanding 98.21% test accuracy, outperforming pre-trained models like VGG-19, ResNet-50, and MobileNetV2. Prioritizing disease detection, SF-CXRnet demonstrates exceptional performance with a 98% recall across all three classes, surpassing other models used in this research. This proposed model’s remarkable accuracy in distinguishing COVID-19 from pneumonia, along with its highest recall score compared to state-of-the-art, underscores its potential to enable timely medical interventions. The study advances COVID-19 and Pneumonia detection, paving the way for enhanced automated solutions, ultimately improving accessibility and precision in lung disease diagnosis.