A Pilot Study on Quantifying Signal Quality in High-Density Surface Electromyography

Adequate electromyography (EMG) signal quality is important for obtaining correct interpretations for EMG diagnostics and EMG-based control. In this pilot study, a dataset (
$N=2$
) of high-density recordings of the biceps and triceps was rated on a four-point scale by two expert EMG raters. The intrarater reliability was good-to-excellent (intra-class correlation (ICC) (2,1)
$>$
0.76) and the interrater reliability was good (ICC (3,k) = 0.85). Four regression models were developed to label electrodes automatically: 1) linear regression; 2) random forest regression; 3) support vector regression; and 4) a combination of the models obtained via majority voting. Using the human raters as ground truth, the random forest and support vector regression obtained a very strong correlation (
$r_{s}$
= 0.90) and excellent reliability (ICC (3,k) = 0.95). These results demonstrate the potential for the development of an automated process to quantify EMG channel quality.