Leveraging Large Language Models for Automated Feature Extraction and Model Training in EMG-Based Motion Decoding.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Electromyography (EMG) based motion decoding relies heavily on feature extraction and model training.
• Traditional methods require significant domain expertise and programming skills for signal processing and model optimization.

Purpose of the Study:

• To investigate the feasibility of using Large Language Models (LLMs) to automate EMG feature extraction and motion decoding model development.
• To compare the performance of LLM-generated features and models against manually developed methods.

Main Methods:

• Utilized LLMs for automated feature extraction from EMG data.
• Developed machine learning models for motion decoding using LLM-generated features.
• Compared LLM-based approaches with traditional, manually coded methods.

Main Results:

• LLM-extracted features and their corresponding models achieved performance comparable to traditional methods.
• Demonstrated the potential of AI-driven biosignal processing to accelerate research.

Conclusions:

• LLMs show promise in automating aspects of EMG data analysis and motion decoding.
• Highlights LLMs' capabilities and limitations for muscle-machine interfaces and biomedical signal processing workflows.