DoctorAgent-RL: A Multi-Agent Collaborative Reinforcement Learning System for Multi-Turn Clinical Dialogue

This repository contains the DoctorAgent-RL model, which is a reinforcement learning (RL)-based multi-agent collaborative framework designed to revolutionize clinical dialogue. The model is presented in the paper DoctorAgent-RL: A Multi-Agent Collaborative Reinforcement Learning System for Multi-Turn Clinical Dialogue.

Code: https://github.com/JarvisUSTC/DoctorAgent-RL

Introduction

DoctorAgent-RL addresses the critical limitations of static clinical dialogue systems by modeling medical consultations as dynamic decision-making processes under uncertainty. It enables:

1. Adaptive Information Gathering: Intelligent adjustment of dialogue paths based on patient responses.
2. Clinical Reasoning Alignment: Autonomous development of interaction strategies consistent with medical logic.
3. Overcoming Static Paradigms: Moving beyond superficial pattern imitation in existing dialogue datasets.

Through continuous multi-turn interactions between doctor and patient agents, optimized via reinforcement learning, DoctorAgent-RL achieves significant improvements in diagnostic accuracy and interaction efficiency.

Key Features

• 🧠 Multi-Agent Collaboration: Doctor and patient agents with distinct roles and objectives.
• 📈 Dynamic Strategy Optimization: Reinforcement learning-based policy updates for adaptive behavior.
• 🎯 Comprehensive Reward Design: Multi-dimensional consultation evaluation metrics guiding optimal strategies.
• 📊 Medical Knowledge Integration: Clinical reasoning logic embedded in decision-making processes.
• 📄 MTMedDialog Dataset: The first English multi-turn medical consultation dataset designed with simulation capabilities.

Usage

You can use the DoctorAgent-RL model with the Hugging Face transformers library for text generation in a multi-turn dialogue context.

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

# Load the model and tokenizer
model_id = "Jarvis1111/DoctorAgent-RL" 
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.bfloat16, device_map="auto")

# Prepare a sample conversation
messages = [
    {"role": "user", "content": "Hello Doctor, I have a headache and feel tired."},
]

# Apply the chat template defined in the tokenizer_config.json
# This is crucial for proper multi-turn dialogue with Qwen models
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)

# Generate response
input_ids = tokenizer(text, return_tensors="pt").input_ids.to(model.device)
output = model.generate(input_ids, max_new_tokens=256, do_sample=True, temperature=0.7, top_p=0.9)
response = tokenizer.decode(output[0], skip_special_tokens=True)

print(response)

Citation

If DoctorAgent-RL contributes to your research, please consider citing our work:

@article{feng2025doctoragent,
  title={DoctorAgent-RL: A Multi-Agent Collaborative Reinforcement Learning System for Multi-Turn Clinical Dialogue},
  author={Feng, Yichun and Wang, Jiawei and Zhou, Lu and Li, Yixue},
  journal={arXiv preprint arXiv:2505.19630},
  year={2025}
}