Covid-19 Detection Using Stacked Ensemble Learning Approach

This paper presents a novel methodology that incorporates deep learning models, specifically EfficientNet B2, EfficientNet B3, and Fine-Tuned EfficientNet B3, to analyze chest X-ray images with dimensions <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">260× 260</tex> pixels for EfficientNet B2, <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">300× 300</tex> pixels for EfficientNet B3, and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">300× 300</tex> pixels for Fine-Tuned EfficientNet B3. The primary focus of this study is the classification of these images into COVID-19 affected or normal cases using three distinct datasets specifically designed to cater to the specific input requirements of each model architecture. The study commenced with extensive data pre-processing, which involved collecting chest X-ray images and subsequently dividing them into training and validation sets for model training and evaluation, respectively. The images were then fed into their respective models, and the training process involved computing training accuracy, validation accuracy, training loss, and validation loss metrics. However, the achieved accuracies fell short of expected levels. Nevertheless, this study represents a pioneering effort to employ larger datasets in this research domain, thereby emphasizing the importance of conducting experiments on extensive datasets. The innovative utilization of large datasets lays the foundation for future research in this area, demonstrating the potential of refined model architectures or alternative approaches to effectively leverage larger datasets. This research contributes to the advancement of the scientific community's knowledge of deep learning applications in medical imaging, particularly in the context of COVID-19 detection using chest X-ray images. The significance of this study transcends its immediate outcomes and lies in its contribution to opening the door for future initiatives that aim to improve the model performance by leveraging extensive datasets.