Leveraging ResNet for Disease Detection: Uncertainty Quantification and Sensitivity Analysis in Lung X-ray Image Segmentation

This paper focuses on the diagnosis of pneumonia through lung X-ray image analysis using machine learning techniques, particularly leveraging a Residual Network (ResNet) model. Traditional methods relying on radiologists are subjective and prone to misdiagnosis. By integrating uncertainty quantification and sensitivity analysis into the segmentation process, we aim to enhance the accuracy and reliability of pneumonia diagnosis. The ResNet model's architecture allows for deeper network training without the vanishing gradient problem, significantly improving segmentation accuracy. Uncertainty quantification provides confidence estimates for model predictions, while sensitivity analysis identifies key factors influencing segmentation accuracy, such as image quality and model parameters. Our results indicate that these methods and the advanced ResNet model significantly improve the diagnostic process, offering a robust approach to medical image analysis and aiding clinical decision-making. This work demonstrates the potential of advanced machine learning techniques, specifically ResNet, in medical diagnostics, aiming to optimise lung disease diagnostic processes.