Larger and Denser: An Optimal Design for Surface Grids of EMG Electrodes to Identify Greater and More Representative Samples of Motor Units

Results from the 200 simulated motor units (MUs) with 84 configurations of grids of electrodes. A, Each solid line represents a motor unit territory, the scatters being the muscle fibers. Blues lines are the theoretically identifiable motor units with a grid of 21.6 cm² and an interelectrode distance (IED) of 18 mm, while the orange lines are the motor units revealed with a grid of 21.6 cm² and an IED of 2 mm. Gray lines represent the nonidentifiable motor units. The percentage of theoretically identifiable motor units (B) and their distance from the skin (C) are reported for the 84 configurations.

Discharge times of the maximum number of motor units identified in one participant (S1) at 30% (A) and 50% maximal voluntary contraction (MVC) (B), with 79 and 58 identified motor units, respectively. The motor units were identified with separated decompositions of the four grids of 64 electrodes (4-mm interelectrode distance). C, Discharge times of the 30 first recruited motor units during the ascending ramp of force (black curve) at 30% MVC (black box in A).

Effect of the electrode density on the number of identified motor units N at 30% (A, B) and 50% maximal voluntary contraction (MVC) (C, D). The boxplots in the left column report the absolute number N of identified motor units per participant (gray dots) and the median (orange line), quartiles, and 95% range across participants. In the right column, the normalized number of motor units N¯ logarithmically decreases with interelectrode distance d (4, 8, 12, and 16 mm in abscissa) as N¯=195−68log(d)(r2=1.0,p=2.5⋅10−5) at 30% MVC (B) and N¯=196−71log(d)(r2=0.99,p=0.001) at 50% MVC (D). The standard deviation of N¯ across subjects is displayed with vertical bars. Moreover, the quality of the motor unit pulse trains (i.e., decomposition accuracy, estimated by the PNR) increased when increasing the density of electrodes (Extended Data Fig. 4-3 for more details). Two decomposition procedures were considered for the 256-electrode condition; the grid of 256 black electrodes indicates that the 256 signals were simultaneously decomposed and the grid of 256 electrodes of four different colors indicates that four subsets of 64 electrodes were decomposed. To maintain consistency with the computational study, the trendlines were fitted with the 4*64 condition, which returned the higher number of identified motor units (Extended Data Fig. 4-1 for the other fitting condition). It is worth noting that computationally increasing the density of electrodes by resampling the EMG signals with a spatial interpolation did not reveal any previously hidden motor units (Extended Data Fig. 4-2).

Effect of the size of the grid on the number of identified motor units N at 30% (A, B) and 50% maximal voluntary contraction (MVC) (C, D). The boxplots in the left column report the absolute number N of identified motor units per participant (gray dots) and the median (orange line), quartiles, and 95% range across participants. In the right column, the normalized number of motor units N¯ logarithmically decreases with the size of the grid s (2, 3.8, 7.7, and 36 cm² in abscissa) as N¯=−20+33log(s)(r2=0.99,p=3.0⋅10−4) at 30% MVC (B), and N¯=−19+32log(s)(r2=0.98,p=0.001) at 50% MVC (D). The standard deviation of N¯ across subjects is displayed with vertical bars. Moreover, the quality of the identified motor unit pulse trains (i.e., decomposition accuracy, estimated by the pulse-to-noise ratio) increased when increasing the size of the grid (Extended Data Fig. 4-3 for more details). Two decomposition procedures were considered for the 256-electrode condition; the grid of 256 black electrodes indicates that the 256 signals were simultaneously decomposed and the grid of 256 electrodes of four different colors indicates that four subsets of 64 electrodes were decomposed. To maintain consistency with the computational study, the trendlines were fitted with the 4*64 condition, which returned the higher number of identified motor units (Extended Data Fig. 4-1 for the other fitting condition).

Effect of the number n of electrodes on the normalized number N¯ of identified motor units at 30% (A) and 50% maximal voluntary contraction (MVC) (B). The discrete results per participant are displayed with gray data points. The average values N¯ per condition are displayed with black crosses. Weighted logarithmic trendlines were fitted to the data and returned (A) N¯=−104+37log(n)(r2=0.98,p=0.018) , and (B) N¯=−113+38log(n)(r2=0.95,p=0.016) . Two decomposition procedures were considered for the 256-electrode condition; the grid of 256 black electrodes indicates that the 256 signals were simultaneously decomposed and the grid of 256 electrodes of four different colors indicates that four subsets of 64 electrodes were decomposed. To maintain consistency with the computational study, the trendlines were fitted with the 4*64 condition, which returned the higher number of identified motor units (Extended Data Fig. 4-1 for the other fitting condition).

A, Typical frequency distribution of motor unit force recruitment thresholds in a human TA. The black dashed lines denote the theoretical portions of the population of motor units recruited at 30% and 50% maximal voluntary contraction (MVC). Effect of the grid density (B, D, F) and grid size (C, E, G) on the percentage of early recruited motor units identified at 30% (B, C, F, G) and 50% MVC (D, E). The boxplots report the results per participant (gray dots) and the median (orange line), quartiles, and 95% range across participants. (F) At 30% MVC, the percentage of early recruited identified motor units logarithmically decreases with interelectrode distance d (4, 8, 12, and 16 mm in abscissa) as 44.6−13.1log(d)(r2=0.91,p=2.8⋅10−3) . (G) At 30% MVC, the percentage of early recruited identified motor units does not vary with the size of the grid s (2, 3.8, 7.7, and 36 cm² in abscissa), the logarithmic trendline fitting (20.5+1.2log(s) ) returning a negligible slope and low r2=0.28(p=8⋅10−4) . The standard deviation across subjects is displayed with vertical bars. Two decomposition procedures were considered for the 256-electrode condition; the grid of 256 black electrodes indicates that the 256 signals were simultaneously decomposed and the grid of 256 electrodes of four different colors indicates that four subsets of 64 electrodes were decomposed. To maintain consistency with the computational study, the trendlines were fitted with the 4*64 condition, which returned the higher number of identified motor units (Extended Data Fig. 4-1 for the other fitting condition). We did not report the results when five or fewer motor units were identified in one condition for three or more participants.

Effect of the electrode density on the correlation ρ between the profiles of motor unit action potentials (MUAP) detected over adjacent electrodes (A) at 30% (B) and 50% maximal voluntary contraction (MVC) (C). The profile of the MUAP detected over the red electrode was compared with those detected over the four adjacent electrodes separated by a 4 mm (orange), 8 mm (blue), 12 mm (green), and 16 mm (purple) interelectrode distance (A). The boxplots denote the correlation coefficient ρ per participant (gray dots) and the median (orange line), quartiles, and 95% range across participants.