Reproduce “An LLM-Powered Agent for Physiological Data Analysis: A Case Study on PPG-based Heart Rate Estimation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Mohammad Feli, Iman Azimi, Pasi Liljeberg, Amir M.Rahmani
Researchers at the University of Turku and UC Irvine developed an LLM-powered agent that orchestrates established analytical tools for physiological data analysis, effectively overcoming token limits and a lack of analytical rigor found in direct LLM applications. In a case study on heart rate estimation from PPG signals, the agent achieved a Mean Absolute Error of 2.83 BPM, significantly outperforming GPT-4o-Mini (9.41 BPM) and GPT-4o (8.93 BPM), while also reducing outlier rates to 0.08%.
Key insights
- LLMs are more effective when used as intelligent orchestrators to interface with and leverage specialized, analytically rigorous tools rather than being directly applied for complex numerical reasoning or waveform interpretation.
- Decomposition of complex health analysis queries into executable tasks by an LLM-powered Task Planner, combined with robust external processing pipelines, leads to more reliable and context-aware results.
- Integrating validated, domain-specific processing pipelines significantly reduces analytical errors and improves the robustness of physiological parameter estimation compared to LLMs generating and executing their own code.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.