ResNet and ResNeSt-Based Deep-Learning Models for Accurate COVID-19 Detection from Chest X-ray Radiographs

The emergence of the COVID-19 pandemic has had a notable influence on the health of populations worldwide and has also affected numerous socioeconomic factors globally. Early identification and timely management of COVID-19 are essential to prevent the proliferation of the disease and protect people’s lives. Recent studies indicate that utilizing deep-learning models gives a feasible solution for the diagnosis of COVID-19 by employing chest X-ray imaging. Combining advanced deep-learning models with radiological imaging can enhance the accuracy of COVID-19 detection and address the shortage of specialized physicians in remote areas. In this study, four distinct deep-learning models are used to detect COVID-19 from chest X-ray images: ResNet50, ResNet101, ResNeSt50, and ResNeSt101 using transfer learning techniques. Specifically, the ResNeSt models are the new variant of ResNet that utilizes a split-attention network. The main objective is to compare the potential of ResNet and ResNeSt models for detecting COVID-19 from chest X-ray images. The models used in this study underwent training and validation using the most extensive publicly accessible repository of COVID-19 chest X-ray (CXR) images. We evaluated the models’ ability to generalize to new data by testing their performance on independent data not utilized during training or validation. Evaluated the performance of all models in which ResNet models showed higher precision, recall, and accuracy scores than ResNeSt models. Our result indicates that deep-learning models show significant potential for COVID-19 medical research, providing a promising avenue for a deeper understanding of COVID-19 disease.