Abstract Due to the COVID-19 outbreak, there is an urgent need for quick, accurate diagnosis methods, especially CXRs. This study investigates the need for advanced technologies for accurate and timely COVID-19 diagnosis, crucial for isolation protocols, medical interventions, public health management, and distribution of healthcare resources. The novel “Hybrid Attention-Based Ensemble Architecture” addresses the difficulties in distinguishing between CXR and COVID-19 radiographs. The architecture combines the best features of ResNet-50 and XGBoost models to provide a robust and effective system for the classification of COVID-19 CXR images. Data is collected from various sources and cleaned through normalization, resizing, and data augmentation methods. ResNet-50 and XGBoost models are selected for their strengths in deep learning and ensemble techniques. Rough training is done using the collected dataset, and the model is ensembled by the combination of trained models into a hybrid ensemble. Based on the F1 score, performance is also evaluated for accuracy, and the model is evaluated on diverse datasets. The research emphasizes the need for a globally flexible model, considering demographic and geographic variations in the data. Accurate COVID-19 identification is achieved by utilizing deep learning, attention mechanisms, ensemble techniques, and Several databases of CXR images. The model’s generalizability is confirmed by cross-dataset assessments, and the diagnostic procedure is clarified by explanatory techniques. The Hybrid Attention-Based Ensemble Architecture showed increased performance with an accuracy of 98.3%, demonstrating its enormous potential to revolutionize the field of COVID-19 diagnosis.
Yenurkar et al. (Wed,) studied this question.