Fig. 2: Machine and human learning effects for MI-based robotic control tasks.
From: EEG-based brain-computer interface enables real-time robotic hand control at individual finger level
A Offline simulation results (n = 21 subjects) for Session 2 fine-tuned runs using Session 1 base models (S1-B), Session 1 fine-tuned models (S1-F), and Session 2 base models (S2-B), as well as the online decoding results for Session 2 fine-tuned runs (S2-F). The center lines indicate the median values. The boxes extend from the lower quartile to the upper quartile. Diamonds indicate outliers that are more than 1.5 times the interquartile range above the third quartile or below the first quartile. Comparisons between the sessions and the models were made using a two-tailed Wilcoxon signed-rank test with Bonferroni multi-comparison correction (*** if p < 0.001, ** if p < 0.01, * if p < 0.05, n.s. if no statistical significance is found). P-values: 0.0003 (S2-B vs. S2-F, 2-class), 0.0254 (S1-B vs. S2-B, 2-class), 0.0003 (S1-F vs. S2-F, 2-class), 0.0449 (S1-B vs. S1-F, 3-class), 0.0074 (S2-B vs. S2-F, 3-class), 0.0009 (S1-B vs. S2-B, 3-class), 0.0002 (S1-F vs. S2-F, 3-class). B Offline FBCSP decoding results on MI robotic finger control data (n = 21 subjects). S1-B, S1-F, S2-B, and S2-F denote decoding results for Session 1 base runs, Session 1 fine-tuned runs, Session 2 base runs, and Session 2 fine-tuned runs respectively. The center lines indicate the median values. The boxes extend from the lower quartile to the upper quartile. Diamonds indicate outliers that are more than 1.5 times the interquartile range above the third quartile or below the first quartile. Statistical analysis was conducted using a two-way repeated-measures ANOVA with main effects of session and model. Main effect of the model: ### if p < 0.001, n.s. if no statistical significance is found. Post hoc comparisons between models within the same session were made using a two-tailed Wilcoxon signed-rank test with Bonferroni correction (*** if p < 0.001, ** if p < 0.01, * if p < 0.05, n.s. if no statistical significance is found). P-values: 0.0005 (model, 3-class), 0.0114 (S1, 3-class). C Group-averaged performance trends across eight consecutive online runs within the same paradigm. The center lines indicate the mean values. The shaded areas represent the standard errors.
