Pneumonia Detection with Explainable Artificial Intelligence

This repository provides MATLAB implementations for pneumonia detection using deep learning models, enhanced with explainable AI techniques.

Dataset: https://www.kaggle.com/datasets/paultimothymooney/chest-xray-pneumonia

This repository presents an approach for pneumonia detection using a MobileNetV2 pretrained model in MATLAB, combined with explainable AI techniques to improve interpretability. The deep learning model is trained on chest X-ray images, leveraging Grad-CAM, Grad-CAM++, Score-CAM, and Saliency Map to highlight critical regions that influence the model’s predictions.Key Features:

1. MobileNetV2 Pretrained Model – A lightweight and efficient deep learning architecture for pneumonia classification.
2. Grad-CAM & Grad-CAM++ – Gradient-based visualization techniques to generate class-discriminative heatmaps.
3. Score-CAM – A perturbation-based method that improves upon Grad-CAM by removing the dependency on gradients.
4. Saliency Map – Highlights pixel-wise contributions to the model’s decision.
5. MATLAB Implementation – Fully coded in MATLAB, making it accessible for medical imaging researchers and engineers.

This project aims to enhance the transparency, reliability, and trustworthiness of AI-based medical diagnosis by providing clear visual explanations of model decisions, assisting clinicians in understanding why and how the model detects pneumonia.