Machine learning-based estimation of EMG baseline noise standard deviation without rest trials.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Neuroscience

Background:

• Accurate estimation of resting noise standard deviation (σnoise) in surface electromyography (EMG) is vital for reliable amplitude estimation, especially in low-level contractions.
• Traditional methods require explicit rest-state EMG recordings, which are not always feasible in real-world applications.

Purpose of the Study:

• To compare the accuracy of a novel machine learning (ML) approach and a fixed noise value against direct rest-state measurements for estimating σnoise.
• To evaluate the effectiveness of these methods in reducing baseline EMG noise.

Main Methods:

• Compared direct rest-state σnoise measurement with an ML model trained on simulated and real EMG data from 62 subjects across three systems.
• Utilized a fixed σnoise value of 3% maximum voluntary EMG (MVE) as a third comparison method.
• Assessed noise reduction performance in a separate evaluation.

Main Results:

• The ML approach showed a median absolute difference of 1.4% MVE from rest-based σnoise.
• The fixed σnoise approach yielded a median absolute difference of 1.71% MVE, not significantly different from the ML method.
• Both ML and fixed σnoise methods achieved 45% noise reduction, while rest calibration achieved 75%.

Conclusions:

• Machine learning and fixed σnoise value are feasible alternatives to direct rest-state measurements for EMG noise estimation.
• These methods enhance the reliability of EMG analysis in settings lacking explicit rest data.
• Further research may optimize these alternative noise estimation techniques.