---
license: mit
tags:
  - medical-imaging
  - chest-xray
  - pneumonia-detection
  - efficientnet
  - pytorch
  - image-classification
datasets:
  - chest-xray-pneumonia
metrics:
  - accuracy
  - auc
---

# MediScan AI — EfficientNetB4 Chest X-Ray Classifier

Classifies chest X-rays as **NORMAL** or **PNEUMONIA**.

## Model Details
- Architecture: EfficientNetB4 (transfer learning, two-phase fine-tuning)
- Input: 380×380 RGB chest X-ray image
- Output: NORMAL | PNEUMONIA + confidence score
- Explainability: Grad-CAM heatmap overlay

## Performance (Kaggle Chest X-Ray Test Set, n=624)
| Metric | Value |
|--------|-------|
| Accuracy | 87.66% |
| AUC-ROC | 0.9428 |
| Avg Precision | 0.9605 |
| Pneumonia Recall | 93.59% |

## Training
- Dataset: Kaggle Chest X-Ray Images (Pneumonia) — 5,863 images
- Optimizer: AdamW + Cosine Annealing
- Epochs: 7 (early stopping)
- Hardware: Kaggle T4 GPU (8.5 min)

## Usage
```python
import torch
from inference import engine

engine.load("mediscan_v5.pth")
result = engine.predict(open("xray.jpg", "rb").read())
print(result["predicted_class"], result["confidence"])
```

## Disclaimer
For research and educational purposes only.
Not a certified medical device.