Abstract
Pneumonia is a serious lung infection that requires early diagnosis for effective treatment. This project presents an AI-based pneumonia detection system that analyzes chest X-ray images using a D4 Equivariant Vision Transformer model. The system enhances images using CLAHE preprocessing and applies D4 transformations to improve robustness to image orientation. The model classifies the images into three categories: bacterial pneumonia, viral pneumonia, and normal. Explainable artificial intelligence techniques such as Grad-CAM and SHAP are integrated to highlight the lung regions responsible for the prediction. A Flask-based web application is developed to allow users to upload chest X-ray images and obtain predictions with confidence scores and visual explanations.
Keywords
Pneumonia Detection, Chest X-Ray Analysis, Vision Transformer, Deep Learning, Explainable AI, Grad-CAM, SHAP, Medical Image Classification


Introduction
Pneumonia is a serious lung infection that affects millions of people worldwide and requires early detection for proper treatment. Chest X-ray imaging is commonly used for diagnosing pneumonia, but manual analysis by radiologists can be time-consuming and may lead to errors. Therefore, automated systems using artificial intelligence can help improve the accuracy and speed of diagnosis.

In this project, an AI-based pneumonia detection system is developed using a D4 Equivariant Vision Transformer model. The system analyzes chest X-ray images and classifies them into three categories: bacterial pneumonia, viral pneumonia, and normal. Image preprocessing techniques such as CLAHE are used to enhance the quality of the images.
The system also integrates Explainable Artificial Intelligence techniques such as Grad-CAM and SHAP to highlight important lung regions that influence the prediction. A Flask-based web application is developed to allow users to upload chest X-ray images and obtain predictions with visual explanations. This system aims to assist healthcare professionals in faster and more accurate pneumonia diagnosis.

Objectives
• To develop an automated system for detecting pneumonia from chest X-ray images.
• To implement a deep learning model using a Vision Transformer architecture.
• To enhance X-ray images using CLAHE preprocessing.
• To classify images into bacterial pneumonia, viral pneumonia, and normal categories.
• To integrate explainable AI techniques such as Grad-CAM and SHAP.
• To develop a Flask-based web application for real-time prediction.

Block Diagram