Fig. 2: EMG classification accuracy.
From: Biomimetic versus arbitrary motor control strategies for bionic hand skill learning
EMG data when testing the trained classifier. a,b, Example visualizations of the EMG signals acquired from the bionic hand system (a) and the real-time EMG signals and classification decisions while an example participant engaged and held each hand gesture for 20 s (b) (Methods). During this assessment, the Coapt controller outputted a real-time classification decision (updated every 50 ms) for which gesture was being performed. Comparing this output with the gesture participants were instructed to perform by the experimenter, we computed the classification accuracy for each gesture. The time window of interest was approximately the first 2.5 s of each 20 s gesture trial (shown in b as a red window). c, Biomimetic (n = 19) and arbitrary (n = 17) group average classification accuracy matrices for 3 gesture classes (rest, open and close) were assessed on D1, immediately following controller calibration. d, Both trained groups (n = 34) showed a significant increase in average classification accuracy (main effect of day, F(1,31) = 5.647, P = 0.024, η2 = 2.6 × 10−5) between the first (D1) and last day (D4) of training. No significant differences were found between the two training strategies before training (W = 167.0; P = 0.631; rrb = 0.09; 95% CI, [−0.28, 0.45]; BF10 = 0.37) or after training (W = 121.50; P = 0.437; rrb = −0.15; 95% CI, [−0.50, 0.22]; BF10 = 0.37). e, Biomimetic (n = 19) and arbitrary (n = 16) average classification accuracy matrices for 5 gesture classes (rest, open, close, pinch and tripod), assessed on D4, at the end of training. f, No differences between trained groups (n = 34) in average classification accuracy (average of the 5-motion-class diagonal; W = 117.50; P = 0.517; rrb = −0.13; 95% CI, [−0.49, 0.25]; BF10 = 0.42). The dashed line denotes chance level (3 classes, 33%; 5 classes, 20%). Data plotted in d and f reflect group means ± s.e.m. All statistical group comparisons were two-tailed Mann–Whitney tests. Circles depict individual participant means (across relevant items). Values indicate group means ± s.e.m.
