CVD Risk Prediction API v1.0.0

CVD Risk Prediction API v1.0.0

Overview

This release introduces the CVD Risk Prediction API, a FastAPI-based service that provides cardiovascular disease (CVD) risk prediction from DICOM images using the Tri2D-Net deep learning model.

Key Features

• DICOM Processing: Upload and process DICOM images in ZIP format
• Heart Detection: Automatic heart region detection using RetinaNet
• Risk Prediction: CVD risk prediction using Tri2D-Net model
• Visualization: Grad-CAM visualization of regions contributing to prediction
• Attention Scoring: Analysis of attention scores for each image slice
• REST API: Complete REST API with documentation

API Endpoints and Response Formats

POST /api_predict

Upload a ZIP file containing DICOM images for CVD risk prediction.

Response Format:

{
  "session_id": "uuid-string",
  "predictions": [
    {
      "score": 0.75  // CVD risk score between 0-1
    }
  ],
  "overlay_images": "http://server/download_zip/uuid-string",
  "attention_info": {
    "attention_scores": [
      {
        "file_name_pred": "pred_image1.png",
        "attention_score": 0.85
      }
    ],
    "total_images": 120,
    "returned_images": 45
  },
  "message": "Prediction successful."
}

GET /download_zip/{session_id}

Download a ZIP file containing visualization results.

Response:

• Binary file (application/zip) containing Grad-CAM visualizations

GET /preview/{session_id}/{filename}

Preview a specific visualization image.

Response:

• Image file (image/png) showing the requested visualization

Technical Details

• Built with FastAPI for high-performance API endpoints
• Integrates SimpleITK for medical image processing
• Implements PyTorch-based deep learning models
• Provides comprehensive logging and error handling
• Includes automatic cleanup of temporary files

Installation

# Clone the repository
git clone https://github.com/your-username/cvd-risk-prediction.git

# Install dependencies
cd cvd-risk-prediction
python setup.py

Usage

# Start the API server
python api.py

API Documentation

Once running, API documentation is available at:

• Swagger UI: http://localhost:8000/docs
• ReDoc: http://localhost:8000/redoc

Model Information

The Tri2D-Net model is based on research published in Nature Communications.

License

This project is licensed under the MIT License - see the LICENSE file for details.