Computational protocol: Task-dependent neural representations of salient events in dynamic auditory scenes

Similar protocols

Protocol publication

[…] EEG signals were down sampled to 512 Hz and lowpass filtered at 104 Hz using a Decimator software provided by Biosemi, and then further preprocessed using FieldTrip (Oostenveld et al., ), MATLAB (MATLAB Release 2013a, the MathWorks, Inc., Natick, Massachusetts, United States), and EEGLab (Delorme and Makeig, ). Each data segment included 1065 ms (duration of 5 music notes) immediately before or after the deviant point in the sound stream. Data were re-referenced to the averaged mastoid electrodes as the reference, and a de-mean was applied to each segment to align the average at zero. A denoising procedure was applied by first bandpassing all signals between 0.7 and 30 Hz. Bad channels were marked by an experienced experimenter and flagged as channels with unusual high frequency noise or large drifts. They were replaced by the arithmetic mean of the surrounding channels. No more than two bad channels were replaced for each participant. Eye artifacts were removed by excluding correspondent one or two eye-blink components and one eye-movement component after ICA (Independent Component Analysis) decomposition. Finally, a second 0.7–30 Hz bandpass filter was applied to smooth over minor distortions occasionally introduced by ICA reconstruction. In order to ensure that trial onset effects were not contaminating our analysis, we only analyzed 320 trials out of 400 experimental trials which contained more than one second of the auditory stream.After data preprocessing, ASSR and ERP responses were analyzed separately. ASSR was derived by first concatenating all trials in each condition for each participant over the 5 notes duration before and after a salient change. A Fourier transform was then applied to obtain the spectral distribution. The amplitude at 4.7 Hz (1/213 ms) was calculated to capture the stimulus-locked steady-state neural response (Figure 3A). The ASSR amplitude was confirmed by verifying that a peak at 4.7 Hz was greater than the averaged energy at surrounding frequencies across all participants and conditions (2–6 Hz excluding 4.7 Hz served as baseline). The ASSR analysis was derived from 24 central electrodes (Cz group) around the vertex as shown in Figure (including electrodes A1 (Cz), A2, A3, A4, B1, B2, B19, B20, B21, B32, C1, C2, C11, C22, C23, C24, D1, D2, D13, D14, D15, D16, D17, D18 in the layout of 128-channel Biosemi headcap). The lateralization of ASSR was calculated by averaging ASSR peaks in the 56 left and 56 right electrodes excluding the midline electrodes.Event-related potentials (ERPs) were analyzed by averaging all trials over the duration of two music notes (i.e., 426 ms) immediately before or after the deviant point with baseline correction using 100 ms prior to the segments. This analysis was called 1-note analysis in the results section in order to highlight that the focus is on the early and late components relative to the onset of 1-note. Because of the closeness in timing between successive music notes in the stimulus (SOA 213 ms), the analysis of the late ERP component of a given note was contaminated by onset effects of the following note. The ERP waveform at 0–213 ms was therefore subtracted from the ERP waveform at 213–426 ms in order to analyze any putative late components, including a P300 component. This waveform correction for analysis of the P300 component was repeated using other forms of correction (no correction, or by subtracting onset effects of a standard note instead of deviant one). Though different forms of correction resulted in slightly different shapes of the corrected waveform and peak latencies, all methods of a putative P300 analysis resulted in the same statistical results in terms of significant changes in the P300 component as a function of top-down attention and acoustic saliency.The analysis based on 1-note required averaging across fewer trials (320 trials) and was repeated using an average of three consecutive music notes (3-notes) with more data (960 trials). In the 3–note analysis, data from 3 notes was averaged then baseline-corrected using 100 ms prior to the onset of the averaged segments. In order to avoid overlapping between the auditory stream and salient change, data starting from the 2nd last music note before the salient event served as standard and the 1st music note after the salient event served as deviant for the 1-note analysis. The data starting from 2nd, 3rd, and 4th last music notes before the salient event served as standard and the 1st, 2nd, and 3rd music notes after the salient event served as deviant for the 3-note analysis. The enlarged N1 was prominently visible only in the 1-note analysis and concealed changes in the positive P1 peak and the presence of a frontal MMN-like negativity which were only revealed with the 3 note analysis. The late P300 component was consistent across the 1-note and 3-note analysis.ERP components were defined as follows: (i) An early P1 positivity was defined as a positive peak in standard and deviant ERP curves over the range 30–100 ms from the central Cz electrode groups. We defined the peak latency of the component as the average peak latency across all participants. Once a positive peak was identified, the amplitude over a window of ±15 ms around this peak was averaged and checked for statistical significance relative to the variance in the data. The peak latency for the P1 component was found to be 64 ms for all 3 experiments (1-stream attend, 2-stream attend and 2-stream ignore). The central topography of the P1 peak was confirmed by full head topography to verify that the maximal positivity is indeed localized in the central electrodes. (ii) An early negativity (N1) was analyzed similarly as a significant negative trough in both the standard and deviant ERP curves over the time range 100–213 ms in the Cz electrode group, with analysis window ±15 ms around lowest negativity. The peak latency for the N1 component was found to be at 156 ms for 1-stream attend, 176 ms for 2-stream attend and 166 ms for 2-stream ignore. (iii) An MMN-like component was defined as a significant negativity over 100–213 ms in the difference curve between standard and deviant in the frontal electrodes from the Fz electrode group as shown in Figure (electrodes B30, B31, B32, C2, C3, C4, C11, C12, C13, C20, C21, C22, C23, C24, C25, C26, D2, D3, D4, D11, D12, D13). The frontal distribution of the MMN-like component was confirmed by full head topography with either the averaged mastoid reference or the nose reference. The peak latency for this component was found to be at 141 ms for 1-stream attend, 197 ms for 2-stream attend and 182 ms for 2-stream ignore. (iv) A late P300 positivity was defined as a significant positive peak in the difference curve over the time window 300–426 ms in the central Cz electrodes (corrected by ERP curves of the following note at 0–213 ms, as explained earlier). The peak latency for the P300 component was found to be at 360 ms for 1-stream attend, 410 ms for 2-stream attend and 410 ms for 2-stream ignore.The complex nature of the variables manipulated in the current study required an across-experiment strategy without a full factorial design. Therefore, we conducted a number of statistical tests to alleviate concerns of comparisons across pools of subjects in different experiments. A One-Way ANOVA was performed on the behavioral results in order to detect any differences of performance across experiments. For the analysis of neural responses using ASSR and ERPs, we first conducted pairwise T-tests to investigate the effects of the salient event. Since bottom-up attention was one of the most important factors in our study, we reported the effect of salient change in each group as well with Bonforroni correction on multiple comparisons. Apart from the effect of salient change, we also examined the influences of (1) top-down attention or (2) auditory scene complexity on: (a) the auditory stream before the salient change by a univariate analysis on ASSR or MANOVA on ERPs and (b) the increase introduced by the salient change by looking at the interaction between saliency and attention or scene complexity in the Two-Way ANOVA. We also compared the lateralization of auditory stream and the salient change by using pairwise T-tests. […]

Pipeline specifications

Software tools FieldTrip, EEGLAB
Application Clinical electrophysiology
Organisms Homo sapiens