Integrating Explainable Artificial Intelligence with Machine Learning for Reliable Pneumonia Detection in Medical Diagnostics

Abstract

Pneumonia is a serious lung infection that impacts millions of people annually, and is a significant problem in modern day health care. It is essential to diagnose the disease early and correctly to minimalize complications and maximize recovery of the patient. It is very common to use a chest X-ray from the imaging procedure in the diagnosis, but, due to the high load of patients and the complexity of images, manual reading might be delayed or might yield different readings from one reader to another. Recently, the use of intelligent diagnostic systems using machine learning has been seen as a potential support to the medical expert in the disease detection. Most of the early systems, however, are black-box systems and do not offer explanations for their predictions. This study presents a dependable pneumonia detection system engineered by integrating machine learning and Explainable Artificial Intelligence techniques to assist medical diagnosis. To assess the robustness and adaptability of the proposed approach, the study employs two chest X-ray datasets. The image preprocessing and augmentation methods made it easier to handle the images and make their visualization as uniform as possible. The model training methods presented in this work were able to assist to obtain better results of the detection process and better understanding of the image data. The explainability analysis involved in the framework assisted to understand the behavior of the object. Visualization methods are used to identify the infected regions that are used in reaching model decisions to make the model more interpretable. Performance metrics are used in experimental analysis.