Collaborative multi-agent conversational artificial intelligence for clinical support in Parkinson disease

Area of Science:

• Artificial Intelligence in Medicine
• Neurodegenerative Disorders
• Clinical Decision Support Systems

Background:

• Parkinson disease (PD) presents significant challenges in diagnosis, treatment, and care due to information gaps.
• Patients and healthcare providers require timely, context-specific information for effective PD management.

Purpose of the Study:

• To develop and evaluate a collaborative multi-agent conversational AI system for Parkinson disease.
• To enhance clinical decision-making and personalize the management of Parkinson disease.

Main Methods:

• A multi-agent system with generation, critique, and synthesis agents using Qwen3-Medical-GRPO.
• Retrieval-Augmented Generation (RAG) framework accessing 80 curated medical resources.
• Integration of user profiling and knowledge graphs for personalized responses.

Main Results:

• The system achieved 95% clinical accuracy on 50 representative queries.
• Diagnostic suggestions scored 4.8/5, and treatment recommendations scored 4.6/5.
• Average response time was 6.5 seconds.

Conclusions:

• The AI system offers explainable, scalable, and personalized conversational support for Parkinson disease.
• It addresses gaps in care continuity, personalization, and accessibility.
• The system has the potential to improve clinical workflows and patient outcomes in Parkinson disease management.