We studied one man (53 years old) and five women (74.2 $\pm$ 2.4 years old) who were admitted to the ICU for different pathologies (see Table 1). The patients were selected according to the following criteria:

(1) Critically ill patients under sedo-analgesia.

(2) There were no asymmetries in bioelectrical activity, e.g., no significant interlobar difference for more than two EEG bands in more than one lobe. In other cases, the patients were considered asymmetric and consequently rejected.

(3) There were no previous antecedents of auditory system impairment.

(4) Brain responsive to nociceptive stimulation.

Three patients who fit the above criteria were excluded due to the absence of a brain response to pain.

All patients were studied by computerized tomography (CT) and/or 1.5 T magnetic resonance imaging (MRI, General Electric®, Fairfield, CT, USA).

EEG was indicated by the ICU staff responsible for the clinical assessment of the patient to evaluate the brain activity, including level of sedation, presence of epileptic seizures, or nonconvulsive epileptic status [21]. Hemogram, coagulation, and biochemical analyzes were usually within the normal physiological ranges. Although some values were out of the normal range, we checked that the serum ionic concentrations were always in range.

The experimental procedure was approved by the medical ethics review board of the Hospital Universitario de La Princesa (reference 4274), and informed consent was obtained from the patients’ relatives.

EEG records were obtained using a 32-channel digital system (EEG32U, NeuroWorks, XLTEK®, Oakville, ON, Canada) with 19 electrodes placed according to a 10–20 international system. In addition, a differential derivation Einthoven lead I was positioned for ECG. If necessary, surface electromyography (EMG) channels were added. Recordings were performed at a 512 Hz sampling rate, with a filter bandwidth of 0.5 to 70 Hz and a notch filter of 50 Hz. Electrode impedances were below 25 k$\mathrm{\Omega }$.

The cortical response to nociception was performed by means of strong pressure with a solid object onto the ungual bed for 10 s [22].

Artefacts were scarcely present (due to the clinical state of patients), and respirator-induced movement, muscle-tone increases by endotracheal tube biting, and movement of people around the patient were carefully avoided. Artefact-free periods were selected by visual inspection and exported to ASCII files to be quantified.

The algorithm used has been published previously [23, 24]. Classical EEG bands used in the analysis were defined as delta ($\delta$) = 0.5–4.0 Hz, theta ($\theta$) = 4.0–8.0 Hz, alpha ($\alpha$) = 8.0–13.0 Hz, and beta ($\beta$) = 13.0–30.0 Hz. All auditory stimuli (music) were applied for 120 s, which allowed 130 windows to be computed. Numerical analysis of EEG recordings was performed with custom-made MATLAB® R2019 software (MathWorks, Natick, MA, USA).

A double banana differential EEG montage was reconstructed for analysis [11]. All recordings were divided into different lengths of moving windows of 1 s width, with 10% overlap.

For each window (n) and frequency (k), we computed the fast Fourier transform of the voltage $V^m(n)$ obtained from each channel (m) to obtain the power spectrum $S_{n,k}^m$, in $\mu$V${}^2$/Hz. We used the following expression (Eqn. 1):

(1)$$S_{n,k}^m={\sum }_{n=0}^{N-1}{V}^m(n)e^{-i\frac{2\pi }{N}kn};m=F{p}_1,F_3$$

We also computed Shannon’s spectral entropy (SSE) according to (Eqn. 2):

(2)$$SS{E}_k^m=-{\sum }_{k=0}^F{p}_k{\log }_2{p}_k$$

where F is the maximum frequency computed and p${}_k$ is the probability density of S, obtained from the expression (Eqn. 3):

(3)$$p_k=\frac{S_{n,k}^m}{{\sum }_{k=0}^F{S}_{n,k}^m\mathrm{\Delta }k}$$

We computed the area under the $S_{n,k}^m$ according to the classical segmentation of EEG bands. We used the following expression (Eqn. 4):

(4)$$A_j(k)={\sum }_{k=\text{inf}}^{\text{sup}}{S}_n^m(k)\mathrm{\Delta }k;j=\delta ,\theta ,\alpha ,\beta$$

The expression sup refers to the upper limit of each EEG band and where $\mathrm{\Delta }$k is the increment of frequency.

The absolute value of Pearson’s correlation coefficient ($\rho$) is computed for each pair of channels (i,j) according to the expression (Eqn. 5):

(5)$${\rho }_{\mathrm{ij}}^{\mathrm{k}}=\frac{{\sum }_{\mathrm{k}=1}^{{\mathrm{N}}_{\mathrm{window}}}\left({\mathrm{x}}_{\mathrm{i}}(\mathrm{k})-{\overline{\mathrm{x}}}_{\mathrm{i}}\right){\sum }_{\mathrm{k}=1}^{{\mathrm{N}}_{\mathrm{window}}}\left({\mathrm{x}}_{\mathrm{j}}(\mathrm{k})-{\overline{\mathrm{x}}}_{\mathrm{j}}\right)}{\sqrt{{\sum }_{\mathrm{k}=1}^{{\mathrm{N}}_{\mathrm{window}}}{\left({\mathrm{x}}_{\mathrm{i}}(\mathrm{k})-{\overline{\mathrm{x}}}_{\mathrm{i}}\right)}^2{\sum }_{\mathrm{k}=1}^{{\mathrm{N}}_{\mathrm{window}}}{\left({\mathrm{x}}_{\mathrm{j}}(\mathrm{k})-{\overline{\mathrm{x}}}_{\mathrm{j}}\right)}^2}}$$

where N${}_{\text{window}}$ is the number of points included in a window (usually 128) and ${\overline{x}}_i,{\overline{x}}_j$ represent the means of both channels.

The mean value of all windows was then computed to obtain the mean correlation matrix.

These formulae are discrete versions from well-known expressions for Fourier’s transform, Shannon’s spectral entropy, probability from a mass density function, equivalence for integration, and Pearson’s correlation coefficient, respectively [25].

Channels were grouped by cerebral lobes. In the case of the left hemisphere (shown as an example), we grouped the frontal, $F=\left\{\frac{(F{p}_1-F_3)+(F_3-C_3)+(F{p}_1-F_7)}{3}\right\}$; parieto-occipital, $PO=\left\{\frac{(C_3-P_3)+(P_3-O_1)+\left(T_5-O_1\right)}{3}\right\}$; and temporal, $T=\left\{\frac{\left(F_7-T_3\right)+\left(T_3-T_5\right)+\left(T_5-O_1\right)}{3}\right\}$. Channels from the right hemisphere were grouped accordingly. For each of the lobes, we obtained values for all four bands, as well as the SSE and $\rho$. We denoted these variables as $Lob{e}_{side}^{band};Lobe=F,PO,T;side=r,l;band=\alpha ,\beta ,\delta ,\theta$ (Fig. 1, Ref. [11]).

We selected three pieces of music with clearly different musical properties: (i) dodecaphonic music (Schönberg, Klavierstück Op. 33a; denoted as DodecM), (ii) classical music (Mozart, Sonata for two pianos in D, K. 448, denoted as ClassM), and (iii) heavy metal music (Volbeat, The Devil’s Bleeding Crown, denoted as HeavyM).

We selected these pieces for the following reasons: Mozart’s work has been used as a paradigm for the so-called Mozart’s effect [26, 27] and is well-known in Spain. However, not all the familial relatives knew it. The other two pieces were unknown to the patients’ relatives and were, therefore, also most likely unknown by the patient. Schönberg and Volbeat are poorly known by the general public.

A two-minute duration of each of these pieces was selected and played to the patient through headphones that had been gently placed over the ears. A minimum two-minute period with headphones placed on-site without music was used as the basal state. All patients listened to the music at around the same time of day (between 10:30 and 12:30 h). After the basal period, music stimulation with each of the three types of music (DodecM, ClassM, or HeavyM) was carried out, separated by periods of two minutes of wash-out. The sequence was randomized and different for every patient. Therefore, we can exclude the possibility of bias induced by the order or appearance of the music.

The auditory intensity delivered by the headphones was assessed (Smart Sensor, Intell Instruments Pro®, China), resulting (in dB) ClassM = 68.2 $\pm$ 4.6, DodecM = 67.6 $\pm$ 4.5, and HeavyM = 69.4 $\pm$ 1.7. This sound level corresponds to normal conversation. No differences in intensity were observed between stimuli. We defined the response to music as a statistically significant difference (either positive or negative) in any of the musical stimulations (DodecM, ClassM, or HeavyM) compared with the basal period.

Numerical variables during music stimulation were normalized to the basal period acquired with headphones placed over the ears with no music playing. To evaluate the response, we computed the difference in the mean normalized value during music stimulation minus the basal state (considered as 100%), for example, for the left frontal lobe:

(6)$$d{F}_l^{\delta }=F_{l,music}^{\delta }-F_{l,basal}^{\delta }$$

where variables are power, SSE or $\rho$.

All the variables obtained from the power spectra (i.e., EEG bands) were normalized to the basal state. SSE and $\rho$ (taking into account that their ranges were narrowed) were analyzed by raw values, without normalization.

The chi-square test (${\chi }^2$) was used to assess the differences in the EEG bands between the different music stimulations. Then, the statistical significance of the response to music was evaluated by the Wilcoxon range test. SigmaStat® 3.5 software (SigmaStat, Point Richmond, CA, USA) was employed for statistical analysis.

Pearson’s correlation coefficient was used to study the linear dependance between variables. Linear regression significance was evaluated by means of a contrast hypothesis against the null hypothesis, $\rho$ = 0, using the following formula:

(7)$$t=\frac{r\sqrt{n-2}}{\sqrt{1-r^2}}$$

This describes a t-Student distribution with n-2 degrees of freedom [28].

The significance level was set at p = 0.05, and the results are shown as the mean $\pm$ SEM.

The EEGs recorded during music stimulation did not differ from the basal bioelectrical activity during de visu inspection (Fig. 2A,B). However, numerical analysis revealed significant changes induced by music (compare with Fig. 2C,D). In Fig. 2E,F, as an example, superimposed mean spectra was present for every channel obtained at basal conditions (blue lines) during the playing of ClassM (red lines) and HeavyM (green lines) from the same patient (#2). It is relevant that ClassM decreased the delta component over the whole scalp. However, the more significant results were the changes in spectra induced by HeavyM in most channels, especially in the right fronto-temporal regions, with increases in frequencies higher than delta. However, in this example, neither SSE (basal = 3.20/3.14, ClassM = 3.25/3.22, DodecM = 3.26/3.19 and HeavyM = 3.21/3.19, respectively, for left/right hemispheres) nor $\rho$ were modified by stimulation (basal = 0.613/0.591, ClassM = 0.640/0.588, DodecM = 0.608/0.585 and HeavyM = 0.614/0.586, for left/right hemispheres, respectively).

Although not fully analyzed, we observed that HeavyM stimulation induced the appearance of well-defined components at specific places (see Fig. 2E,F), rather than only modifying the power of preexisting bands.

This different behavior for lobar spectra and SSE/synchronization measures was a significant finding in all patients (Fig. 3). As we observed, clear differences in bioelectrical activity were induced by different types of music.

This different behavior for lobar spectra and SSE/synchronization measures was a significant finding in all patients (Fig. 3). As we observed, clear differences in bioelectrical activity were induced by different types of music.

As shown in Fig. 3, ClassM (Fig. 3A) mostly reduced the power of EEG bands over the whole scalp, except for $F_r^{\delta }$ and $T_l^{\delta }$, although neither of these changes were statistically significant. However, for DodecM, most of the bands from the whole scalp increased in power (except $T_r^{\delta }$), although not all of these changes were significant. In fact, only $F_r^{\beta }$, $P{O}_r^{\beta }$, $T_r^{\alpha }$, and $T_r^{\beta }$ were significant. Therefore, DodecM induced changes in 4/24 possibilities. It is interesting to note that changes were observed only in the faster bands ($\alpha$ and $\beta$) and exclusively in the right hemisphere.

Finally, HeavyM induced increasing spread changes, always increasing the power, mainly for the slower bands (delta and theta, but not only these bands) and in both hemispheres. This kind of music induced significant increases of power in $F_l^{\delta }$, $F_l^{\theta }$, $P{O}_l^{\delta }$, $P{O}_l^{\beta }$, and $T_l^{\alpha }$ in the left hemisphere and $F_r^{\delta }$, $F_r^{\theta }$, $F_r^{\beta }$, and $P{O}_r^{\alpha }$ in the right hemisphere. Therefore, HeavyM induced modifications in 9/24 possibilities. No modifications in SSE were observed regarding the kind of music stimulation.

We compared the effects of the three types of music for the same EEG band in the same lobe. We only observed differences between ClassM and HeavyM at these bands and locations ($F_l^{\delta }$, $P{O}_l^{\alpha }$, $P{O}_l^{\beta }$, and $P{O}_r^{\delta }$).

Finally, although the changes in $\rho$ were in the left hemisphere in opposite directions for different types of music (Fig. 4), none of the changes were significant.

We wanted to check whether the changes in the different EEG bands to different kinds of music stimulation were correlated (e.g., whether the increase in $\delta$ band was associated with a similar increase in the rest of bands). To evaluate this possibility, we performed correlation analysis by means of least-square regression between the band with the highest change (delta) and the rest of the bands throughout the scalp (Fig. 5).

The correlation coefficients (r) were 0.0689, 0.0477, and 0.0748 for the regression between delta vs. theta, delta vs. alpha, and delta vs. beta, respectively, for ClassM, 0.0758, 0.1782, and 0.0474 for DodecM, and 0.1546, 0.2250, and 0.1138 for HeavyM. From Eqn. 7, with n = 36 and p = 0.05, we noted that only values of r $\ge$0.3275 were significant, which was not the case for any of the results obtained here. Therefore, from this result, we concluded that there was no relationship between the changes induced by music in the delta band and the rest of the bands, meaning that changes are band specific.

To characterize the patterns of response in the different lobes of the scalp, we compared the changes induced by the three types of music in all four EEG bands. For the analysis, we performed the ${\chi }^2$ test with three degrees of freedom between pairs of music stimuli (ClassM vs. DodecM, ClassM vs. HeavyM, and DodecM vs. HeavyM). These comparisons are listed in Table 2.

As we can observe from Table 2, the patterns of response were completely different for the left frontal, right parieto-occipital, and both temporal lobes. It is interesting to observe that only 2/18 possibilities were not significant and only for ClassM vs. DodecM in the right frontal lobe and DodecM vs. HeavyM in the left parieto-occipital lobe. All comparisons between ClassM and HeavyM were highly significant in all the lobes.