mou11/pneumonia-detector

🫁 Pneumonia Detector

🎯 Live Demo

👉 Try it on Hugging Face Spaces

🔍 Overview

A medical image classification system that detects pneumonia from chest X-ray images using Vision Transformer (ViT) with Grad-CAM explainability. The model highlights which regions of the X-ray influenced the prediction, making it transparent and trustworthy for medical screening.

📊 Results

Metric	Score
Overall Accuracy	87%
AUC-ROC	0.9805
PNEUMONIA Recall	1.00
NORMAL Precision	0.99

📈 Classification Report

Class	Precision	Recall	F1-Score
NORMAL	0.99	0.66	0.79
PNEUMONIA	0.83	1.00	0.91

✨ Features

• ✅ Chest X-ray classification: NORMAL vs PNEUMONIA
• ✅ Vision Transformer (ViT) fine-tuned on medical images
• ✅ Grad-CAM heatmap visualization for explainability
• ✅ Handles class imbalance with weighted loss
• ✅ Gradio demo for interactive predictions

🏗️ Architecture

Chest X-Ray Image → Preprocessing (224x224) → ViT Base Patch16 → Fine-tuned Classifier Head → Prediction + Grad-CAM Heatmap

🛠️ Tech Stack

Component	Tool
Model	Google ViT Base Patch16 224
Framework	PyTorch
Explainability	Grad-CAM
Dataset	Chest X-Ray Images (Pneumonia) — Kaggle
UI	Gradio
Training	Google Colab T4 GPU

📁 Dataset

• Source: Chest X-Ray Images (Pneumonia)
• Total Images: 5,863
• Classes: NORMAL (1,341) vs PNEUMONIA (3,875)
• Split: Train / Val / Test

🚀 How to Use