Cortical alpha oscillations in cochlear implant users reflect subjective listening effort during speech-in-noise perception

Abstract

Listening to speech in noise is effortful for individuals with hearing loss, even if they have received a hearing prosthesis such as a hearing aid or cochlear implant (CI). At present, little is known about the neural functions that support listening effort. One form of neural activity that has been suggested to reflect listening effort is the power of 8-12 Hz (alpha) oscillations measured by electroencephalography (EEG). Alpha power in two cortical regions has been associated with effortful listening-left inferior frontal gyrus (IFG), and parietal cortex-but these relationships have not been examined in the same listeners. Further, there are few studies available investigating neural correlates of effort in the individuals with cochlear implants. Here we tested 16 CI users in a novel effort-focused speech-in-noise listening paradigm, and confirm a relationship between alpha power and self-reported effort ratings in parietal regions, but not left IFG. The parietal relationship was not linear but quadratic, with alpha power comparatively lower when effort ratings were at the top and bottom of the effort scale, and higher when effort ratings were in the middle of the scale. Results are discussed in terms of cognitive systems that are engaged in difficult listening situations, and the implication for clinical translation.

Fig 1. Experimental design.

A) Participants were presented with triplets of digits spoken by a female talker at the front-facing speaker, while multi-talker babble noise of varying SNR was presented in the seven peripheral speakers. B) During EEG recording, trials of the digit triplet test were presented, with SNR varying on each trial. After the stimulus presentation ended, participants reported the digits they perceived as well as their listening effort rating on a 1–10 verbal report scale.

Fig 2. Behavioural data.

A) Correlation between CNC Word Score (a clinical test of word recognition) and performance on the task. The line represents the least squares fit. “Rho” indicates Spearman’s correlation coefficient. B) Performance predictions from the logistic mixed effects model plotted against individual trials (grey dots). Note that dots are randomly jittered around each effort level and performance outcome for visibility. The fitted line represents model predictions of performance from subjective effort, with the orange shaded region bounding the 95% confidence interval. C) The same as panel B, but plotting performance as a function of SNR.

Fig 3. Single-subject effort–SNR correlations.

Scatterplots for each participant showing the relationship between the trial SNR and effort ratings. Spearman’s rank correlation coefficients are provided below each plot with significance values. All lines represent the least squares fit. Correlations for all participants were significant, and survived false discovery rate correction. Note that values are jittered around each effort level and SNR level for visibility.

Fig 4. Parietal EEG alpha power as a function of listening effort in CI users.

A) The three time-frequency plots depict wideband power (expressed as a dB change relative to baseline) averaged across low, medium, and high effort rating groupings in seven posterior channels (channel locations plotted adjacent to the top plot). These bins, though unbalanced between categories, were chosen for the sake of illustration. Red shading indicates higher power compared to baseline, and blue shading represents lower power. Above the time-frequency plots is a schematic describing the acoustic waveform across the trial structure, including the babble noise and digits. Note that effort groupings are for visualization, as single trials were the basis for inferential analysis. Further, these plots are an average of time frequency representations across participants, where trials were averaged before time frequency decomposition. Single-trial time frequency representations were submitted to mixed modeling. B) Topographic maps showing alpha power averaged over the 3- to 6- second time window. The left column is a top view of the scalp, and the right column is the right view of the scalp. C) The upper panel plots the inverted U function between listening effort ratings and alpha power (z-scored). The grey dots represent averaged alpha power at each effort rating level. The black line and orange shaded area are the regression lines from the single-trial analysis, bounded by 95% confidence intervals. The lower panel plots the same regression line and confidence intervals as in the upper panel, but the grey dots are z-scored alpha power for each trial. Note that individual trials are jittered around each effort level for visibility.

Fig 5. Left IFG alpha power as a function of listening effort in CI users.

A) Same as Fig 4A; the location of left IFG is shown adjacent to the top time-frequency plot. B) The upper panel plots the function between listening effort ratings and alpha power. The grey dots represent an average of z-scored alpha power values per each effort rating level. The black line and orange shaded area are the regression lines from the single-trial analysis, bounded by 95% confidence intervals. The lower panel is the same as the upper panel, but the grey dots represent z-scored alpha power values for individual trials. Note that the regression line and confidence intervals are identical between the upper and lower panels. Note that individual trials are jittered around each effort level for visibility. n.s. = not significant.