An LLM-Powered Agent for Physiological Data Analysis: A Case Study on PPG-based Heart Rate Estimation.

Area of Science:

• Artificial Intelligence in Healthcare
• Biomedical Signal Processing
• Wearable Health Technology

Background:

• Large language models (LLMs) are increasingly used in healthcare for tasks like diagnosis and patient care.
• Applying LLMs to physiological time-series data (e.g., wearables) presents challenges due to token limits and analytical limitations.
• Current methods for integrating time-series data with LLMs often yield generic or unreliable health insights.

Purpose of the Study:

• To develop an LLM-powered agent for accurate physiological time-series analysis.
• To bridge the gap between LLMs and established analytical tools for health insight extraction.
• To enhance the reliability and accuracy of health data interpretation using AI.

Main Methods:

• Developed an LLM-powered agent using OpenAI's GPT-3.5-turbo model within the OpenCHA framework.
• Implemented an orchestrator to integrate user interaction, data sources, and analytical tools.
• Conducted a case study on heart rate (HR) estimation from Photoplethysmogram (PPG) signals, using Electrocardiogram (ECG) as the gold standard.

Main Results:

• The developed agent significantly outperformed benchmark LLMs (GPT-4o-mini, GPT-4o) in heart rate estimation accuracy.
• Achieved lower error rates and more reliable HR estimations compared to baseline models.
• Demonstrated the agent's effectiveness in a remote health monitoring context using PPG and ECG data.

Conclusions:

• The LLM-powered agent offers a robust solution for analyzing physiological time-series data, overcoming limitations of existing methods.
• This approach enhances the potential of LLMs in extracting meaningful health insights from complex biomedical signals.
• The agent's implementation is publicly available, promoting further research and development in AI-driven healthcare.