README.md

---
title: Pneumonia Detector
emoji: 🫁
colorFrom: purple
colorTo: red
sdk: gradio
sdk_version: 6.12.0
app_file: app.py
pinned: false
license: mit
short_description: Pneumonia detection from chest X-rays using ViT and Grad-CAM
---

# 🫁 Pneumonia Detector

Medical AI system for detecting pneumonia from chest X-ray images using Vision Transformer (ViT) with explainable Grad-CAM heatmaps.

## Features

- **Upload chest X-ray** — supports PNG, JPEG
- **AI prediction** — Normal or Pneumonia with confidence score
- **Grad-CAM heatmap** — visual explanation showing which regions the model focused on
- **Fast inference** — runs on CPU or GPU

## Model Details

| Attribute | Value |
|-----------|-------|
| Architecture | Vision Transformer (ViT) |
| Backbone | ViT-B/16 |
| Fine-tuned on | Chest X-ray dataset |
| Accuracy | 87% |
| AUC-ROC | 0.98 |
| Input size | 224x224 pixels |

## Training Details

| Parameter | Value |
|-----------|-------|
| Dataset | Chest X-ray (Pneumonia) |
| Train/Val/Test split | 80/10/10 |
| Epochs | 20 |
| Batch size | 32 |
| Learning rate | 1e-4 |
| Optimizer | AdamW |
| Hardware | Kaggle P100 (16GB) |
| Training time | ~3 hours |

## Performance Metrics

| Metric | Score |
|--------|-------|
| Accuracy | 87% |
| AUC-ROC | 0.98 |
| Precision | 85% |
| Recall | 89% |
| F1-Score | 87% |

*Tested on held-out test set of 624 images*

## How to Use

1. Upload a chest X-ray image (frontal view preferred)
2. Click **Submit**
3. View prediction + Grad-CAM heatmap

## Example Output

**Prediction:** Pneumonia (confidence: 94%)  
**Heatmap:** Highlights upper right lung region

## Technical Stack

- PyTorch
- Hugging Face Transformers
- timm (Vision Models)
- Grad-CAM
- Gradio

## Run Locally

```bash
git clone https://huggingface.co/spaces/mou11/pneumonia-detector
cd pneumonia-detector
pip install -r requirements.txt
python app.py