gptplan

This project, gptplan, is designed for automated radiotherapy treatment planning using advanced AI techniques. It integrates multi-agent systems, retrieval augmented generation, and specialized radiotherapy libraries to streamline and potentially optimize the planning process. You can find our paper A feasibility study of automating radiotherapy planning with large language model agents for more information.

Core Functionality

The system uses AI agents, built with the pyautogen framework, to automate various aspects of treatment planning. Key functionalities include:

• Patient Data Management: Interacts with patient databases (e.g., database_Lung.py for lung cancer cases) to retrieve and manage necessary data.
• AI Agent Collaboration: Employs a team of AI agents (agent_team.py) that collaborate on planning tasks, guided by predefined prompts (prompts.py) and LLM configurations (llm_config.py).
• Retrieval Augmented Generation (RAG): Utilizes RAG techniques (rag.py) to provide agents with relevant information from a knowledge base, enhancing the quality and context-awareness of their decisions.
• Radiotherapy Planning with PortPy: Leverages the portpy library (use_portpy.py) for core radiotherapy calculations, plan optimization, and simulation, using specific configuration files found in portpy_config_files/.

Dependencies

The project is built in Python and relies on several external libraries.

• Python 3.x
• Key Libraries:
  • pyautogen: For building and managing AI agent teams. (Note: pyautogen has undergone significant changes. The version specified in requirements.txt (0.2.33) is recommended to ensure compatibility with this project.)
  • portpy: For radiotherapy planning, optimization, and dose calculation.
  • google-generativeai: For interacting with Google's Generative AI models.
  • pandas, numpy, scipy: For data manipulation and scientific computing.
  • cvxpy: For optimization.
  • matplotlib, seaborn: For plotting and visualization.
  • scikit-image: For image processing.
  • And others as listed in requirements.txt.

Setup Instructions

1. Clone the Repository:

   # git clone <repository_url>
   # cd gptplan

2. Install Dependencies: It is recommended to use a virtual environment.

   python -m venv venv
   source venv/bin/activate  # On Windows use `venv\\Scripts\\activate`
   pip install -r requirements.txt

3. LLM Configuration: The project uses Large Language Models (LLMs) via pyautogen. You need to configure your LLM API keys.

   • Examine llm_config.py. This file likely sets up the configuration for the LLM provider (e.g., Google Generative AI, given the google-generativeai dependency).
   • You may need to create or update llm_config.json or set environment variables with your API keys as required by llm_config.py and pyautogen. Refer to the pyautogen documentation for details on configuring LLM providers.

How to Run

The primary way to run a full auto-planning workflow is likely through the main execution scripts. Individual modules also contain test routines.

1. Main Auto-Planning Workflow: The script auto_planning_patient48_rag.py appears to be an example of an end-to-end automated planning process.

   python auto_planning_patient48_rag.py

   Ensure your LLM configuration is correctly set up before running. This script will likely demonstrate the collaboration of agents, RAG, and PortPy for a specific patient case.

2. Running Individual Modules / Unit Tests: Several modules have if __name__ == "__main__": blocks, often calling unit_test() functions. These can be run directly to test or demonstrate the functionality of that specific module:

   • python agent_team.py
   • python database_Lung.py
   • python rag.py
   • python use_portpy.py

Project Structure Overview

• agent_team.py: Defines and manages the team of AI agents.
• auto_planning_patient48_rag.py: Example script for running the full auto-planning workflow with RAG.
• database.py / database_Lung.py: Handles database interactions and patient data.
• llm_config.py: Configures the Large Language Models for pyautogen.
• llm_config.json: (If used) Stores LLM API keys or other related configurations.
• prompts.py: Contains the prompts used to guide the AI agents.
• rag.py: Implements the Retrieval Augmented Generation logic.
• use_portpy.py: Demonstrates and utilizes functionalities of the portpy library.
• utils.py: Contains utility functions used across the project.
• portpy_config_files/: Directory containing configuration files for portpy, such as clinical criteria and optimization parameters.
• requirements.txt: Lists all Python package dependencies.
• README.md: This file.