X-PNEUMO: A mobile application for pneumonia detection with meaningful explainability.

Abstract

This thesis addresses a critical challenge in the application of deep learning in medical imaging: developing pneumonia detection systems that provide not just accurate diagnoses but also meaningful explanations that align with clinical reasoning. While current explainable artificial intelligence methods can identify pneumonia, they often highlight irrelevant regions in chest X-rays that lack clinical relevance.This meaningless explanations fail to support medical decision-making and diminishing confidence in AI systems. To solve this problem, we developed X-Pneumo, a mobile application that not only detects pneumonia effectively but also provides clear, medically relevant explanations by focusing specifically on lung areas where pneumonia appears, unlike standard XAI explanations that might point to irrelevant areas. X-Pneumo’s visual guides highlight only clinically significant regions, matching how doctors examine X-rays, following a complete development process from initial research to final implementation, including careful analysis of existing technologies, design of a specialized deep learning model, and the creation of a practical mobile tool for healthcare settings. Testing showed our approach produces better explanations than conventional methods while maintaining high diagnostic accuracy, representing an important step toward making AI more transparent and useful in clinical practice, with future work potentially expanding this approach to other diseases and further improving the system’s reliability across diverse patient cases.