Topology of surface electromyogram signals: hand gesture decoding on Riemannian manifolds

Area of Science:

• Biomedical Engineering
• Neuroscience
• Signal Processing

Background:

• Noninvasive surface electromyogram (sEMG) signals are crucial for upper limb gesture decoding.
• Existing methods struggle with inter-individual and inter-session signal variability.
• Decoding gestures has applications in prosthetics, limb augmentation, and virtual reality control.

Purpose of the Study:

• To develop a novel method for decoding hand gestures from sEMG signals.
• To leverage spatial patterns of motor unit (MU) activity for improved gesture classification.
• To address the challenge of signal variability in sEMG analysis.

Main Methods:

• Analyzing sEMG signals recorded from an array of electrodes around the forearm.
• Constructing symmetric positive definite covariance matrices to represent spatial MU activity.
• Utilizing Riemannian manifolds to analyze multivariate sEMG time-series.

Main Results:

• Distinct hand gestures were classified using both supervised and unsupervised learning.
• The method robustly models complex interactions across spatially distributed MUs.
• The approach demonstrated superior performance compared to current benchmarks.
• The technique addresses signal variability across individuals and sessions.

Conclusions:

• Analyzing spatial sEMG patterns on Riemannian manifolds offers a robust framework for gesture decoding.
• The proposed method provides a flexible and transparent way to quantify sEMG differences.
• This computationally efficient technique enhances gesture recognition for various applications.