README.md

---
title: StrokePainSideDetector
emoji: 🧠
colorFrom: red
colorTo: blue
sdk: streamlit
sdk_version: 1.44.1
app_file: app.py
pinned: false
license: mit
---

# 🧠 Stroke Patient Pain Intensity Detector

This app predicts the **pain intensity (PSPI score)** of stroke patients based on facial expressions using deep learning. It also detects the **affected side** of the face and uses the **unaffected half** for analysis.

## 🚀 What This App Does

1. 🧠 **Detects affected side** of the face using a Keras CNN model.
2. 🎯 **Crops the unaffected side** of the face using OpenCV + Haar cascades.
3. 📈 **Predicts pain score (PSPI)** using a PyTorch ResNet model.

If the face is tilted, the app will automatically rotate it for correct alignment before prediction.

---

## 🖼️ Input

Upload a **full-face photo** of a stroke patient. The app:
- Detects the face
- Splits the face in half (face point of view)
- Uses the expressive side for pain prediction

---

## 📊 Output

- **Affected Side** (face point of view)
- **Unaffected Side**
- **Predicted PSPI Score**: A number between 0 and 6
- **Raw pain model output**
- **Raw stroke model output** (value closer to 0 → left side affected; closer to 1 → right side affected)

---

## 📚 PSPI Score Scale

| Score | Pain Level        |
|-------|-------------------|
| 0     | No pain           |
| 1–2   | Mild pain         |
| 3–4   | Moderate pain     |
| 5–6   | Severe pain       |

The PSPI (Prkachin and Solomon Pain Intensity) score is based on facial action units like eye tightening, brow lowering, and cheek raising.

---

## 💾 Models Used

| Model                     | Task                            | Format         | Link                                                                 |
|--------------------------|----------------------------------|----------------|----------------------------------------------------------------------|
| `cnn_stroke_model.keras` | Detect affected facial side      | TensorFlow     | [Download](https://huggingface.co/AdhamQQ/cnn_stroke_model/resolve/main/cnn_stroke_model.keras) |
| `pain_model.pth`         | Predict PSPI pain intensity      | PyTorch        | [Download](https://huggingface.co/AdhamQQ/cnn_stroke_model/resolve/main/pain_model.pth)         |

---

## 🛠️ Technologies

- **Streamlit** for the web interface
- **OpenCV** for face detection
- **TensorFlow/Keras** for stroke side classification
- **PyTorch** for PSPI pain intensity prediction

---

## 🤝 Credits

Developed by [Your Name or Team Name]  
With support from Hugging Face Spaces

---

## 🔗 Live App

Launch it on Hugging Face Spaces:  
[👉 Try It Now](https://huggingface.co/spaces/YourUsername/StrokePainSideDetector)