Infant dominant rhythm desynchronization to faces and objects

Abstract

Infants’ electroencephalography (EEG) dominant rhythm oscillates between 6 and 9 Hz. The desynchronization of this rhythm from baseline to the processing of visual stimuli is used as an index to better understand the development of visual attention. However, development trajectories of desynchronization remain underexplored. Additionally, it is unclear whether development of desynchronization is sensitive to task demands or if it reflects broader developmental changes. To investigate these questions, EEG data were collected from infants aged 6, 9, and 12 months while they passively viewed a fixation cross followed by 10-s trials of a female face or novel object tracked down the screen. Dominant rhythm desynchronization was calculated by subtracting power during the fixation period from power during each task condition. The results revealed significant desynchronization in response to faces at occipital electrodes for all age groups. The magnitude of the desynchronization also increased from 6 to 9 to 12 months of age in response to faces over right occipital electrodes. No significant desynchronization was observed for object stimuli. These findings suggest that dominant rhythm desynchronization develops across infancy and is sensitive to stimulus type. The increased desynchronization for faces compared to objects highlights infants’ general preference for faces relative to objects.

Introduction

During infancy, visual attention plays a crucial role in directing learning (Colombo 2001; Colombo 2002; Markant and Scott 2018; Xie et al. 2018). From the moment infants begin to see, they encounter a continuous flow of objects, people, sounds, and events, which they use to build representations of the world (Amso and Tummeltshammer 2020). Infants utilize attention mechanisms to support their learning and behavior by filtering relevant from irrelevant information and to maintain focus on a specific stimulus despite distractions (Amso and Johnson 2006; Yu and Smith 2011; Johansson et al. 2015; Amso and Tummeltshammer 2020; Brandes-Aitken et al. 2023; Oakes 2023; Schroer and Yu 2023; Mendez et al. 2024). The development of attention enables infants to manage, prioritize, and integrate sensory information effectively and is viewed as an essential skill for learning.

However, the complex nature of attentional processing poses challenges for measurement in infancy. The measurement of electroencephalography (EEG) provides a useful method for quantifying attentional deployment and investigating the development of cognitive processes (Stroganova et al. 1999; Marshall et al. 2011; Bell and Cuevas 2012; Xie et al. 2018; Köster et al. 2019; Jones et al. 2020; Borge Blystad and van der Meer 2022; Brandes-Aitken et al. 2023). The dominant rhythm in infants, sometimes called infant alpha, is an important marker of visual attention and falls within the 6 to 9 Hz frequency range (Stroganova et al. 1999; Xie et al. 2018; Elhamiasl et al. 2024). The dominant rhythm is comparable in function to the alpha activity (8 to 13 Hz) seen in adults. However, during infancy, this rhythm differs from adult alpha in its response properties including its frequency range, spatial distribution, and functional response to different conditions (Kuhlman 1980). Given the difficulties in recording alpha under the same conditions as adults’ alpha rhythm in infants (Stroganova et al. 1999), the term “dominant rhythm” was used to describe the 6 to 9 Hz activity in infants (Elhamiasl et al. 2024).

The peak frequency of the dominant rhythm increases across the first year of life (Stroganova et al. 1999; Marshall et al. 2002; Freschl et al. 2022; Elhamiasl et al. 2024). For example, Stroganova et al. (1999) showed that the spectral peak over the occipital, frontal, and central regions increased from 8 to 12 months of age. A longitudinal study following infants into early childhood (from 4 months to 4 years) also found a clear developmental increase in the frequency of peak activity within the 6 to 9 Hz range over central and posterior regions (Marshall et al. 2002). More recently, the peak frequency of the dominant rhythm, as measured by the center of gravity (CoG), was found to increase from 6 to 9 to 12 months of age over occipital, frontal, and central brain regions (Elhamiasl et al. 2024). In this investigation, developmental increases in peak frequency were present independent of the visual stimulus infants viewed (eg objects, faces, or fixation cross), suggesting general changes in the peak frequency (Elhamiasl et al. 2024). Finally, in a meta-analysis of 40 papers, the peak frequency of the dominant rhythm was shown to increase from early infancy to late adolescence (Freschl et al. 2022). Together, these findings suggest that dominant rhythm peak frequency increases across the first year of life and may represent a reliable index of brain development.

Another critical aspect for understanding the developmental trajectory of the dominant rhythm is the extent to which it is sensitive to visual categories. Dominant rhythm desynchronization has been reported to be larger under conditions that elicit attention (Neubauer and Tetzlaff 1999; Klimesch et al. 2007; Michel et al. 2015). In adults, alpha oscillatory activity (8 to 13 Hz) is dominant over the visual cortex (i) in the absence of visual stimuli and (ii) during a relaxed but awake state (Klimesch 2012). Opening the eyes and thus increasing sensory information attenuates alpha activity, a phenomenon known as alpha desynchronization (Niedermeyer and Lopes da Silva 2005). Event-related desynchronization (ERD) reflects an increase in cortical excitability that is predicted to arise from the activation of cortical areas engaged in information processing, task, and motor response preparation (Orekhova et al. 2001; Klimesch 2012; Arutiunian et al. 2023). Alpha event–related synchronization (ERS) is thought to be related to filtering or ignoring task-irrelevant processes. Using various tasks in adult samples, voluntary attention to visual stimuli and enhanced perceptual processing have been shown to result in occipital alpha desynchronization (Ergenoglu et al. 2004; Trenner et al. 2008; Mazaheri et al. 2014). In addition, alpha desynchronization over occipital regions has been linked to involuntary attention orientation toward a prominent but spatially unpredictable sound, suggesting top–down modulation of desynchronization (Feng et al. 2017). It is therefore plausible that alpha desynchronization may index general priming mechanisms, which gate visual stimuli in the cortex and enhance the processing and perception of the stimuli following attention (Mazaheri et al. 2014; Feng et al. 2017).

In one investigation, 7 to 8 months exhibited decreased power within 6 to 9 Hz activity over precentral regions during both anticipation and reappearance phases of dynamic peek-a-boo task (Stroganova et al. 1999). Desynchronization of the dominant rhythm over frontal and occipital regions was also recently demonstrated while they watched a Sesame Street video (Xie et al. 2018). In this previous study, desynchronization effects began to emerge around 10 months and were well established by 12 months of age. These findings highlight the presence of attention-dependent infant dominant rhythm desynchronization during infancy. However, the developmental trajectory of the dominant rhythm desynchronization across the first year of life and its distribution over the cortex is not well understood.

Although desynchronization of the dominant rhythm is thought to reflect an increase in attentional deployment (eg Klimesch et al. 2007; Mazaheri et al. 2014), developmental studies have primarily used a single stimulus or condition making it difficult to determine if the relative decrease in power is sensitive to experimental tasks and/or stimuli. In a previous study, infants’ dominant rhythm activity (5 to 8 Hz) was found to desynchronize in response to object-directed and object-averted eye gaze in 4- and 9-month-olds (Michel et al. 2015). Additionally, increased alpha desynchronization in response to faces looking at objects compared to faces looking away from objects was also shown in 4- and 9-month-old infants but not 2- and 5-month-old infants (Michel et al. 2015). Enhanced alpha desynchronization to faces looking at objects (relative to away from objects) was interpreted as an indicator of increased attention to objects cued by gaze. These results suggest that the magnitude of infant dominant rhythm desynchronization may be influenced by tasks and stimuli that elicit more or less visual attention. The impact of visual stimuli on infants’ attention has been supported in event-related potential (ERP) studies. For example, infants’ negative central (Nc) component is sensitive to stimulus familiarity (Nelson and Collins 1992; Reynolds and Richards 2005) and attention allocation (Ackles and Cook 2007; Dennis et al. 2009; Conte et al. 2020). Infants typically showed larger Nc amplitudes for female faces compared to infant faces, toys, and familiar stimuli like their mother’s face (de Haan and Nelson 1999; Xie and Richards 2016). Conversely, other studies reported reduced Nc amplitudes for faces compared to objects (McCleery et al. 2009; Jones et al. 2016; Di Lorenzo et al. 2020). Behavioral studies also reveal a notable preference in infants for faces over objects, evident across various developmental stages. Research showed that both 6- and 12-month-old infants exhibited longer peak looks—the duration of their longest gaze—toward faces compared to toys (Jones et al. 2017). Although the peak look duration decreased from 6 to 12 months, the preference for faces persisted. In addition, 4- to 8-month-old infants displayed a consistent preference for faces during the first second of looking (DeNicola et al. 2013). Moreover, research on 6-month-olds found that while faces, body parts, and animals initially captured attention, faces were most effective at sustaining attention over time compared to other stimuli (Gluckman and Johnson 2013). These findings indicate that infants generally prefer to look at faces than at objects. Accordingly, the magnitude of dominant rhythm desynchronization is expected to be larger in response to faces than objects.

In the present investigation, dominant rhythm desynchronization in response to faces and novel objects was examined in 6-, 9-, and 12-month-old infants. We aimed to assess whether the dominant rhythm significantly desynchronized from baseline to task conditions (faces and objects) across age groups. To investigate this, we conducted a permutation-controlled topographical one-sample t-test on a dense array of EEG data (109 electrodes) to identify significant desynchronization across all electrodes. Additionally, we examined age-related changes in the desynchronization of the dominant rhythm to determine if its magnitude increased with age. We used mass univariate techniques to provide greater spatial detail in identifying these changes. Given that dominant desynchronization is primarily noted in the occipital region, a region-of-interest (ROI) analysis was conducted to evaluate the development of desynchronization in this area across ages and conditions. Based on previous findings, we predicted that the dominant rhythm would desynchronize from baseline to task conditions and that its magnitude would increase from 6 to 12 months of age (Michel et al. 2015; Xie et al. 2018). Further, faces were expected to elicit increased desynchronization relative to objects (DeNicola et al. 2013; Gluckman and Johnson 2013; Jones et al. 2017). The results of the current study are expected to inform our understanding of age- and stimulus-related differences in dominant rhythm desynchronization across the first year of life. These findings may offer valuable insights into the developmental trajectory of visual attentional and sensory processing and their underlying neural rhythms during infancy.

Materials and methods

Participants

The study was conducted in accordance with the ethical standards set forth by the World Medical Association’s Code of Ethics (Declaration of Helsinki) and received approval from the local Institutional Review Board (IRB201701431). The study included a final sample of 111 infants, comprising 49 6-month-olds (M = 188.29 days, SD = ±7.38; 24 females), 32 9-month-olds (M = 274.22 days, SD = ±9.18; 17 females), and 30 12-month-olds (M = 366.97 days, SD = ±8.67; 14 females). The demographic data are presented in Table 1. All infants were born full term, and their parents reported no history of visual, auditory, or neurological issues.

Participants were recruited from a pre-existing database of parents who had previously consented to be contacted for research purposes. Recruitment methods for the database included mailings and direct engagement with families at community gatherings. Families with infants aged 6, 9, or 12 months were invited to participate in a larger study on brain development and learning. Prior to participation, written informed consent was obtained from all parents, who were compensated with $10 and given a small toy for their infant.

Stimuli and task

Stimuli and apparatus

The task involved presenting two types of stimuli: novel objects and faces (Fig. 1A). The novel objects, called “Sheinbugs,” included 36 computer-generated objects across three species (Red, Yellow, and Blue), with 12 exemplars per species (Jones et al. 2020; Elhamiasl et al. 2022; Kutlu et al. 2023; Boylan et al. 2024). Object exemplars were created and edited using Modo© (Luxology, LLC) and varied in shape, color pattern, size (between 8 to 12 cm [width] × 8 to 12 cm [length]), and orientation. The face stimuli comprised nine colorful forward-facing, smiling female portraits selected from the Karolinska Directed Emotional Faces database (Lundqvist et al. 1998). The faces were divided into three distinct groups to create counterbalanced sets. Importantly, the grouping was conducted randomly, ensuring no discernible differences between groups based on facial characteristics such as race, age, or other features. Each face was 11 cm × 11 cm. Additionally, all stimuli were uniformly cropped and resized to fit within a 500 by 500 pixel frame, presented in full color against a gray background. Both object and face stimuli were displayed at a visual angle of approximately 8.27 to 12.30 degrees horizontally and vertically. To maintain uniformity in luminance across stimuli, a modified version of the lummatch function in the SHINE toolbox in MATLAB (Willenbockel et al. 2010) was utilized to equalize the mean and standard deviation of luminance.

Using Psychtoolbox-3 for Windows 7 in MATLAB R2016b (MathWorks Inc.), we presented the stimuli on a 22.5-inch VIEWPixx liquid crystal display (LCD) monitor (VIEWPixx Technologies, QC, Canada). The VIEWPixx LCD monitor encompassed a resolution of 1,920 × 1,200 pixels at a refresh rate of 120 Hz, 12-bit red, green, blue (RGB) intensity, and display properties ensuring luminance and color uniformity across 95% of the display area. The monitor also featured a wide gamut light emitting diode (LED) and a scanning LED backlight with a direct RGB LED array.

Task: floating stimuli

Each block of the task began with a 5-s fixation cross, followed by two 10-s trials of either faces or objects (Fig. 1B). During each trial, stimuli tracked or floated down the screen for 10 s, with a 3-s interblock interval. The task comprised four trials (two object trials and two face trials), with the assignment of exemplars counterbalanced across participants. Counterbalancing was achieved through the creation of 18 different sets and subsets of stimuli combinations (for more information, see Elhamiasl et al. 2024).

Procedure

Infants were seated on their parent’s lap approximately 55 to 65 cm from the monitor. The EEG nets were applied by one experimenter, with a second experimenter engaging the infant to ensure their comfort and proper placement of the net. If necessary, electrodes were adjusted until impedances were below 50 kΩ. After net placement and the experiment began, the lights were dimmed. The infant remained with their parent and an experimenter focused on redirecting the infant’s attention to the screen if they became distracted.

E‌EG recording

EEG activity was recorded using a 128-channel Hydrocel Geodesic Net while they passively observed the stimuli float down the screen. The net was connected to a DC-coupled high-input impedance amplifier (Net Amps 400 TM, Electrical Geodesics Inc., Eugene, OR, now Magstim EGI). Amplified signals were digitized every millisecond, corresponding to a sample rate of 1000 Hz. All electrodes were referenced online to the vertex (Cz).

Data preprocessing

The raw EEG data were preprocessed utilizing a modified version of the Harvard Automated Processing Pipeline for Electroencephalography (HAPPE) 2.2 pipeline (Lopez et al. 2022; GitHub: PINE-Lab/HAPPE) and custom in-house code employing EEGLAB functions (https://osf.io/adp3m/?view_only=cd16498dbf3b4abfbc1fa4837a2590e7). The outer band electrodes including electrodes E17, E43, E48, E49, E56, E63, E68, E73, E81, E88, E94, E99, E107, E113, E119, and E120 were removed due to heightened noise level. The outermost ring of electrodes typically exhibits poor connections and increased noise in infant data, making it standard practice to remove this ring prior to data preprocessing in infant EEG studies (Fujioka et al. 2011; Nyström et al. 2011; Maffongelli et al. 2018; Debnath et al. 2020; Leach et al. 2020; Maffongelli et al. 2020; Routier et al. 2023). Next, a 1 Hz highpass second-order Butterworth filter was applied. Signals were downsampled to 500 Hz, and a 30 Hz, eighth-order Butterworth lowpass filter was employed. Then, wavelet denoising was applied to the data before bad electrode selection to retain more data by improving the robustness of artifact detection while minimizing the loss of valid data.

For wavelet denoising, each electrode’s timeseries underwent wavelet transformation (using Coif4), wherein a wavelet function was fitted to the EEG data, generating a set of coefficients. This transformation segmented the EEG data into various frequency ranges, employing a decomposition level of 10. Subsequently, the wavelet transform coefficients within each frequency bin were subjected to thresholding to distinguish artifacts from neural signals using an empirical Bayesian method for setting the threshold (Clyde and George 2000). For each frequency range, the threshold was individually determined (level-dependent threshold), and the hard threshold rule was employed to ensure complete separation between the neural signal and the artifact. Following this, an inverse wavelet transform was conducted to convert artifact-related coefficients back to the timeseries domain, maintaining the integrity of the signal’s frequency space, phase, and amplitude. This inverse transform produced an artifact timeseries that was subtracted from the original timeseries, yielding the artifact-corrected timeseries.

After wavelet thresholding, bad channels were detected by applying a 1-s moving window algorithm to all EEG channels in continuous EEG data. A −300/300 μV threshold was used as the criteria for bad channel detection. If the length of the total artifactual section of the channel was greater than 25% of the length of the recording, that channel was marked as bad. Then, waveleting quality control was done to assess the performance of wavelet thresholding. Bad channels were interpolated using spherical splines only if the number of bad channels was less than 15% of all 109 channels (16 channels). Data with more than 15% of bad channels were labeled “failed” and did not proceed to the next steps. The average number of bad channels in our sample was 1.6 channels (SD = ±1.91). Data were then re-referenced to the average to minimize the impact of noise at the reference site and to render more intuitively interpretable scalp topographies.

Eye electrode channels (E8, E9, E14, E21, E22, and E25) were excluded from segment artificial rejection. Data were then segmented for each trial and time-locked to the onset of the stimuli (−0.2 to 10 s for face and object trials and − 0.2 to 5 s for fixation trials). Within each trial, artifact detection was completed on the time series data in three steps. First, bad trial epochs were determined using a voltage threshold of ±300 μV. If an epoch exceeded this threshold, a second round of bad channel detection was run to determine the channels (up to 5) that best explained more than 50% of present artifacts. These channels were then interpolated using the spherical spline method. Finally, segment artifact detection was re-run and any trial epochs still exceeding the voltage threshold were excluded from further analyses. An average of 1.98 face trials (out of 2, SD = 0.13) and 1.99 object trials (out of 2, SD = 0.09) were retained across participants.

In the next step, the 10-s face and object trials were each divided into 2 5-s trial segments to equate the time series to the length of the fixation trials. This trial segmentation is crucial to ensure that comparable amounts of data enter the desynchronization calculation in later steps. The 5-s trial-segmented face and object data as well as the 5-s fixation trial data were then submitted to 1-s moving window fast Fourier transformations (FFTs), which resulted in a frequency resolution of 1 Hz. The frequency domain data from the 5 1-s FFTs per trial (fixation) or trial segment (face, object) were then averaged together. Thus, resulting frequency data for face or object presentation and fixation are composed of the same amount of time series data. The spectrum of the processed data, averaged across occipital electrodes and the combined conditions of faces and objects, is shown in Fig. 2.

The magnitude of desynchronization was calculated by subtracting the preceding fixation trial power from the task trials according to the following formula:

In this formula, F refers to the power values within the faces condition and O refers to the power values within the objects condition. A negative output indexed desynchronization (power attenuation from baseline), while a positive output showed synchronization (power increase from baseline). For the data analysis, only desynchronization within the dominant rhythm (6 to 9 Hz) frequency band was examined. Desynchronization within the dominant frequency band (6 to 9 Hz) was averaged for each participant for each condition and electrode. Desynchronization averaged across all participants and occipital electrodes for the combined conditions of faces and objects is shown in Fig. 3.

We also removed 1/f noise for further inspection, confirming that 1/f distribution did not affect the shape of the desynchronization. Consequently, we proceeded with the raw desynchronization. For the desynchronization plot after 1/f noise removal, see Supplementary A.

Data analysis

First, we examined whether the dominant rhythm significantly desynchronized from baseline to task conditions (faces and objects) within each age group. To explore desynchronization patterns across the topography, we conducted a permutation-controlled, one-tailed one-sample t-test against a null hypothesis value of zero for each electrode, separately for each age group and condition. This method enabled us to compare developmental differences at each scalp location using powerful corrections for multiple comparisons and detect effects with greater spatial information (Groppe et al. 2011a; Groppe et al. 2011b). To establish the statistical significance threshold, we employed permutation methodology, in which per 1,000 draws, participant data were randomly shuffled via a coin-toss procedure, ie a 50–50 chance of the original data being included or replaced with 0, to create a shuffled/null dataset for each age group and condition. Next, a one-sided t-test against zero was conducted on the shuffled dataset for each electrode. For the desynchronization t-test, the minimum t-value across the topography was extracted for each draw, forming a distribution of minimum t-values. The lower 5% quantile value served as the cutoff for statistical significance for the desynchronization analyses. This means that any observed t-value falling below this threshold was considered statistically significant, indicating robust evidence against the null hypothesis and demonstrating that the desynchronization of the dominant rhythm at those electrodes was significantly less than zero.

Second, to investigate age-related differences in the desynchronization of the dominant rhythm, we employed a between-group Bayesian bootstrapping linear model at each electrode in the conditions where significant desynchronization was observed in the previous step. This method provided a robust evaluation of between-age differences in desynchronization for our models (model weights = [3, 2, 1], corresponding to ages 6, 9, and 12 months, respectively; Fig. 5A). Our model assessed the increasing magnitude of desynchronization from 6 to 9 to 12 months, indicating that desynchronization values for 12-month-old infants were more negative than those for 9-month-old infants, and those for 9-month-olds were more negative than those for 6-month-olds. In this procedure, a bootstrapped test distribution and a bootstrapped null distribution were created over 5,000 draws. During each of the 5,000 bootstrap iterations, we randomly sampled subjects within each age group to compute mean desynchronization values for each condition. This process allowed us to generate distributions of inner products for both the tested models and a null model, where weights were shuffled randomly. After completing the bootstrapping, we calculated Bayes factors (BFs) for each EEG channel and condition to quantify the evidence supporting the tested models compared to the null model. Using a BF10 threshold of ≥3, suprathreshold electrodes provided substantial evidence for the tested model (Jeffreys 1961) and demonstrated that the tested model was at least 3 times as likely as the null model of no pattern across age. Thus, electrodes exceeding the threshold of BF ≥3 suggest that the magnitude of dominant rhythm desynchronization was significantly greater for 12-month-old infants than for 9-month-olds and greater for 9-month-olds than for 6-month-olds.

Third, in addition to the topographical analysis, we were interested in investigating to what extent the desynchronization of the occipital dominant rhythm changes with age and in response to different conditions. Accordingly, we averaged the desynchronization values over seven occipital electrodes (E70, E71, E74, E75, E76, E82, and E83) and conducted an ROI-based repeated-measures ANOVA with Condition (objects, faces) as a within-subjects factor and Age (6, 9, 12 months) as a between-subjects factor. Significant effects and interactions were followed up with Bonferroni-corrected analyses, as appropriate.

Results

The results of the permutation-controlled one-sample t-test indicated significant desynchronization of the dominant rhythm to faces over three posterior electrodes at 6 months of age (threshold t-value = −4.01, min t-value = −4.32, max t-value = −4.02, significant electrodes = E69, E70, and E75 (including and around O1 and Oz); see Fig. 4A, left). At 9-month-old infants, seven occipital electrodes showed significant desynchronization (threshold t-value = −4.62, min t-value = −5.38, max t-value = −4.63, significant electrodes = E70, E74, E75, E76, E82, E83, and E85 [including and around Oz]; see Fig. 4A, middle). At the age of 12 months, the dominant rhythm desynchronized significantly at six occipital electrodes (threshold t-value = −3.90, min t-value = −4.30, max t-value = −3.92, significant electrodes = E70, E71, E72, E75, E83, and E89 [including and around Oz]; see Fig. 4A, right). For the objects condition, no significant desynchronization was detected for all three age groups (6 months threshold t-value = −2.72, 9 months threshold t-value = −2.95, 12 months threshold t-value = −2.91) (Fig. 4B).

Since the desynchronization of the dominant rhythm was significant only in response to faces, we examined the topographical Bayesian bootstrapping model (Fig. 5A) in this condition. The results indicated that the tested model was supported over the null model across occipital electrodes, including E62, E71, E72, E76, E77, E83, E84, E89, E90, and E95 (around Oz). At these electrodes, there was a significant age-related increase in the magnitude of dominant rhythm desynchronization, with greater desynchronization observed at 12 months compared to 9 and 6 months (Fig. 5B and C).

The results of ANOVA assessing Age * Condition interaction in the occipital desynchronization showed significant main effects of age [F (2,108) = 3.59, P = 0.031] and condition [F (1,108) = 27.24, P < 0.001]. The magnitude of occipital desynchronization increased for 9-month-olds [t (52.90) = 2.469, P = 0.016] and marginally increased in 12-month-olds [t (42.91) = 1.930, P = 0.060] compared to 6-month-old infants (Fig. 6A). The occipital desynchronization was greater in response to faces than objects (Fig. 6B). The interaction between age and condition was not significant [F (2,108) = 0.54, P = 0.583].

Discussion

The present study examined the developmental trajectories of the dominant rhythm desynchronization across infancy while infants were presented with floating faces and objects. The magnitude of desynchronization was significantly lower than zero only in the faces condition. While these significant differences were present over a few occipital electrodes at 6 months, the number of electrodes with significant desynchronization was greater over the occipital region for 9- and 12-month-old infants. The examination of age-related changes in desynchronization during the face condition revealed that the magnitude of dominant rhythm desynchronization increased from 6 to 9 to 12 months of age, especially over the right occipital electrodes, consistent with our tested model. The assessment of occipital dominant rhythm also indicated that occipital desynchronization was present in infants as young as 6 months, and the magnitude increased from 6- to 9- to 12-month-old infants. In addition, the occipital dominant rhythm attenuated more in response to faces compared to objects, supporting the sensitivity of desynchronization to different types of visual stimuli.

The present results showed that the dominant rhythm significantly desynchronized from baseline in response to faces. Significant desynchronization was present over some occipital electrodes at the age of 6 months, and the number of occipital electrodes with significant desynchronization increased as infants got older at the age of 9 and 12 months old. These results are in line with previous research showing the presence of dominant rhythm desynchronization in infants as young as 6 months of age, with maturation observed by the end of the first year (Stroganova et al. 1999; Michel et al. 2015; Xie et al. 2018). Xie et al. (2018) found that desynchronization of the dominant rhythm over the occipital region occurred in 10-month-old infants while they watched a dancing character and this response fully developed by the time they reached 1 year of age. In contrast, the present study identified that significant desynchronization began at 6 months of age. However, when comparing the results of different studies, it is essential to consider the various methods used for data preprocessing and analysis, as these might influence the findings. In the current study, we used absolute power, which is consistent with the approach adopted by several other studies (eg Stroganova et al. 1999; Michel et al. 2015). However, some other studies, have used relative power (eg Orekhova et al. 2001). Relative power is the proportion of power within a specific frequency band at each electrode site, normalized by the total power across all frequency bands at that site, and is used to account for among-subject differences (Duffy 1994; Orekova et al. 2001). However, relative power can introduce operational interdependence among the spectral values of different frequency bands, meaning that the values of one band are mathematically linked to those of another, which can complicate the interpretation of the results (Marshall et al. 2002). The choice between analyzing EEG band power as absolute or relative has been debated (Marshall et al. 2002). Some studies suggest that relative power may provide better test–retest reliability than absolute power (John et al. 1980). Additionally, factors like changes in bone thickness, skull resistance, and impedance in developmental populations could affect absolute power measurements (Benninger et al. 1984). However, other studies found no significant age-related effects on total power due to these factors and, therefore, used both absolute and relative power in their analyses (Gasser et al. 1988; Clarke et al. 2001). Preprocessing techniques can also influence the data and final results. Some studies might apply minimal preprocessing steps, while others perform more extensive artifact rejection to retain more data. Although the HAPPE pipeline is a widely recognized and peer-reviewed tool specifically designed for infant EEG data (Lopez et al. 2022), we re-preprocessed our data as a follow-up to our previous analysis using a version of the pipeline that excluded the wavelet denoising procedure, thereby reducing the level of artifact rejection. We found that the spectra generated from both preprocessing pipelines were similar, with only minor increases in amplitude within the dominant rhythm range, rather than pronounced peaks.

Furthermore, using a computerized visual task that minimized muscular activity, we noted that desynchronization predominantly occurred over the occipital region. This contrasts with previous findings, which reported desynchronization of 6 to 9 Hz activity in the precentral regions during dynamic tasks like peek-a-boo games (Stroganova et al. 1999). The increased muscular activity and movement in such tasks were thought to reflect attenuation of the mu rhythm (Stroganova et al. 1999). Event-related modulation of the dominant rhythm power is thought to reflect the flow of information in the cortex via selective attention and suppression of sensory signals (Klimesch et al. 2007; Foxe and Snyder 2011; Klimesch 2012; Mazaheri et al. 2014), the present results suggest that as infants get older, they become better at applying top–down processes (attention) to task-relevant information. The thalamus and visual cortex are predicted to be two main neural sources of dominant rhythm generation that modulate dominant rhythm oscillations through thalamo-cortical and cortico-cortical networks (Andersen and Andersson 1968; Lopes da Silva et al. 1973; Lopes da Silva 1991). In a previous study utilizing EEG effective connectivity source analysis (iCoh), researchers assessed the number and strength of both increasing and decreasing connections from 3 months (baseline) to 6 and 12 months of age by subtracting the iCoh values at 3 months from those at 6 and 12 months (Bosch-Bayard et al. 2022). The results revealed that the thalamus establishes connections with all brain regions except the middle occipital area, while all regions project to the thalamus (Bosch-Bayard et al. 2022). As infants got older, connections from the thalamus showed a declining trend, suggesting that while the thalamus is extensively connected at birth, inefficient connections are pruned away with age (Bosch-Bayard et al. 2022). Conversely, the number of connections directed to the thalamus increased, supporting the prediction that maturation and myelination enhanced feedback from the cortex to the thalamus (Bosch-Bayard et al. 2022). The study also observed that the reduction in connections was particularly notable at 6 months, especially within the theta and alpha frequency bands, suggesting that pruning is more significant during the first half of the infant’s first year. At 12 months, the increasing connections were stronger, suggesting a continued growth of connections toward the end of the first year (Bosch-Bayard et al. 2022). As the dominant rhythm power can be modulated through the thalamus, cortex, and their connections, developmental changes in connectivity could account for the age-related differences in the desynchronization of the dominant rhythm. In line with this, functional connectivity analysis from adults’ EEG data explained alpha desynchronization in terms of the number of connections. The results showed that from eye-closed to eye-open conditions, the number of nodes and connections decreased in bilateral posterior areas for the alpha band (Tan et al. 2013). Additionally, a study involving healthy adults and individuals with pharmacologically refractory mesial temporal lobe epilepsy revealed that power-based connectivity consistently discriminated between the eyes closed and eyes open conditions for temporal and interhemispheric electrodes in both patients and the healthy control group. Phase-based connectivity, on the other hand, was sensitive to the transition from eyes closed to eyes open only in interhemispheric and frontal electrodes (Gómez-Ramírez et al. 2017). Thus, the developmental changes in brain connectivity may serve as a hypothetical underlying mechanism for the shifts observed in dominant rhythm desynchronization.

The present results showed greater desynchronization in response to faces than objects. Previous studies demonstrated that visual attention and perception modulate dominant rhythm activity, with greater attenuation in response to increased external attention and perception (Vanni et al. 1997; Sauseng et al. 2005; Babiloni et al. 2006; Magosso et al. 2019). Accordingly, the current results suggest that infants may deploy more attentional recourses to faces than novel objects, leading to greater suppression of dominant rhythm activity. This differentiation of faces and objects during infancy is in line with developmental ERP data. For example, the Nc is suggested to be greater during attention deployment than inattention (Reynolds and Richards 2005; Reynolds et al. 2010; Guy et al. 2016). Some studies reported that infants showed larger Nc amplitudes to female faces than to infant faces and toys (Xie and Richards 2016), as well as to novel stimuli (unfamiliar faces and unfamiliar toys) compared to familiar stimuli (mother’s face and own toys) and to familiar faces (mother) than to novel faces (de Haan and Nelson 1999), while others reported decreased Nc amplitude to faces compared to objects [symmetrical nonface pictures (Jones et al. 2016); toys (McCleery et al. 2009), houses (Di Lorenzo et al. 2020)]. Even though it might be difficult to interpret the opposite directions of the Nc effect, these contradictory results reflect the effect of the stimuli to which faces are contrasted (Di Lorenzo et al. 2020). Different patterns of Nc amplitudes for faces compared to objects demonstrate differences in either attention or familiarity (Di Lorenzo et al. 2020). Since both the exemplars of female faces and novel objects used in our study were unfamiliar to infants, greater desynchronization in response to faces than objects might indicate that babies preferred faces to objects.

The different desynchronization responses to faces and objects are similar to ERP findings showing face-sensitive components (adult N170; infant N290/P400). Traditionally, ERPs are thought to reflect cortical neuron responses to sensory input, while event-related de/synchronization is viewed as changes in local interactions between neurons regulating EEG frequency (Pfurtscheller and Lopes da Silva 1999). However, previous research suggests that modulations in P300 and alpha amplitude may be linked in adults (Studenova et al. 2023; Başar et al. 1999). Alpha oscillations are thought to contribute to the generation of P300 through the baseline-shift mechanism (BSI), which suggests that stimulus-modulated oscillations with a nonzero mean can produce evoked responses (Nikulin et al. 2007; Studenova et al. 2023). In line with our findings, several developmental studies have shown differential ERP responses to faces relative to objects (de Haan and Nelson 1999; Guy et al. 2016; Conte et al. 2020; Di Lorenzo et al. 2020; Yrttiaho et al. 2022; Glauser et al. 2022). However, the task used in the present study does not present discrete stimuli like an ERP task does. Here, stimuli tracked or floated down the screen and there were not enough trials presented to look at ERP responses. However, given similar patterns of responses for the present results and previous infant ERP studies, it is likely that the desynchronization of the dominant rhythm and ERP components may both be modulated by attentional processes. Future studies examining the link between rhythm desynchronization and ERP components during face and object processing could clarify how these modulations are interconnected.

Furthermore, the current results indicated a significant age-related increase in the magnitude of dominant rhythm desynchronization associated with face processing, predominantly occurring in the right occipital electrodes. These results align with previous research that highlights right hemisphere specialization for face perception in infancy (Scott and Nelson 2006; de Heering and Rossion 2015; Buiatti et al. 2019). In adults, the right hemisphere, particularly the ventrolateral occipitotemporal cortex, plays a crucial role in face perception (Kanwisher et al. 1997; Haxby et al. 2000; Meng et al. 2012; Jonas et al. 2016; Lesinger et al. 2023). Studies have also reported right lateralization in face processing (Scott and Nelson 2006; de Heering and Rossion 2015; Buiatti et al. 2019). For instance, Scott and Nelson (2006) demonstrated that in 8-month-old infants, the latency of the N290 component was picked later in the right hemisphere, with a pronounced sensitivity to configural changes in faces compared to featural alterations. This pattern contrasts with the left hemisphere’s response, which favored featural changes, indicating a lateralized processing strategy. Expanding on these findings, de Heering and Rossion (2015) utilized fast periodic visual stimulation with 4- to 6-month-old infants, uncovering a robust face-specific response in the right occipito-temporal cortex. Importantly, this response was absent when phase-scrambled images were presented, effectively discounting low-level visual explanations for the observed effects. These results suggest that even at a young age, infants are tuning into face-specific features, reinforcing the idea of specialized processing. Additionally, Buiatti et al. (2019) explored face processing in newborns aged 1 to 4 days, employing a frequency-tagging design with face-like patterns. The findings showed that upright face-like stimuli produced significantly stronger power than inverted controls across multiple electrodes and revealed the engagement of a partially right-lateralized network overlapping with adult face-processing regions. The current results on dominant rhythm desynchronization further support the right lateralization of face processing, indicating that this hemisphere is dominant for face processing by the end of the first year of life.

These results imply that dominant rhythm desynchronization might be considered a stimuli-sensitive index that can be used to investigate differential cortical activity to various visual stimuli and experimental conditions. This is in line with ideas that state alpha desynchronization is associated with tasks requiring the processing of relevant information and is a more task-related phenomenon than has been previously believed (Vanni et al. 1997; Sauseng et al. 2005; Babiloni et al. 2006; Magosso et al. 2019). For example, in 4- and 9-month-old infants, the dominant rhythm desynchronized more in response to faces looking at objects compared to faces looking away from objects, suggesting the allocation of more attention to the relevant object (here, an object that is cued by eye gaze) (Michel et al. 2015). The task used in the current study can be easily applied to answer questions related to attention allocation in development.

In the current study, desynchronization was not only limited to the 6 to 9 Hz activity but also occurred at lower frequency bands (Fig. 3). This raises important considerations for interpreting the current results. An important challenge in studying EEG activity during infancy is the inconsistent frequency boundaries (Orekhova et al. 2006), which leads to variability in how studies define and categorize frequency bands. For instance, some research categorized the 4 to 8 Hz range as theta, while others labeled it as alpha, and some did not assign a label at all (Futagi et al. 1998; Bell 2002; Schmidt et al. 2003). Additionally, the choice of frequency bands in early infancy was suggested to depend on the specific brain regions and phenomena being investigated (Orekhova et al. 2006). For example, while the 6 to 9 Hz band was suggested to capture central sensorimotor rhythms from around 5 months of age, the dominant rhythm in posterior regions was supposed to be observed in a lower range, such as 4 to 6 Hz (Orekhova et al. 2006). This variability has led to terms like “alpha-like” or “dominant rhythm” being used instead of alpha activity (Kuhlman 1980). Consistent with this, the posterior dominant rhythm, which is thought to be a precursor to the alpha rhythm, is believed to be established by approximately 2 months of age (Britton et al. 2016). This rhythm begins at a frequency of 3 to 4 Hz, increases to 4 to 5 Hz by 6 months, reaches about 5 to 7 Hz by 12 months, and finally transitions into the alpha range of around 8 Hz by age 3 (Britton et al. 2016). Consequently, the dominant rhythm in the first year of life may span a broader frequency range, despite developmental studies generally categorizing the dominant rhythm or alpha activity within the 6 to 9 Hz range. The desynchronization observed in lower frequencies in the current study suggests that these frequencies, especially those below 6 Hz, might represent the dominant rhythm rather than distinct activities like theta. Additional research is needed to ascertain whether the characteristics of frequencies below 6 Hz are consistent with those in the 6 to 9 Hz range and whether they reflect elements of the dominant rhythm or signify different types of brain activity.

Additionally, the observed desynchronization in lower frequencies may reflect theta activity in response to stimuli. Both theta and alpha oscillations play critical roles in attention. Although some studies report theta synchronization in response to attention (Michel et al. 2015; Xie et al. 2018), others have found evidence of theta desynchronization (van der Meer et al. 2008; Tsai et al. 2013; Agyei et al. 2015; Mateos et al. 2022). One study observed that 8-month-old infants displayed theta desynchronization over the visual cortex in response to motion stimuli, while theta synchronization occurred with static dot patterns (van der Meer et al. 2008). The authors suggested that the infants’ responses to motion stimuli indicate a more complex processing mechanism compared to static patterns, leading to more suppression of synchronous neuronal firing and theta desynchronization during motion perception (van der Meer et al. 2008). Similarly, a study involving infants aged 3 to 4 months and 11 to 12 months found theta desynchronization in response to optic flow in both age groups (Agyei et al. 2015). In the current study, where stimuli were tracked or floated down the screen, the observed theta desynchronization may also reflect infants’ visual motion perception. This aligns with the findings that proposed increased task complexity contributes to enhanced theta desynchronization (Pfurtscheller and Lopes da Silva 1999).

The findings from the present study, coupled with our earlier research on the development of dominant rhythm peak frequency (Elhamiasl et al. 2024), suggest that desynchronization and peak index different aspects of the dominant rhythm. The peak frequency of the dominant rhythm, measured using the CoG, was not modulated by faces, objects, and fixation conditions for all age groups, supporting that it may serve as a reliable indicator of infant brain development (Elhamiasl et al. 2024). On the other hand, dominant rhythm desynchronization demonstrated sensitivity to task stimuli, exhibiting greater desynchronization in response to faces compared to objects. Moreover, there was a developmental increase in occipital desynchronization from 6 to 9 to 12 months of age for faces over the right occipital region. Thus, dominant rhythm desynchronization may be more sensitive to condition differences. It is predicted that this sensitivity reflects the increase in cortical excitability due to information processing and enables us to investigate the development of cognitive processes such as attention (Orekhova et al. 2001; Klimesch 2012; Arutiunian et al. 2023).

Acknowledgments

L.S. Scott is the corresponding author for this paper. The authors would like to express their gratitude to Alexia Brown, Jessica Sanches Braga Figueira, and all members of the Brain, Cognition, and Development Lab for their valuable support with data collection, insightful discussions, research contributions, as well as technical and programming assistance.

Author contributions

Mina Elhamiasl (Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Software, Validation, Visualization, Writing—original draft), Maeve R. Boylan (Data curation, Formal analysis, Methodology, Software, Supervision, Validation, Writing—review & editing), Ryan Barry-Anwar (Conceptualization, Data curation, Methodology, Software, Validation, Writing—review & editing), Zoe Pestana (Data curation, Methodology, Software, Writing—review & editing), Andreas Keil (Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Resources, Software, Supervision, Validation, Writing—review & editing), and Lisa S. Scott (Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Supervision, Validation, Writing—review & editing).

Funding

Funding for this research was provided to L.S. Scott and A. Keil from a National Science Foundation (BCS:1728133). Also, M. Elhamiasl was awarded the University of Florida Graduate School Funding Award (GSFA) and the Jacquelin Goldman Dissertation Fellowship from the Department of Psychology at the University of Florida. M.R. Boylan was supported by the UF Substance Abuse Training Center in Public Health from the National Institute on Drug Abuse (NIDA) of the National Institutes of Health under award number T32DA035167.

Conflict of interest statement

None declared.

References