Fig. 1: Experimental paradigm, online BCI performance, and decoding accuracy. | Communications Biology

Fig. 1: Experimental paradigm, online BCI performance, and decoding accuracy.

From: Learning to operate an imagined speech Brain-Computer Interface involves the spatial and frequency tuning of neural activity

Fig. 1

a Experiment Paradigm. Each trial began with the presentation of the trial number, followed by a fixation cross. The last second of the fixation cross was considered as a baseline for the EEG analysis. Next, participants were instructed to imagine pronouncing either the syllable /fɔ/ or /gi/ as indicated by a text on the screen. They had to continue performing the imagery for 5 s in the offline session, or until the battery was filled during the real-time control (online session). The end of the imagery period was signaled by the tip of the full battery turning yellow. b Average BCI online performance (%) over the 5 training days. BCI performance is computed, separately for each trial, as the percentage of classifier outputs in accordance with the cued syllable. Dots represent the average performance across participants with error bars indicating the standard error of the mean, and the significant linear regression (dashed line). c Online BCI performance (%) for each participant was ordered according to the value of the learning slope (from lowest to highest). Dots represent the individual average BCI-control performance for each training day, with the corresponding regression line. d Correlation between individual learning slopes and average BCI performance across the 5 training days. Positive and negative learning slopes are plotted in cyan and orange, respectively for learners and non-learners. e Model cross-validation (CV) accuracy was obtained by computing the classifier considering the entire dataset from the offline (blue) and online (red) sessions on each day (left). The box plot (right) shows the learning slopes obtained by fitting a linear model per participant and session (offline in blue and online in red) using CV accuracies on each training day. There was no significant difference between the two sessions. Boxes represent the interquartile range (IQR), with the horizontal line indicating the median, and whiskers extending to data points that are within 1.5× the IQR from the upper and lower quartile. Individual points represent data from a single participant, and gray lines connect data points from the same participant. f Impact of discontinuous feedback on learning dynamics. Comparison of average BCI online performance (%) with the continuous (green, n = 15, same as b vs discontinuous real-time feedback (purple, n = 10). Error bars in (b, e, f) indicate the standard error of the mean.

Back to article page