Scale-(In)variance in the EEG: A Concert of Scale-Free and
Scale-Dependent Processes Underlie Perception and
Processing of Complex Music Stimuli
Thomas Brockmeier, Master Student Brain & Cognitive Sciences: Cognitive Neuroscience,
Universiteit van Amsterdam
Student Number VU: 2566337; UvA: 10102957 Internship Final Report, 36 ECTS Semester 1, Academic Year 2014–2015
Supervised by: Filipa Texeira Borges, Mona Irrmischer, Dr. Klaus Linkenkaer-Hansen
Abstract
Music is a near universal product of human culture and mind, and—like many phenomena that arise from nature, cognition, and behavior—its structure can be quantified by power-law distributions. The emergence of power-law scaling is an important indicator of the presence of a (near) critical state in the underlying network that governs it. Neuronal mass activity in the cortex adheres to power-law distributions as well; whereas changes in these distributions their exponents have been found to correlate with cognitive and behavioral measures. The questions how music perception and subjective music experience affect neuronal mass activity in the cortex—and scale-free dynamics found herein—however, have remained unanswered. These were addressed in the current study by obtaining ongoing EEG activity was obtained from 28 participants who first rested with their eyes closed, and subsequently listened to 12 pieces of music selected for their scaling properties. It is found that changes in power-law scaling behavior of elicited neural activity correlate with cognitive measures of focus and familiarity. Furthermore, it is found that these effects occur both independently from, as well as in tandem with changes in spectral power in classical frequency bands. These results indicate that long-range temporal correlations and spectral power are distinct phenomena that can interact, but do not do so by default—suggesting that distinct properties of neural activity work in concert to orchestrate perception, cognition, and behavior.
Keywords: Criticality, EEG, Neural Oscillations, Music Cognition
Acknowledgements
Filipa Teixeira Borges, Mona Irrmischer, Thomas Brockmeier and Klaus Linkenkaer-Hansen designed research. T.B., M. I. and F.T.B. performed research. T.B. and F.T.B. analyzed data and Figures 4 and 5 were produced by F.T.B.. T.B. wrote the manuscript with comments from M.I. and K. L.-H. A part of the introduction and Methods was written by F.T.B. If the data in this report are to be used for publication, Thomas Brockmeier will be granted co-authorship.
Introduction
In work conditions, positive effects of music on performance have been found (Lesiuk, 2005), but little is known about the underlying neurocognitive mechanics. Furthermore, without a demanding or interesting task, our mind seeks additional entertainment (Mason et al., 2007). Whereas mind wandering is generally disruptive to other activities, music listening usually is in the background and unrelated to the task at hand. Experience sampling has shown that mind wandering (Killingsworth & Gilbert, 2010) or listening to music (Sloboda et al., 2001) both occupy more than 40% of our time.
Studies of different music excerpts revealed that musical pitch (Hsü & Hsü, 1990; Su & Wu, 2007), loudness (Voss & Clarke, 1975), and musical rhythms (Levitin et al., 2012) obey 1/fα power-law scaling in the power spectrum with an α exponent ranging from approx. 0.5–1. This hallmark in the temporal organization of music could arise from a necessity in balancing predictability (highly correlated structure) and surprise within the music flow in order to create an aesthetic experience and hence, an emotion. Indeed, when contrasting compositions in which the frequency and duration of each note either follow a 1/f, 1/f2 and white noise spectrum, those characterized by 1/f are judged the most pleasing (Voss & Clarke, 1978).
Systems that operate at, or near, a critical level have been shown to express behavior that adheres to the power-law distribution described above (e.g., Bak et al., 1987; 1988; Lux & Marchesi, 1999). Power-law distributions occur widely in nature (see Clauset et al., 2009 for several examples), cognition (Kello et al., 2010), and—as noted— music: an (almost) universal product of the human mind. Furthermore, critical-state dynamics have been found to be an emergent property of (neural) networks (Barabási &
Albert, 1999; Poil et al., 2012), and the cortex has been shown to operate at a (near) critical level, with its activity exhibiting scale-free dynamics that adhere to a power-law distribution (Priesemann et al., 2014; Scott et al., 2014). The functional significance of this notion is that information transmission and storage, and computational strength are thought to be optimal in systems that operate at their critical point (Beggs, 2008), but it remains unclear how cognitive processes are reflected in the critical dynamics observed in cortical activity.
Resting-state neuronal oscillations exhibit scale-free properties in the amplitude modulation of cortical oscillations akin to those found in music (Linkenkaer-Hansen et al., 2001). Only recently, we have begun to discover how individual variation in scale-free neuronal oscillations relate to individual differences in cognitive (Palva et al., 2013) or behavioral (Smit et al., 2013) performance. MEG research has indicated that brain responses can better match the statistical properties of tone sequences that obey music-like statistical properties (Patel & Balaban, 2000); however, whether people with high scaling exponents in their ongoing oscillations can better match music with high scaling exponents—or vice versa—remains unknown. Furthermore, it is not known whether an individual’s ability to correlate his/her brain activity to the musical dynamics is predictive of behavioral responses. These two problems, as well as a potential link between cognition and criticality in the brain will be addressed in this paper.
Materials and Methods
Participants: Participants were 12 female and 16 male (mean age: 26.8 years, S.D.: 4.17 years; 27 right handed, 1 left handed) healthy members of the general population. They provided written informed consent before participating in the experiment. Participants
who were diagnosed with any psychiatric disease, or who reported use of any psychotropic medication or (illicit) drugs were excluded from participating in the study.
Eyes-closed rest paradigm: An eyes-closed rest (ECR) paradigm provided a baseline measure to which the subsequent music presentation trials could be compared. A standardized task (Diaz et al., 2013) was used to maintain congruency with previous research at the institute. Participants were asked to sit in front of a computer, and given the following instruction: “For this part of the experiment it is required that you are able to sit relaxed with your eyes closed during a period of 3 minutes. Once the resting session has ended, you will be notified by a beep. You can stop the sound by clicking ‘Stop.’ Afterwards you can click on ‘Next’ to proceed to the questions. Please relax and try not to fall asleep.” (Diaz et at., 2013). After the rest the following instructions were provided: “The 3 minute rest is over. Now several statements will follow regarding potential feelings and thoughts you may have experienced during the resting period. Please indicate the extent to which you agree with each statement.” (Diaz et at., 2013). The end of the ECR period was signaled by a beep; participants were asked to keep their eyes shut until they heard this auditory cue. After the ECR period, the participants were presented with a standardized questionnaire (Amsterdam Resting-State Questionnaire (ARSQ), see Diaz et al., 2013) to collect information about the way in which they had experienced the ECR. After completing the ECR paradigm, participants moved on to the music phase of the experiment.
Music paradigm: The music paradigm was developed to expose participants to music fragments with different degrees of statistical predictability as to how they unfolded over time—with respect to fluctuations in rhythm, pitch, and loudness—as well as to obtain
subjective ratings on how the presented music pieces were perceived.
12 distinct 120-second segments were presented to each participant in a randomized order, followed by 3 repetitions—again in random order. See Table A for stimulus details. The repeated stimuli were included to test for potential fatigue and familiarity effects. The following instructions were provided before starting the experiment: “For this part of the experiment it is once again required that you are able to sit relaxed with your eyes closed. You will be presented with a fragment of classical music on each trial, which will last for two minutes. Please keep your eyes closed throughout the trial and concentrate on the music. You will again hear a beep after the music has finished. Please open your eyes only after hearing this tone, and answer the questions about the music. Please press ‘Submit’ and close your eyes again when you have finished answering the questions. We will first start with a practice trial, to familiarize you with the procedure. Please use this trial to adjust the volume to your liking as well. Hereafter, 15 experimental trials will follow, with short breaks in between. The experiment will end with a short questionnaire regarding your musical taste and experience.” A visual cue for the participant to close his or her eyes was presented on-screen. 2 seconds after the cue a stimulus was played, followed by an auditory cue to open his or her eyes again. After hearing each segment, participants were asked to rate their enjoyment of, familiarity with, and concentration on the music on 7-point Likert scales (ranging from 0: ‘not enjoyed’/’not familiar’/’not focused at all’ to 6: ‘very enjoyable’/’very familiar’/’very focused’). Subjects were also asked to indicate whether or not they had kept their eyes closed on a 2-point scale (‘had’/’had not opened eyes’). Participants were familiarized with the procedure beforehand, using a music fragment that was not used again in the
experimental phase of the paradigm. Finally, all participants were asked to fill out a questionnaire examining their experience with, and taste in music (Appendix A).
All visual cues, instructions, and questionnaires were presented on a computer screen. Auditory cues and music stimuli were presented over KRK Rokit 8 RPG 2 studio monitors. Stimulus presentation and acquisition of behavioral data was done using custom scripts in the OpenSesame environment (Mathôt et al., 2012).
Music stimuli selection: Long-range temporal correlations (LRTC) in rhythm, pitch, and loudness fluctuations were quantified using detrended fluctuation analysis (DFA). DFA is a scaling analysis technique used to estimate LRTC of power-law form, which provides the scaling exponent of a given signal as its output. Exponents between 0.5 < α < 1 indicate the presence of these long-range temporal correlations in the input signal (Peng et al., 1995; Hardstone et al., 2012). Previous research on scale-free features of music have used linear fits to loglog spectral power (Voss & Clarke, 1975; Hsü & Hsü, 1990; Su & Wu, 2007; Levitin et al., 2012), which give an exponent (beta) that is related to the DFA exponent (alpha) through the relationship alpha = (1 + beta)/2 (Rangarajan and Ding 2000).
The rhythm and pitch analyses were performed for all solo piano scores contained in the Humdrum Kern database (Huron, 2010) that exceeded 200 separate note onsets (Levitin et al., 2012) and were at least 2 minutes in length, and Mozart’s sonata for two pianos in D Major, K448 (cf. Rauscher et al, 1993). Three bins were created for low (0.5 < α < 0.65), middle (0.65 < α < 0.85), and high (0.85 < α < 1) exponent values. Because of limited presence of pieces with rhythm exponents that fit both into the ‘high’ bin and combined with ‘high’ pitch values, the rhythm
bin was extended to include two segments with a rhythm exponent of α = 1.06.
Based on these results, 12 compositions were selected in such a way that a variety of different rhythm and pitch scaling exponent combinations existed within the stimulus pool. Where possible, low, medium, and high rhythm exponents were combined with low, medium, and high pitch exponents, leading to different combinations of rhythm and pitch scaling being present in the stimulus pool (e.g., a piece with high rhythm DFA and low pitch DFA, one with high rhythm DFA and medium pitch DFA, and one with high rhythm DFA and high pitch DFA, and so on). As a further restriction to ensure stylistic diversity, every selected composer was present with two segments of different combinations of rhythm and pitch exponents. See Table A for further information concerning the stimuli.
Recorded performances of the 12 pieces were obtained from Qobuz.com in CD-quality (lossless, 16-bit, 44.1 kHz). DFA was performed on the left and right channels of these recordings—as well as on an averaged mono signal—to analyze scaling behavior of the loudness envelopes. The stimuli were created by extracting the first 2 minutes of music from the recordings using Audacity, equalizing these at a 65 dB output level (Mitterer, 2014), and adding 0.5 second long fade-in and fade-outs to prevent speaker popping—both using Praat (Boersma & Weenink, 2001). Measures of loudness DFA were obtained by running the DFA algorithm (see below) on the loudness envelopes of the recordings. This was performed post-hoc, as loudness fluctuations within a single performance cannot be determined from the score.
EEG data acquisition, preprocessing, and statistical analysis: EEG data were recorded at 1000 Hz using an EGI Geodesic EEG system with HydroCel sensor nets consisting of 128 Ag/AgCl electrodes.
Impedances were kept below a 50–100 kΩ range. Electrodes were referenced to Cz at the time of recording, and later re-referenced to a common average. Preprocessing of the EEG signals and statistical analysis were performed using the Neurophysiological Biomarker Toolbox (NBT, http://www.nbtwiki.net/; see Hardstone et al., 2012) for Matlab (Mathworks).
The continuous recordings were epoched in 16 segments comprising a segment of 178 seconds of eyes-closed resting state and 15 segments of 108 seconds of the different music pieces presented. The first 2 seconds of all segments were clipped while epoching (minimizing possible amplifier saturation effects). Next, bad channels of each segment were detected using the FASTER approach, which excludes channels based on a threshold of +/- 3 z-scores of any of the 3 parameters (channel’s correlation with its neighbors, variance and Hurst exponent; see details in Nolan et al., 2010). Bad channels were reconstructed by interpolating from neighboring electrodes using EEGLAB’s spherical spline interpolation function. To mitigate transient artefacts, each segment was epoched in multiple sub-epochs of 250 ms and sub-epochs with high epoch’s amplitude range, variance or difference from the epoch’s mean were removed. This procedure excluded on average about 5% of the original segment’s length.
Both amplitude and DFA analysis were done in the following frequency bands: delta (1–4 Hz), theta (4–8 Hz), alpha (8–13 Hz), beta (13–30 Hz), gamma (30–45 Hz), high-gamma 1 (55–125 Hz), and high-gamma 2 (60–90 Hz). The band-pass filtering used finite impulse response filters with a Hamming window, and filter orders equal to 2000 (delta-band), 500 (theta-band), 250 (alpha-band), 154 (beta-band), 67 (band) and 33 (high gamma-band) samples.
Time-averaged power was calculated by applying the Welch’s modified periodogram method implemented in Matlab as pwelch() function with non-overlapping Hamming windows of 1 second. The results are displayed as amplitudes (square root of the power spectrum obtained).
The instantaneous amplitudes of the fluctuations were calculated using the magnitude of the analytic signals quantified with resource to the Hilbert transform on the previously band-passed signals. Next, monofractal scaling exponents were estimated using DFA (Peng et al., 1994, 1995)—a well-established technique first used for the amplitude dynamics of neuronal oscillations by Linkenkaer-Hansen and colleagues (2001). Details of the method are described elsewhere (Peng et al., 1994; Kantelhardt et al., 2001; Hardstone et al., 2012). The exponent of the decay of temporal correlations was quantified within the range of 1–20 seconds for all segments of resting state or music. Throughout the report, amplitude or DFA exponent are presented for each channel averaged across all 28 subjects and averaged across subjects and across the 12 non-repeated music pieces.
The distributions of amplitudes and DFA exponents (delta, theta, alpha, beta, gamma and high-gamma bands) of resting-state and music trial EEG recordings were compared using Wilcoxon signed rank sum tests (significance threshold 0.05), a nonparametric test which makes no assumption about the distribution. No multiple comparison correction method was applied, instead the differences between the median amplitudes or DFA exponents of music and resting-state were deemed significant using binomial testing (Maris and Oostenveld, 2007). Differences were considered significant if at least 10 channels have a p-value below 0.05.
DFA on music scores was calculated in windows ranging from 0.5 seconds to the entire length of the signal with 50% overlap between windows. The DFA exponent was obtained by fitting from a minimum of 1 second to a maximum of 60 seconds of the entire signal, depending on the length of a given piece.
Behavioral responses were correlated using Pearson's linear correlation coefficient, whereas correlations between behavioral responses and EEG data were computed using Spearman’s rank correlation coefficient. Group differences in EEG data were assessed using Wilcoxon signed-rank test due to the non-parametric nature of the data.
The study was approved by the Ethics Committee of the Faculty of Psychology and Education at the VU University Amsterdam. Results
In order to study the effect of music on human brain activity in general and its scale-free properties in particular, we selected music pieces with varying strengths of LRTC as determined by rhythm, loudness or note-interval fluctuations (Fig. 1; Table A).
LRTC in music. Further analysis of selected compositions revealed the consistent presence of LRTC in unfolding musical rhythms, harmonic progressions, and loudness contours. This is represented by the observed log-log linear increase of fluctuation present in a given curve with increasing window size. When this line has a slope between 0.5 < α < 1.0, the presence of these dependencies can be confirmed (Fig. 1, harmonic progressions shown; see Appendix B for all music analysis results). Power spectra suggesting 1/f scaling in rhythm and harmony contours disappeared after the curves were shuffled, indicating that the observed scaling behavior is the result of the way music was ordered, rather than the relative frequencies in which note lengths or
intervals were present in the signals (data not shown).
Behavioral feedback on the music. Participants reported moderate familiarity with classical music (mean = 3.42, S.D. = 1.45) in the exit questionnaire. Judgments of pleasure derived from, familiarity with, and focus on stimuli were found to be independent of music DFA results (Fig. 2). Pleasure and familiarity show a highly significant positive correlation (R = 0.18, N = 28, p < 0.001; Fig. 3), as do focus and pleasure (R = 0.59, N = 28, p < 0.001; Fig. 3), and familiarity and focus (R = 0.10, N = 28, p < 0.05; Fig. 3).
Differences in electrophysiological activity between task and ECR conditions: An increase in alpha activity over somatosensory areas, and left parietotemporal cortex can be observed in music conditions compared to ECR (approx. + 0.2 µV, N = 28, p < 0.05; Fig. 4), as well as an increase in theta power in the auditory regions (approx. + 0.2 µV, N = 28, p < 0.05; Fig. 4).
High frequency (> 30 Hz) power-law exponents decreased in experimental trials (approx. Δα = -0.1, N = 28, p < 0.05; Fig. 5), as compared to ECR, whereas the low frequency (< 8 Hz, and most notably < 4 Hz) exponents increased (delta: approx. Δα = 0.05, N = 28, p < 0.005, theta: approx. Δα = 0.03, N = 28, p < 0.05; Fig. 5). This effect is visible across the scalp in all mentioned frequency bands. No differences in power-law exponents of intermediate frequency bands were observed between music and ECR trials.
Correlations between behavioral responses and electrophysiological data: Amplitude in the beta range showed a positive correlation with familiarity judgments over frontoparietal regions (0.1 < R < 0.2, N = 28, p < 0.05; Fig. 6). Furthermore, changes in gamma-power showed a similar correlation (0.1 < R < 0.2, N = 28, p < 0.05; Fig. 6) over parietotemporal regions.
Amplitude in the alpha band showed a positive correlation with focus across the scalp, apart for over bilateral primary auditory cortices, and several central and frontal areas (0.1 < R < 0.2, N = 28, p < 0.05; Fig. 7).
Changes of scaling exponents in frequency bands from theta upwards showed a significant negative correlation with focus ratings (-0.1 < R < -0.2, N = 28, p < 0.05; Fig. 8). The effect was visible across the whole scalp in alpha, low gamma, and high gamma 1 bands, whereas the effect disappeared over posterior areas in the theta band, over the right parietal cortex in the beta band, and over areas approximately surrounding Cz in the high gamma 2 band.
No significant effects were found for correlations between LRTC in the rhythm, pitch, and loudness dimensions of the music— and the mean of these values—and LRTC in the EEG when listening to this music (-0.1 < R < 0.1, N = 28, p > 0.1; Fig. 9). Neither was a consistent pattern of significant effects found for correlations between pleasure and LRTC in the EEG (-0.1 < R < 0.1, N = 28, p > 0.1; Fig. 10). Discussion
Music analysis results were in line with a long-standing line of research showing the presence of LRTC in Western tonal music (Voss & Clarke, 1975; Hsü & Hsü, 1990; Su & Wu, 2007; Levitin et al., 2012). Behavioral results showed that pleasure, familiarity, and focus are all correlated to one another, with the relation between focus and pleasure being the strongest. The most salient effect in differential amplitudes between ECR and music conditions is the significant increase in alpha power over somatosensory cortices. This could reflect the suppression of task-irrelevant sensory input in the haptic domain (Jensen & Mazaheri, 2010; Foxe & Snyder, 2011). It was observed that spectral exponents of high-frequency cortical activity declined during task presentation in compliance with
previous literature (He et al., 2011; Fagerholm et al., in press; Irrmischer et al., unpublished data), and the opposing effect was visible in low frequency bands. The observed independence between spectral power in a given frequency band and the LRTC of the activity in the same range is striking, and corroborates prior research (Miller et al., 2009) as well. The functional significance of these three effects, however, has remained elusive.
While the hypotheses regarding the reflection of perceived scaling behavior of music in the brain have yet to be examined fully, the salient effects mentioned above will be discussed in the remainder of this article. Music Familiarity Correlates with Spectral Power, but not with LRTC
The observed correlation between familiarity judgments and beta power over frontoparietal regions and gamma-band oscillations over posterior regions matches previous literature concerning the mediation of top-down attention (Hipp et al., 2011), as well as in processes supporting the active maintenance and accumulation of sensory information in the brain (Siegel et al., 2011; 2012), and auditory working memory performance (Roberts et al., 2013).
The maintenance and accumulation of a sensory input exhibiting LRTC will intuitively lead to the formation of expectancies how it will progress. Evidence for these expectations has indeed been found: Vuust and colleagues (2009) for example show that a mismatch negativity event-related field component is elicited by stimuli that violate rhythmic expectancy, whereas Bharucha and Stoeckig (1986) show that a harmonic context primes the processing of contextually relevant chords, leading to ‘increased sensitivity (faster and more accurate reactions) to related targets relative to unrelated targets’. These results indicate that expectancies of the way in which
a presented—and, importantly, novel—music piece will progress are formed while the piece is being processed.
Expectations of how a composition will progress can be formed more readily when one is familiar with the piece at hand: familiar information can be held and processed in working memory more easily (Case et al., 1982; Lewellen et al., 1993). Familiarity therefore facilitates stimulus maintenance in working memory, which may be reflected in its observed correlation with observed beta and gamma-band effects. Spectral Power and LRTC of Neural Signals Reflect Distinct Processes
Declining LRTC in gamma band activity is associated with increased task-performance (He et al., 2011; Fagerholm et al, in press). The concurring increase of LRTC in delta-band activity, however, appears to be absent from the existing literature. Changes in delta power have traditionally been associated with the sleep-wake cycle, as well as with a wide array of developments disorders and other pathologies, the integration of cerebral activity with homeostatic processes, and—in the cognitive domain—with attention, salience detection, and subliminal perception (Knyazev, 2012).
Significant changes in delta power, however, have not been observed in the current study. This finding further solidifies the notion that LRTC and spectral power are distinct neural processes that underlie different functional operations. Further research will be required to examine the exact situations wherein this effect occurs, and what its functional significance is.
Spectral Power and LRTC Interact to Shape Behavior and Cognition
LRTC in cortical activity correlated negatively with how focused participants reported to be on music they heard in a given trial. These results are in line with studies examining focused attention meditation and behavioral performance in sustained attention tasks (Irrmischer et al., unpublished data).
Interestingly, focus and spectral power hardly correlate—it is the temporal structure of the EEG that changes. This change occurred over a widely distributed network of frontal, parietal, and temporal regions, as well as parts of the cingulate, and thalamic areas that has been implicated in the maintenance of focused attention (see Lutz et al., 2008 for a review). The observed effect matches the cortical distribution mentioned above best in the theta band, whereas higher frequencies show an even broader distribution.
While delta and theta band LRTC show an overall increase during music presentation, the negative correlation with focus suggests an attenuated increase in temporal coherence when an individual is more concentrated, rather than an absolute drop in LRTC. Activity in these bands thus becomes more coherent under influence of sensory input, but this increase is attenuated by focus. By moving less closely to a critical state—in favor of subcriticality—the brain reduces its sensitivity to being greatly disrupted by minor perturbations. The processing of sensory information is thus reflected in the overall increase of the scaling exponent, whereas the ability of an individual to maintain focus on this information is manifested in the tendency of the concentrating brain to stay in a more subcritical state than its easily distracted peers.
Lowered gamma-band DFA exponents show a decay of LRTC of signals in these ranges during stimulus presentation; the negative correlation between exponent value and focus thus suggests an amplification of this decay, leading to values implying near
randomness. The question why temporal coherence—i.e., the theoretical ability to process and transfer information—of high frequency activity—usually associated with local processing of information—diminishes when sensory input is provided, as compared to ECR is an interesting one.
Current research suggests that reduced LRTC in a focused cognitive task—and thus a lower dynamic range in the system— may reduce interferences that could affect task performance (Fagerholm et al., in press). Furthermore, the authors of this paper argue that the near-critical dynamics observed during rest may be advantageous, leaving the system sensitive to any relevant stimuli that may enter it. Whether small—task relevant— cortical regions do show near-critical activity during focused behavior continues to be an open question (Fagerholm et al., in press).
If this were the case, one may expect to see a larger suppression of local processing with increased concentration, reflected by a positive correlation between focus and alpha power (Jensen & Mazaheri, 2010; Foxe & Snyder, 2011). This effect is indeed visible across the scalp, achieving significance almost everywhere—except over bilateral auditory cortices, and some frontal and central locations. Furthermore, the areas where the anticorrelation between gamma band LRTC and focus does not reach significance appear to overlap with the areas where the correlation between increased alpha power and focus does not reach significance either. This suggests that information processing and transfer in high frequency bands could be more efficient in areas that show a decrease in alpha power. Further research will be required to investigate the interactions between these effects, which lies beyond the scope of the current project.
Conclusion
The current study does not only provide evidence that the amplitude and temporal structure of neural activity reflect distinct functional operations during a music-listening task; it furthermore suggests that these two measures interact to shape behavior and cognition. This finding sheds light on the existence of a newly uncovered dimension comprised of a symphony of interacting neural processes—one that only now starts to unfold before our eyes. The extent to which these different aspects of neural activity work together to shape neural states, orchestrating perception, cognition, and behavior remains an open question, demanding further investigation.
Outlook
The notion that different neural processes interact to shape cognition demands further investigation. Other measures of neural activity such as phase coherence and locking will have to be included in this line of research as well.
Further research is required to investigate the functional significance of:
- changes in delta band scaling behavior - the global decrease in coherence of gamma band activity across the cortex Prior research has shown neural activity mirroring pitch changes through phase tracking (Patel & Balaban, 2000). The current dataset requires further analysis to search for the presence of an analogous effect in our more realistic paradigm.
References
Bak, P., Tang, C., & Wiesenfeld, K. (1987). Self-organized criticality: An explanation of the 1/f noise. Physical review letters, 59(4), 381.
Bak, P., Tang, C., & Wiesenfeld, K. (1988). Self-organized criticality. Physical review A, 38(1), 364.
Ball, P. (2008). Science & music: facing the music. Nature, 453(7192), 160–162. Barabási, A. L., & Albert, R. (1999). Emergence
of scaling in random networks. Science, 286(5439), 509–512.
Beggs, J. M. (2008). The criticality hypothesis: how local cortical networks might
optimize information
processing. Philosophical Transactions of the Royal Society of London A: Mathematical, Physical and Engineering Sciences, 366(1864), 329– 343.
Bharucha, J. J., & Stoeckig, K. (1986). Reaction time and musical expectancy: priming of chords. Journal of Experimental Psychology: Human Perception and Performance, 12(4), 403.
Boersma, P., & Weenink, D. (2001). Praat, a system for doing phonetics by computer. Glot International, 5(9/10), 341–345.
Bullock, T. H., McClune, M. C., & Enright, J. T.
(2003). Are the
electroencephalograms mainly rhythmic? Assessment of periodicity in
wide-band time
series. Neuroscience, 121(1), 233–252. Case, R., Kurland, D. M., & Goldberg, J. (1982). Operational efficiency and the growth of short-term memory span. Journal
of experimental child psychology,33(3),
386–404.
Chialvo, D. R. (2006). The brain near the edge. arXiv preprint q-bio/0610041. Clauset, A., Shalizi, C. R., & Newman, M. E.
(2009). Power-law distributions in
empirical data. SIAM review, 51(4), 661–703.
Diaz, B. A., Van Der Sluis, S., Moens, S., Benjamins, J. S., Migliorati, F., Stoffers, D., ... & Linkenkaer-Hansen, K. (2013). The Amsterdam Resting-State Questionnaire reveals multiple phenotypes of resting-state cognition. Frontiers in human neuroscience, 7.
Fagerholm, E., Lorenz, R., Scott, G., Dinov, M., Hellyer, P., Mirzaei, N., Leeson, C., Carmichael, D., Sharp, D., Shew, W., Leech, R. (in press) Cascades and cognitive state focused attention incurs subcritical dynamics. Journal of Neuroscience
Foxe, J. J., & Snyder, A. C. (2011). The role of alpha-band brain oscillations as a sensory suppression mechanism during selective attention. Frontiers in psychology, 2.
Hardstone, R., Poil, S. S., Schiavone, G., Jansen, R., Nikulin, V. V., Mansvelder, H. D., & Linkenkaer-Hansen, K. (2012). Detrended fluctuation analysis: a scale-free view on neuronal oscillations. Frontiers in physiology, 3. He, B. J., Zempel, J. M., Snyder, A. Z., &
Raichle, M. E. (2010). The temporal structures and functional significance
of scale-free brain
activity. Neuron, 66(3), 353–369. Hipp, J. F., Engel, A. K., & Siegel, M. (2011).
Oscillatory synchronization in large-scale cortical networks predicts perception. Neuron, 69(2), 387–396. Hsü, K. J., & Hsü, A. J. (1990). Fractal geometry
of music. Proceedings of the National Academy of Sciences, 87(3), 938–941. Huron, D. (2010) Humdrum Kern database,
http://kern.humdrum.net. Accessed October 2014.
Jensen, O., & Mazaheri, A. (2010). Shaping functional architecture by oscillatory
alpha activity: gating by inhibition. Frontiers in human neuroscience, 4.
Kantelhardt, J. W., Koscielny-Bunde, E., Rego, H. H., Havlin, S., & Bunde, A. (2001). Detecting long-range correlations with detrended fluctuation analysis. Physica A: Statistical Mechanics and its Applications, 295(3), 441–454. Kello, C. T., Brown, G. D., Ferrer-i-Cancho, R.,
Holden, J. G., Linkenkaer-Hansen, K., Rhodes, T., & Van Orden, G. C. (2010). Scaling laws in cognitive sciences. Trends in cognitive sciences, 14(5), 223–232.
Killingsworth, M. A., & Gilbert, D. T. (2010). A wandering mind is an unhappy mind. Science, 330(6006), 932–932. Kitzbichler, M. G., Smith, M. L., Christensen, S.
R., & Bullmore, E. (2009). Broadband criticality of human brain network synchronization. PLoS computational biology, 5(3), e1000314.
Knyazev, G. G. (2012). EEG delta oscillations as a correlate of basic homeostatic and motivational processes. Neuroscience & Biobehavioral Reviews, 36(1), 677– 695.
Lesiuk, T. (2005). The effect of music listening on work performance. Psychology of Music, 33(2), 173–191.
Levitin, D. J., Chordia, P., & Menon, V. (2012). Musical rhythm spectra from Bach to Joplin obey a 1/f power law. Proceedings of the National Academy of Sciences, 109(10), 3716– 3720.
Lewellen, M. J., Goldinger, S. D., Pisoni, D. B., & Greene, B. G. (1993). Lexical familiarity and processing efficiency: Individual differences in naming, lexical decision, and semantic categorization. Journal of
Experimental Psychology:
General, 122(3), 316.
Linkenkaer-Hansen, K., Nikouline, V. V., Palva, J. M., & Ilmoniemi, R. J. (2001). Long-range temporal correlations and scaling behavior in human brain oscillations. The Journal of neuroscience, 21(4), 1370–1377. Lutz, A., Slagter, H. A., Dunne, J. D., &
Davidson, R. J. (2008). Attention regulation and monitoring in meditation. Trends in cognitive sciences, 12(4), 163–169.
Lux, T., & Marchesi, M. (1999). Scaling and criticality in a stochastic multi-agent model of a financial market. Nature, 397(6719), 498–500. Maris, E., & Oostenveld, R. (2007).
Nonparametric statistical testing of EEG-and MEG-data. Journal of neuroscience methods, 164(1), 177– 190.
Mason, M. F., Norton, M. I., Van Horn, J. D., Wegner, D. M., Grafton, S. T., & Macrae, C. N. (2007). Wandering minds: the default network and stimulus-independent
thought. Science, 315(5810), 393–395. Mathôt, S., Schreij, D., & Theeuwes, J. (2012).
OpenSesame: An open-source, graphical experiment builder for the social sciences. Behavior Research Methods, 44(2), 314–324.
Miller, K. J., Sorensen, L. B., Ojemann, J. G., & Den Nijs, M. (2009). Power-law scaling in the brain surface electric potential. PLoS computational biology,5(12), e1000609.
Mitterer, H. (2014).
http://www.holgermitterer.eu/resear ch.html. Accessed October 2014. Nolan, H., Whelan, R., & Reilly, R. B. (2010).
FASTER: fully automated statistical thresholding for EEG artifact rejection. Journal of neuroscience methods, 192(1), 152–162.
Palva, J. M., Zhigalov, A., Hirvonen, J., Korhonen, O., Linkenkaer-Hansen, K., & Palva, S. (2013). Neuronal long-range temporal correlations and avalanche dynamics are correlated with behavioral scaling laws. Proceedings of the National Academy of Sciences, 110(9), 3585– 3590.
Patel, A. D., & Balaban, E. (2000). Temporal patterns of human cortical activity
reflect tone sequence
structure. Nature, 404(6773), 80–84. Peng, C. K., Buldyrev, S. V., Havlin, S., Simons,
M., Stanley, H. E., & Goldberger, A. L. (1994). Mosaic organization of DNA nucleotides. Physical Review E, 49(2), 1685.
Peng, C. K., Havlin, S., Stanley, H. E., & Goldberger, A. L. (1995). Quantification of scaling exponents and crossover phenomena in nonstationary heartbeat time series. Chaos: An Interdisciplinary Journal of Nonlinear Science, 5(1), 82– 87.
Poil, S. S., Hardstone, R., Mansvelder, H. D., & Linkenkaer-Hansen, K. (2012). Critical-state dynamics of avalanches and oscillations jointly emerge from balanced excitation/inhibition in neuronal networks. The Journal of Neuroscience, 32(29), 9817–9823. Priesemann, V., Wibral, M., Valderrama, M.,
Pröpper, R., Le Van Quyen, M., Geisel, T., ... & Munk, M. H. (2014). Spike avalanches in vivo suggest a driven, slightly subcritical brain state. Frontiers in systems neuroscience, 8.
Rangarajan G., Ding M. (2000). Integrated approach to the assessment of long range correlation in time series data. Phys Rev E Stat Phys Plasmas Fluids
Relat Interdiscip Topics 61, 4991– 5001.
Rauscher, F. H., Shaw, G. L., & Ky, K. N. (1993). Music and spatial task performance. Nature, (365), 611. Roberts, B. M., Hsieh, L. T., & Ranganath, C.
(2013). Oscillatory activity during maintenance of spatial and temporal information in working memory. Neuropsychologia, 51(2), 349–357. Scott, G., Fagerholm, E. D., Mutoh, H., Leech,
R., Sharp, D. J., Shew, W. L., & Knöpfel, T. (2014). Voltage imaging of waking mouse cortex reveals emergence of critical neuronal dynamics. The Journal of Neuroscience, 34(50), 16611–16620.
Siegel, M., Engel, A. K., & Donner, T. H. (2011). Cortical network dynamics of perceptual decision-making in the human brain. Frontiers in human neuroscience, 5, 21.
Siegel, M., Donner, T. H., & Engel, A. K. (2012). Spectral fingerprints of large–scale neuronal interactions. Nature Reviews Neuroscience, 13(2), 121–134.
Sloboda, J. A., O'Neill, S. A., & Ivaldi, A. (2001). Functions of music in everyday life: An exploratory study using the experience sampling method. Musicae scientiae, 5(1), 9–32.
Smit, D. J., Linkenkaer-Hansen, K., & de Geus, E. J. (2013). Long-range temporal correlations in resting-state alpha oscillations predict human timing-error dynamics. The Journal of Neuroscience, 33(27), 11212–11220. Su, Z. Y., & Wu, T. (2007). Music walk, fractal
geometry in music. Physica A: Statistical Mechanics and its Applications, 380, 418–428.
Voss, R. & Clarke, J. (1975) ‘1/f noise’ in music and speech. Nature 258(5533), 317– 318.
Voss, R. F., & Clarke, J. (1978). ‘1/f noise’ in music: Music from 1/f noise. The Journal of the Acoustical Society of America, 63(1), 258–263.
Vuust, P., Ostergaard, L., Pallesen, K. J., Bailey, C., & Roepstorff, A. (2009). Predictive coding of music–brain responses to rhythmic incongruity. Cortex,45(1), 80–92.
Figures Figure 1
Fig. 1: LRTC are present in the harmonic progression of classical piano music. Left, top to bottom: Pitch contours for the first 500 intervals of selected high, middle, and low pitch DFA stimuli. Interval distance in semitones (vertical axis) plotted against successive pitch intervals (horizontal axis) as the music unfolds over time. Right, top to bottom: DFA plots for the corresponding segment (right, same row); data points (black), and regression line (red). The DFA exponent (top) equals the slope of the regression line—straight on a log-log scale.
Figure 2
Fig. 2: Reported ratings of pleasure derived from, familiarity with, and focus on music stimuli—sorted by pitch DFA. Rating distributions did not correlate with LRTC in music. Stimulus codes along vertical axes. Far left: Distribution of pitch DFA α exponents per segment. Middle left: Boxplots of pleasure ratings per segment. Idem for familiarity (middle right), and concentration (far right) ratings. These ratings were provided on a 7-point Likert scale. N = 28; 15 ratings per participant.
Figure 3
Fig. 3: All behavioral dimensions were correlated positively with each other. Responses were provided on a 7-point Likert scale; jittered for improved legibility. N = 28; 15 ratings per participant. Top: Scatterplot with linear fit for pleasure and familiarity ratings. Pearson’s rho (R) and corresponding p-value in title. Middle: Idem for focus and pleasure. Bottom: Idem for familiarity and focus ratings.
Figure 4
Fig. 4: Amplitude and difference topographies in the classical frequency bands and high gamma, between ECR and music conditions. Values in µV; N = 28. Top row: Absolute amplitudes in ECR condition. Upper middle row: Absolute amplitudes in combined music conditions. Lower middle row: Differences between amplitudes in music minus ECR conditions. Bottom row: P-values for locations where significant differences in amplitude are found between music and ECR.
Figure 5
Dichotomous influence of music in the decay of temporal correlations for slow and high frequencies Left column: Amplitude envelopes of delta (first 2 rows) and high-gamma 2 activity (2 bottom rows) during resting-state and while listening a piece of music for a representative recording. Right column, top row: Topographies of the differences between the degree of temporal correlations while listening music* and being at rest for a cohort of 28 subjects. Channels with significantly different medians (p < 0.05) are depicted with white circles. DFA exponents for individual subjects in an exemplary channel (marked with a yellow star in the topographies) are displayed in the bottom.
Figure 6
Fig. 6: Significant positive correlations between familiarity and both beta power over frontoparietal areas as well as gamma power over parietotemporal areas. Values in µV; N = 28. Top row: Topoplots of Spearman’s rank correlation coefficient between measured amplitude per electrode location and familiarity judgments. Bottom row: P-values corresponding to each correlation. Columns (left to right): Classical frequency bands, and high gamma: delta, theta, alpha, beta, gamma, high gamma 1, high gamma 2.
Figure 7
Fig. 7: Correlation and amplitudes of elicited cortical activity in corresponding trials. Significant positive correlations are visible across the scalp between focus responses and alpha power—apart from over primary auditory cortices, and central and frontal areas. Top: Topoplot of Spearman’s rank correlation coefficient (R, N = 28) between measured amplitude per electrode location and focus judgments. Bottom: Corresponding p-values.
Spearman Correlation between Focus and Alpha Power of Elicited Cortical Activity in Corresponding Trials
Figure 9
Fig. 8: Significant negative correlations between the scaling exponent of cortical activity and reported focus on the stimulus are visible over large areas across the scalp from the theta band upwards. Top row: Topoplots of Spearman’s rank correlation coefficient (R, N = 28) between measured DFA exponent per electrode location and focus judgments. Bottom row: P-values corresponding to each correlation. Columns (left to right): Classical frequency bands, and high gamma: delta, theta, alpha, beta, gamma, high gamma 1, high gamma 2.
Fig. 9: No correlations were observed between music DFA (top to bottom: rhythm, pitch, loudness, and mean rhythm, pitch, and loudness) and scaling behavior of elicited cortical activity in classical frequency bands and high gamma (left to right). Top row per pair: Topoplots of Spearman’s rank correlation coefficient (R, N = 28) between measured DFA exponent per electrode location and music DFA. Bottom row per pair: P-values corresponding to each correlation.
Figure 10
Fig. 10: No consistent topology of correlations between pleasure and scaling behavior of elicited cortical activity in corresponding trials was foundin classical frequency bands and high gamma (left to right). Top row per pair: Topoplots of Spearman’s rank correlation coefficient (R, N = 28) between measured DFA exponent per electrode location and music DFA. Bottom row per pair: P-values corresponding to each correlation.
Tables Table 1
Table 1: Stimulus information. Left to right: The codes with which a given stimulus was indicated during the procedure; music composer; artist who performed the piece, title of the piece; rhythm, pitch, and loudness exponents—respectively;
combinations of binned rhythm and pitch exponents in a given piece: L(ow)/M(edium)/High R(hythm), and L/M/H P(itch).
Appendix A Exit questionnaire
What is your age? What is your gender?
Have you had formal musical training? At what age did you start?
How many years of musical training have you had?
In what musical style did you receive your main musical training? What instruments do you play?
Do you work with music professionally? What do you do?
How familiar are you with classical music? What are your favorite music genres?
Appendix B
