Utilizing deep learning architectures for early detection of lung diseases in chest X-ray images

Abstract

In addressing the critical need for early detection of COVID-19 and viral pneumonia, our study introduces an advanced system using deep learning for analyzing chest X-ray images and their respective lung masks. Utilizing the comprehensive COVID-QU-Ex dataset, we extensively trained and evaluated five pre-trained deep learning models (ResNet-18, VGG-16, AlexNet, Inception-V3, and DenseNet-169) and developed a novel ensemble model. This ensemble model, combining the strengths of the top-performing models, achieved a notable 96.002% prediction accuracy and 95% recall, outperforming the individual models. Our research goes a step further by integrating explainable AI methodologies, specifically LIME and SHAP, to provide understandable and transparent explanations of the model's decision-making process. These explanations are crucial for identifying critical areas of lung infections and enhancing the model's credibility. Furthermore, we deployed a user-friendly Flask web application, enabling easy access to our system for medical professionals. This application offers immediate analysis of uploaded X-ray images, delivering predictions, accuracy percentages, and interpretative visual explanations. Our work significantly advances the field by providing an accurate, explainable, and accessible tool for early lung disease detection, potentially revolutionizing the approach to healthcare diagnostics.