Figures and data
Experimental setup and Algorithm for N1 extraction from Temporal Response Functions (TRFs).
A. Neurofeedback training was divided into Baseline blocks (pre- or post-training measures) and Feedback blocks (with neurofeedback or sham feedback, depending on the group). During all blocks, participants were instructed to pay attention to a target audiobook, presented at 0°). After each 22-s segment of speech, participants selected one out of four four-word phrases that they believed was part of the preceding segment. A distractor audiobook was presented at 180° in all Feedback trials, and in half of the Baseline trials. Only in Feedback blocks, feedback was presented to the participant in the form of two visual bars, one for the Target and the other for the Distractor. Participants were instructed to increase the displayed value for the Target and reduce it for the Distractor, and instructed that they can achieve this goal by paying attention to the target audiobook, and by ignoring the other audiobook, respectively. Prior to neurofeedback training, participants completed a short volume calibration (not shown). B. During each 22-s audio segment, EEG data was collected and processed by a Backend computer to extract TRFs estimated from neural responses to target and distracting audiobooks. The N1 component was then extracted separately for each audiobook and converted to feedback (see Online Data Processing) through a Presentation Computer. C. To estimate individual time ranges for the TRF’s N1 component, the largest negative peak was identified first within a search window (80-150ms) and the nearest positive peaks (>10ms) preceding and following the negative peak, respectively. The individual N1 time range was defined as ±25% of the distance between the negative peak and the positive peak on each side, yielding a window centred on the negative peak. N1 time ranges were identified separately for each audiobook (male/female) speaker and condition (Target / Distractor). For each 22-s trial, feedback was calculated as the average N1 in the individual time range, expressed as a percentile relative to the previously obtained values (see Online Data Processing). For illustration, the TRFs shown were obtained from four pilot participants that did not receive neurofeedback and were not included in the final data analysis.
Overview of speech comprehension and TRF results.
A. Performance in the speech comprehension task. Circles show results from individual participants, bars show the group average, and error bars the standard error of the mean (SEM). B,C. TRFs estimated from responses to the derivative of the speech envelope at the EEG channels used for neurofeedback (inset). B. TRFs in the absence (“single”) vs presence (“multi”) of a distractor in Baseline blocks. C. TRFs estimated for target and distracting speech (Feedback trials and “multi” trials from Baseline blocks). Shaded areas indicate the standard error of the mean (SEM) and grey lines highlight the timepoints where the TRFs are significantly different from each other. Topographies of the most prominent components (P1, N1, P2) from the TRFs are shown as insets and were extracted individually for each condition and component (Single Speaker [P1-39 – 47ms, N1-89 – 97ms, P2-156 – 164ms] from B, Target [P1-46 - 54ms, N1-105 – 113ms, P2-179 – 187ms] and Distractor [P1-42 – 50ms, N1-97 – 105ms, P2-148 – 156ms] from C).
Neurofeedback increases the N1 component in the EEG response to target speech.
A.T-values from main effects of group (neurofeedback vs sham feedback) and block type (feedback vs baseline) on N1 amplitudes, as well as their interaction in a linear mixed model. The interaction is significant in a cluster comprising the three EEG channels shown by asterisks. A positive interaction term indicates that the increase in N1 negativity from baseline to feedback blocks is greater in the NF group than in the SF group. B. N1 amplitude values averaged across the three channels in the significant cluster (Panel A) relative to the pre-training baseline (subtraction within participants). Error bars indicate the SEM. C. TRFs estimated from channel FCz, averaged across the three Feedback blocks. Shaded areas around the mean indicate the SEM. Grey lines illustrate the time window used for statistical analysis.
Increased N1 amplitudes in the EEG response to target speech after neurofeedback (during Baseline blocks).
A.T-values from main effects of group (NF vs SF) and time (pre-training vs post-training) on N1 amplitudes as well as their interaction in a linear mixed model. This interaction is significant in a cluster comprising the fifteen EEG channels shown by asterisks. A positive interaction term indicates that the increase in N1 negativity from pre-training to post-training baseline is greater in the NF group than in the SF group. B. Difference (post-minus pre-training Baseline) in N1 amplitude values, averaged across the fifteen channels in the significant cluster (Panel A). Error bars indicate the SEM. C. TRFs estimated from channel Iz. Shaded areas around the mean indicate the SEM. Grey lines illustrate the time window used for statistical analysis.
Only in the NF group, increased N1 amplitudes during feedback are associated with better speech comprehension.
A,B. Correlation between the average N1 amplitude and speech comprehension performance during Feedback blocks (both relative to pre-training Baseline) in the NF group (A) and SF group (B). N1 amplitudes were computed in the three-channel cluster that showed reliable neurofeedback effects (Fig. 3A). The contour lines indicate the bootstrapped Mahalanobis distance from the bivariate mean in steps of six squared units (Brighter colours denote greater distances). Filled circles denote data included in Shepard’s Pi correlations, open circles denote outliers. The solid line is a linear regression over the data after outlier removal. C. The small topographies show correlation coefficients, computed equivalently to panels A,B, but for all 64 EEG channels. The large topography shows t-values from the contrast in correlation coefficients between neurofeedback and sham feedback groups. The contrast is significant in a cluster comprising the six EEG channels shown by asterisks. Positive T-values indicate that a negative correlation between N1 amplitude and speech comprehension performance is stronger in the NF group than in the SF group.
Control analyses.
A,B. Reliability of single-trial N1 values. Single-trial N1 amplitudes from TRFs (multiplied by -1 and relative to a baseline) were estimated for target (A) and distracting speech (B), and their average was divided by their trial-to-trial variability. Circles show outcomes from individual participants, bars and error bars show the mean and SEM across participants, respectively. C. Feedback vs actual N1 values in the sham feedback group. Average distance between displayed feedback values (as percentiles) in the SF group (C) and those that would have been displayed during genuine neurofeedback. Shaded areas represent the SEM.
Increase in P1 amplitude during Feedback only for the neurofeedback group
A. T-values from main effects of group (NF vs SF) and block type (feedback vs baseline) on P1 amplitudes, as well as their interaction in a linear mixed model. This interaction is significant in a cluster comprising the nine EEG channels shown by asterisks. A positive interaction term indicates that the increase in P1 positivity from baseline to feedback blocks is greater in the NF group than in the SF group. B. P1 amplitude values averaged across the nine channels in the significant cluster (Panel A) relative to the pre-training baseline (subtraction within participants). Error bars indicate the SEM.
Neurofeedback does not affect the N1 component in the EEG response to distractor speech.
A. T-values from main effects of group (neurofeedback vs sham feedback) and block type (feedback vs baseline) on N1 amplitudes, as well as their interaction in a linear mixed model. A positive interaction term indicates that the increase in N1 negativity from baseline to feedback blocks is greater in the NF group than in the SF group. B. N1 amplitude values averaged across the nine channels used for feedback, relative to the pre-training baseline (subtraction within participants). Error bars indicate the Standard Error (SEM). C. TRFs estimated from channel FCz, averaged across the three Feedback blocks. Shaded areas around the mean indicate the SEM. Grey shaded area indicates the time window used for statistical analysis.
