Detection of COVID-19 using EfficientnetV2-XL and Radam Optimizer from Chest X-ray Images

Abstract

Automating the detection of the COVID-19 pandemic has become necessary for assisting radiologists and medical practitioners in the diagnosis process. It enables them not only to save time through early diagnosis but also to ensure that they are making more accurate diagnoses. Therefore, this research presents a novel approach for automatically identifying COVID-19 in chest X-ray images by utilizing the EfficientNetV2-XL model in combination with the Rectified Adam optimizer for training. For conducting the experiments, we used the dataset available on Kaggle, known as the “COVID-19 Radiography Dataset.” The totality of this dataset was 21,165, and it included four patterns: COVID-19, viral pneumonia, lung opacity, and normal cases. The dataset was divided into 80% training and 20% testing. The preprocessing stage included resizing images to 512 × 512 pixels and then applying data augmentation techniques to enhance model robustness. Consequently, a fine-tuned multiclass categorization system was implemented. The proposed system's effectiveness is evidenced by the experimental outcomes, which show a 99.31% accuracy rate and a perfect Area Under the Curve score of 1 for identifying COVID-19. Additionally, the Score-CAM visualization method was utilized to enhance the interpretability of model predictions, identifying key regions within the chest X-ray images that influence the classification outcome. This Localization technique aids healthcare professionals in understanding the reasoning behind the model and confirming the accuracy of the diagnosis. The proposed system outperformed the state-of-the-art models for COVID-19 detection.