On-Line Recursive Decomposition of Intramuscular EMG Signals Using GPU-Implemented Bayesian Filtering

Area of Science:

• Biomedical Engineering
• Neuroscience
• Signal Processing

Background:

• Real-time intramuscular electromyography (iEMG) decomposition is crucial for biofeedback and neural interfacing.
• Previous methods relied on sequential algorithms and Bayesian filters, limiting real-time performance.

Purpose of the Study:

• To develop a real-time iEMG decomposition algorithm with enhanced performance.
• To implement the algorithm using parallel computation on a Graphics Processing Unit (GPU).

Main Methods:

• Modified a Hidden Markov Model-based sequential decomposition algorithm.
• Replaced the Kalman filter with a least-mean-square filter for motor unit action potential (MUAP) estimation.
• Introduced heuristics to optimize decomposition scenarios and implemented a GPU-parallelized version.

Main Results:

• Achieved real-time decomposition of simulated and experimental iEMG signals.
• Decomposition accuracy exceeded 85%, varying with muscle activation levels.
• Demonstrated successful decomposition of signals with up to 10 active motor units (MUs).

Conclusions:

• The enhanced algorithm provides accurate, real-time interfacing with spinal motor neurons.
• The GPU implementation significantly broadens the applicability of iEMG decomposition.