RT-Rename

Overview

RT-Rename is a web-based application designed to standardize radiotherapy (RT) structure nomenclature according to the TG263 guidelines using Large Language Models (LLMs). The application provides an intuitive interface for renaming RT structure sets, supporting both local models (via Ollama) and cloud-based models for intelligent structure name suggestions.

Key Features

• 🏥 Rename structure names: Standardizes structure names according to international radiotherapy guidelines such as TG-263.
• 🤖 AI-Powered: Uses state-of-the-art LLMs for name suggestions such as Llama 3.3, Gemma 3, Qwen 3, Deepseek R1, Deepseek V3 ...
• 📁 Multiple Formats: Supports DICOM RTstruct, CSV, and .nrrd files
• 🌐 Web Interface: User-friendly interface
• 🐳 Docker Ready: Containerized deployment with Docker Compose
• 💾 Export Options: Export as CSV or updated DICOM files

Project Structure

Installation & Setup

Option 1: Docker (Recommended)

Prerequisites

• Docker and Docker Compose
• NVIDIA GPU with Docker GPU support (optional, for local LLM models)

Quick Start

1. Clone the repository

   git clone https://github.com/LMUK-RADONC-PHYS-RES/rt-rename.git
   cd rt-rename

2. Start the application

   cd docker
   docker-compose up -d

3. Access the application

   • Web interface: http://localhost:8055
   • Ollama API: http://localhost:11435

Environment Variables

If you want to use cloud inference create a .env file in the root directory for open-ai API compatible cloud-based models:

OPEN_AI_URL=your_api_url_here
OPEN_AI_API_KEY=your_api_key_here

Option 2: Local Installation

Prerequisites

• Python 3.8 or higher
• pip package manager

Installation Steps

1. Clone the repository

   git clone <repository-url>
   cd rt-rename

2. Create virtual environment

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install dependencies

   pip install -r docker/requirements.txt

4. Install Ollama (for local models)

   # On Linux/macOS
   curl -fsSL https://ollama.ai/install.sh | sh
   
   # On Windows
   # Download from https://ollama.ai/download

5. Start Ollama service

   ollama serve

6. Pull required models

   ollama pull llama3.1:70b-instruct-q4_0
   # or other models as configured in config/models.json

7. Run the application

   python app.py

8. Access the application

   • Web interface: http://localhost:8055

Usage Guide

Web Interface

1. Configure Settings

   • Select nomenclature type (TG263 or TG263_reverse)
   • Choose anatomical regions (Thorax, Head and Neck, etc.)
   • Set target volume filtering (yes/no)
   • Select LLM model and prompt version

2. Upload Structure Data

   • DICOM RTstruct files (.dcm)
   • CSV files with structure names
   • Directory with .nrrd files (drag & drop supported)

3. Review Suggestions

   • View AI-generated TG263 name suggestions
   • Check confidence scores and verification status
   • Accept/reject individual suggestions
   • Add comments for manual review

4. Export Results

   • Export as CSV file
   • Export as updated DICOM RTstruct file

Batch Processing

For large-scale processing, modify the example batch script:

python batch_rename.py

Edit the script to specify:

• Input file paths
• Output file paths
• Model and prompt settings
• Batch size

Supported File Formats

DICOM RTstruct

• Standard DICOM RT Structure Set files
• Automatically extracts ROI names
• Can export updated DICOM files

CSV Files

• First column: structure names
• Optional additional columns for existing TG263 names, confidence, etc.
• Exports enhanced CSV with AI suggestions

.nrrd Files

• Directory containing structure files
• Filters out intermediate files related to synthRAD2025 preprocessing (_stitched, _s2_def, _s2)
• Supports batch processing

Configuration

Models Configuration

Edit config/models.json to add or modify available models:

{
   "name": "Qwen 3",
   "parameters": "30B",
   "model_str": "qwen3:30b-a3b", 
   "cloud": false
},

• name: Display name in the UI
• parameters: Model size (e.g., 30B, 70B)
• model_str: if cloud == false Ollama model string else model string used by the cloud inference api (e.g., llama3.3:70b-instruct-q4_0)
• cloud: Set to true for cloud-based models, false for local models

Prompt Engineering

Customize prompts in config/prompt_v*.txt:

• Add domain-specific instructions
• Modify output format requirements
• Include additional context or examples

TG263 Guidelines

The config/TG263_nomenclature.xlsx file contains the official TG263 nomenclature. Update this file to use custom guidelines or newer versions.

Troubleshooting

Common Issues

1. GPU Memory Issues

   # Use smaller models or increase swap
   ollama pull llama3.2:3b-instruct-fp16

2. Port Conflicts

   # Change ports in docker/compose.yaml
   ports:
     - "8056:8055"  # Change external port

3. Model Loading Errors

   # Pull model manually
   docker exec -it rt-rename-ollama-1 ollama pull llama3.1:70b-instruct-q4_0

4. File Upload Issues

   • Check file permissions
   • Verify file format compatibility
   • Ensure sufficient disk space

Performance Optimization

• Use GPU acceleration for local models
• Adjust batch sizes based on available memory
• Use cloud models for faster inference
• Enable caching for repeated operations

Adding New Models

1. Update config/models.json
2. Implement model-specific handling in utils.py
3. Test with sample data
4. Update documentation

Citation

If you use RT-Rename in your research, please cite:

add paper once it's published

License

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

Acknowledgments

• TG263 Working Group for nomenclature standards
• Ollama team for local LLM infrastructure
• Dash/Plotly for the web framework
• PyDICOM developers for DICOM handling capabilities