RGFSAMNet: An interpretable COVID-19 detection and classification by using the deep residual network with global feature fusion and attention mechanism

Artificial intelligence has shown considerable promise in fields like medical imaging. Existing testing limitations necessitate reliable approaches for screening COVID-19 and measuring its adverse effects on the lungs. CT scans and chest X-ray images are vital in quantifying and accurately classifying COVID-19 infections. One significant advantage of deep learning models in medical image analysis for detection and classification is that they are a top-notch way to diagnose diseases. For this purpose, we have utilized the power of a deep residual learning network with a global feature fusion technique and attention mechanism to develop our proposed model named "RGFSAMNet" in this study to diagnose the COVID-19 infected patient accurately from a CT scan and chest X-ray images. We have used two publicly available datasets named "SARS-COV-2," which consists of 2482 CT scan images with two classes, and another chest X-ray dataset that contains 12,576 images with three classes. To check the effectiveness of our model, we have trained and tested the model on two different types of datasets. We also generated the Grad-CAM, LIME, and SHAP visualization based on our proposed model, which can represent the identification of the affected areas regions in images and describe the models interpretability level. These experimental results show that the proposed model architecture can achieve accurate classification of COVID-19 affected CT scans and X-ray images despite a lack of data, with the highest accuracy of 99.60% on test data for CT scans and 99.48% on X-ray image detection and classification. We also developed a web-based User Interface (UI) for the model validation to detect and classify COVID-19 images. Our proposed model exceeds some previous state-of-the-art performance levels. We think our contributions will help clinicians detect and classify COVID-19 images effectively and save human lives.