Mapping neural signatures of face processing in young children: an OPM-MEG study

Abstract

Facial expressions are fundamental to social communication, with emotional face processing developing throughout childhood. However, the neural mechanisms underlying this process in young children remain underexplored due to challenges in neuroimaging this population. Optically pumped magnetometers (OPMs), a wearable magnetoencephalography (MEG) technology, offer potential advantages for studying these responses early in life. This study investigated evoked responses and functional connectivity in 45 children (3–5 years) during an emotional face processing task using OPM-MEG. The M170 component, a key marker of face processing, and whole-brain functional connectivity of eight regions of interest were assessed. Children exhibited a robust M170 response to emotional faces in the bilateral fusiform gyri. Peak amplitude increased with age, but no significant latency changes were observed. A significant network of increased connectivity following emotional face onset, involving connections between the amygdalae, insulae, occipital, and frontal regions was found. This study provides the first evidence of M170 responses and large-scale connectivity to emotional faces in young children using OPM-MEG. Findings highlight the feasibility of OPMs for developmental neuroimaging and provide insights into the maturation of emotion-related neural circuits in early childhood. These results establish a foundation for future face processing research in clinical paediatric populations, such as autism.

Introduction

Facial expressions are an essential channel of nonverbal communication and provide insight into the internal states and intentions of others; thus, the interpretation of emotional information from faces is foundational for successful social functioning (Adolphs 2002, Batty and Taylor 2006, Bigelow et al. 2021). The processing of emotional faces undergoes protracted development: the ability emerges in infancy, with gains in processing speed and accuracy achieved across childhood, with proficiency for recognizing subtle, complex, and implicitly presented emotions reached in adulthood (Walker-Andrews and Dickson 1997, De Sonneville et al. 2002, Herba et al. 2006, Thomas et al. 2007).

The spatial origins of emotional face processing have been investigated for decades using functional magnetic resonance imaging (fMRI) in adults (e.g. Puce et al. 1995, George 2013). The brain regions that play an important role in the early perception of faces include the inferior occipital gyri (IOG), lateral fusiform gyri, and posterior superior temporal gyri (STG) (Haxby et al. 2002, 2000). The fusiform gyri are critical for discriminating faces from objects, face-specific information processing (e.g. face configuration, eye detection), and identity recognition (McCarthy et al. 1997, Haxby et al. 2000, 2002). Other cortical and subcortical brain regions that play key roles in face and emotional face processing include the amygdalae—important for the detection and allocation of attention to salient emotional stimuli (Whalen and Phelps 2009); limbic regions including the insulae and the anterior cingulate gyri (ACC)—part of the salience network—are critical for emotional judgement and empathy (Di Martino et al. 2009, Menon and Uddin 2010); and orbitofrontal regions—involved in the regulation of emotional information and evaluation of reward (Rolls 2000).

Neurophysiological studies, including electroencephalography (EEG) and magnetoencephalography (MEG), have bolstered this body of research as these techniques provide direct measures of neural activity and excellent temporal information, with MEG additionally offering high spatial resolution (Hari and Salmelin 2012). Event-related potential (ERP) studies have found a robust negative deflection component at approximately 170 ms following the onset of face stimuli which occurs over occipitotemporal regions (Bentin and Deouell 2000, Eimer 2000, Itier and Taylor 2004, Rossion 2014), known as the N170 (Bentin et al. 1996, Eimer 2000, George 2013). The component is larger over the right compared to the left hemisphere, is larger for faces compared to non-face objects (Eimer 2000, Itier and Taylor 2002), and has been shown to be modulated in response to facial expressions of emotion (Batty and Taylor 2003, 2006, Rossignol et al. 2005, Leppänen et al. 2007). An analogous component—the M170—is seen in MEG studies of face processing, with similar response properties as in ERPs (Halgren 2000, Taylor et al. 2001a). The sources of the N170/M170 have been localized to the bilateral occipitotemporal cortices, and posterior fusiform gyri (Rossion 2003, Itier et al. 2004, Deffke et al. 2007). MEG source localization studies demonstrate robust neural activation of these areas in adults, particularly the fusiform gyri, to face and emotional face stimuli (Liu et al. 2000, Monroe et al. 2013).

ERP studies of face and emotional face processing in young children have demonstrated an N170 response over occipitotemporal scalp by 4 years of age (Taylor et al. 1999, 2004, Batty and Taylor 2006, Dennis et al. 2009, Kuefner 2010), and this response demonstrates marked age-related changes. Studies have shown a decrease in latency from young childhood to adulthood, and more negative amplitude with age. These maturational changes are consistent with the prolonged developmental course and changes in the efficiency of face-processing with age (Taylor et al. 1999, Itier and Taylor 2004, Batty and Taylor 2006, Hileman et al. 2011). The few MEG studies that have examined evoked neural activity and source localized responses in children (6–13 years) have not found an M170, but a component peaking at 140 ms (M140), seen in 6- to 13-year-olds (Taylor et al. 2010) and 8- to 11-year-olds (Kylliäinen et al. 2006a, 2006b); the M170 was evident starting in the teenage years (Taylor et al. 2010). These studies suggest developmental changes in the neural sources of the M170, such that it is not as readily recorded using MEG in children. A very small number of MEG studies in young children (3–6 years) using child-sized MEG systems, however, have shown a reliable M170 face response (He et al. 2015, He and Johnson 2018, Chen et al. 2022). Despite these findings, there remains a substantial gap in research examining the development of the M170 component in early childhood (e.g. 3–5 years) using MEG.

Although MEG is optimal for investigating source localized neural activity and functional networks, providing high temporal and spatial resolution (Hari and Salmelin 2012, Baillet 2017), there are several challenges associated with recording neural activity in young children using traditional cryogenic MEG. First, the ‘one size fits all’ dewar of cryogenic MEG is built for ∼95% of adults, and thus in young children with smaller head sizes, the signal strength is reduced, and there is inhomogeneous coverage across the head (Brookes et al. 2022).

In addition, participants must remain completely still during testing, which poses a challenge to obtaining artefact-free data in young children.

Optically pumped magnetometers (OPMs)—a new wearable MEG technology—provide potential advantages for recording neural responses in very young children, due to enhanced signal strength, data quality, adaptability to head size, and tolerance to movement compared to conventional MEG systems (Boto et al. 2018, Hill et al. 2019, Rier et al. 2024). OPM sensors offer comparable sensitivity to conventional superconducting quantum interference device (SQUID) sensors used in traditional MEG systems (Allred et al. 2002), which is achieved through the manipulation of the quantum properties of alkali atoms to measure magnetic fields of the brain (Tierney et al. 2019, Schofield et al. 2023), rather than cryogenic cooling. Therefore, OPM sensors can be situated closer to the surface of the head and flexibly positioned, which offers improved sensitivity, signal strength, and greater adaptability in coverage (Boto et al. 2016, Hill et al. 2020, Brookes et al. 2022). Furthermore, with background fields controlled, OPMs are tolerant of participant head movement, which is a key advantage when recording neural responses from young children (Holmes et al. 2018, Rea et al. 2021). The advent of OPMs has revolutionized MEG studies in young children, making MEG now feasible across the lifespan.

To address our limited knowledge of the neural mechanisms underlying emotional face processing in young childhood, we are the first to leverage OPM-MEG to characterize the development of emotional face processing in 3-to 5-year-old children (n = 45). The current study investigated evoked responses and whole-brain functional connectivity with eight regions of interest (ROIs) in 3-to 5-year-old typically developing children. Whole-brain functional connectivity based on phase synchrony was investigated to further characterize the M170 evoked responses, as phase synchrony mediated interregional connectivity supports dynamic communication and coordination of information transfer across the brain regions involved in perceptual and cognitive processing (Fries 2005, 2015). As ROIs, we selected the bilateral fusiform gyri, amygdalae, insulae, and ACC, as these regions are known to underpin the early perception and attention to emotional faces (Adolphs 2002, Haxby et al. 2002). Previous findings from our group have also demonstrated the important role of these regions in emotional face processing networks in typical and atypical development, from 7 years of age (Safar et al. 2021, 2018). Given OPM-MEG offers tremendous improvements surrounding recording neural responses in young children compared to traditional MEG, it was hypothesized that children would demonstrate the classic M170 evoked response to emotional faces in the bilateral fusiform gyri, with amplitude of the M170 component increasing and M170 latency decreasing with age. It was also expected that children would recruit an emotional face network, primarily involving connections among the eight ROIs and linked to other face processing areas (e.g. primary visual, temporal, and limbic regions).

Methods

Participants

Data from forty-five 3- to 5-year-old typically developing children (M age = 4.41 years, SD = 0.83), 21 [47%] boys were included in the analyses: 3-year-olds (n = 17; M age = 3.52 years, SD = 0.33; 8 boys), 4-year-olds (n = 15, M age = 4.57 years, SD = 0.33; 8 boys), 5-year-olds (n = 13; M age = 5.38 years, SD = 0.26; 5 boys). Data from five additional participants were excluded from analysis due to insufficient trials following pre-processing (<20). Exclusion criteria for the study included neurological or neurodevelopmental disorders, brain injury, visual impairment, colour blindness, and children born very preterm (<32 weeks gestation). All participants were tested at the Hospital for Sick Children in Toronto, Canada. The study protocol was approved by the Hospital for Sick Children Research Ethics Board. All parents provided written informed consent according to the Declaration of Helsinki, and children provided verbal assent.

Emotional faces task and OPM-MEG data acquisition

During the OPM recording, participants were presented with an emotional faces task (Fig. 1). A happy or angry face was presented on each trial for a duration of 500 ms followed by an inter-stimulus interval (i.e. fixation cross) with a jittered duration (1250 ± 200ms). Faces were extracted from the NimStim Set of Facial Expressions (Tottenham et al. 2009). The children were re-directed to the screen, as needed, to ensure task attention via an attention-grabbing stimulus that appeared in the centre of the screen between every second trial prior to the fixation cross. Cartoon characters were randomly presented on about 10% of trials to also maintain attention. A total of 80 randomized trials (40 faces per emotion) were presented using Presentation software (Neurobehavioural Systems, California, USA). For more details about the emotional faces task used in the current study, see Safar et al. (2024).

Figure 1.

Emotional faces task. Participants were presented with a happy or angry face on each trial for 500 ms; a fixation cross followed with a jittered duration (1250 ± 200 ms). An attention-getting stimulus appeared between trials to redirect the child’s attention to the screen; intermittently presented cartoon characters also helped keep children’s attention on the stimuli.

Evoked fields to emotional faces were recorded using an 80-channel OPM-MEG system (40 dual-axis zero field magnetometers, QuSpin Inc, USA and Cerca Magnetics Limited, UK). Data were recorded from each OPM channel using a digital acquisition system (National Instruments, Texas, USA), with a 1200 Hz sampling rate in a magnetically shielded room (MSR; Vacuumschmelze, Hanau, Germany). OPMs were mounted in one of two rigid 3D-printed helmets according to participant head measurements (i.e. circumference, nasion to inion, and pre-auricular to pre-auricular). Head circumference measurements were collected for 35 participants, Mcircumference = 51.07 cm; range = 46–54.5 cm. Sensors were distributed approximately uniformly across the whole-head. Wearing the helmet, participants were seated in front of a screen at a viewing distance of 110 cm, such that each image subtended approximately 7.8 × 8.8 degrees of visual angle. A pair of bi-planar coils and reference triaxial OPM sensors were positioned on either side of the participant to control the surrounding magnetic field (Holmes et al. 2018, 2019, QuSpin Inc. triaxial zero field magnetometers). The coils were wound on two 1.6 m square planes, separated 1.5 m apart, and generate three orthogonal magnetic fields and all five independent first-order gradients, to keep static and gradient magnetic field magnitudes below ∼1nT within a 40-cm cube surrounding the participant’s head (Holmes et al. 2018, Hill et al. 2022). This set-up maintained sensor operation within ±3.5nT by dynamically compensating for the background magnetic field and its drift (Safar et al. 2024). A four-camera system (OptiTrack Flex 13, Natural Point Incorporated, Oregon, USA) with infra-red markers placed on the helmet was used to track continuous head movement (head movement data were not obtained for three participants). There was no significant association between head motion and age (r = −0.16, P = .32). An empty room noise recording was acquired prior to recording participant data. Following data acquisition, digitization of the OPM sensors relative to the helmet was acquired using a three-dimensional optical imaging system for co-registration (Einscan H, SHINING 3D, Hangzhou, China).

OPM-MEG data preprocessing

All data preprocessing was performed using an in-house OPM pipeline using the FieldTrip toolbox (version 2022-02-14; Oostenveld et al. 2011) implemented in MATLAB (version R2024b) (MathWorks, Massachusetts, USA). Trials were epoched from −1000 to 1500 ms relative to face onset. Bad channels were identified and removed using an outlier detection algorithm (Safar et al. 2024) and visual inspection; the number of channels removed did not differ between age groups (3 years: M = 1.47 SD = 1.8; 4 years: M = 2.07 SD = 1.91; 5 years: M = 1.67 SD = 1.86; F(2, 42) = 0.51, P = .60). Homogenous field correction was applied to attenuate external interference (Tierney et al. 2021, Seymour et al. 2022). Data from remaining channels were band-pass filtered between 1 and 150 Hz (fourth order, two-pass Butterworth). To suppress interference (i.e. powerline noise at 60 Hz and 120 Hz and additional interference from electrical equipment), noise peaks at frequencies found in both the power spectra of the empty room and participant data recordings were identified and band-stop filtered (fourth or third order, two-pass Butterworth; Safar et al. 2024). Independent component analysis was applied to remove ocular and cardiac artefacts and rejected components were removed based on visual inspection. Artefact rejection was then performed for trials with artefacts that exceeded an adjusted artefact rejection threshold of 4000fT, which is common in studies of young children using conventional MEG (He et al. 2015, Partanen et al. 2017, Alho et al. 2023). This threshold was adjusted by a factor of 13.78 to account for the increased noise floor in OPM data (see Safar et al. 2024), yielding an OPM threshold of 55.12pT. The number of trials included in the analyses post-artefact rejection did not differ between age groups (3 years: M = 57.18 SD = 20.33; 4 years: M = 54.00 SD = 20.54; 5 years: M = 60.92 SD = 16.03; F(2, 42) = 0.45, P = .64). Differences between happy and angry faces were not examined due to 12 children having fewer than 20 trials in each condition, which was the minimum trial count for participant inclusion. A 20-trial cut-off was chosen across emotions to maximize the number of participants included in the analyses and is consistent with previous MEG/EEG studies in children and adults (Fogelson et al. 2019, Márquez‑García et al. 2022).

Source estimation

A single-shell head model (Nolte 2003) was constructed based on age-specific template T1-weighted magnetic resonance images (MRIs; Richards et al. 2016) and used to generate a 2 mm grid in MNI space for each subject. Source reconstructed OPM-MEG data modelled with and without an individual MRI have been shown to be highly similar (Rhodes et al. 2025). The participant head models and source coordinates were co-registered with participant data using surface meshes of the age-specific MRIs. This was conducted using participants’ optical head digitalizations acquired after scanning (Hill et al. 2020, Rhodes et al. 2023). Optical head digitalizations were not acquired for two participants; for these children a perfect fit was assumed. Following co-registration, data were bandpass filtered to the frequency band of interest: 4 to 40 Hz (fourth order, two-pass Butterworth). A lower frequency cutoff of 4 Hz was used given the higher inherent noise level of the OPM-MEG at low frequencies relative to SQUID-MEG, that may obfuscate neuromagnetic fields (Boto et al. 2022). It has been demonstrated that visual evoked responses can be reliably measured at 4 Hz using OPM-MEG, with comparable signal to noise ratio to SQUID-based MEG (Rhodes et al. 2023). Continuous data were used to generate a covariance matrix, with 2% regularization using the Tikhonov method (Tikhonov 1943).

For the evoked analyses, source activity was then computed across the 2 mm grid using a linearly constrained minimum variance (LCMV) beamformer (Van Veen et al. 1997). The voxel whose timeseries showed the maximum percentage change between active (100–400 ms) and baseline (−300 to 0 ms) time windows was selected for each of the 90 parcels in the Automated Anatomical Labelling (AAL; Tzourio-Mazoyer et al. 2002) atlas. For the connectivity analyses, the broadband timeseries were reconstructed using an LCMV beamformer for the centroid of each of the AAL sources, which were used as seed locations.

M170 evoked emotional face response and functional connectivity

To investigate the M170 component, the timeseries of the bilateral fusiform gyri were then extracted across emotional face type and z-scored by the baseline window to determine peak amplitudes and latencies. Prior to statistical analyses the timeseries data were multiplied by −1 in the left or right hemisphere so that the M170 response for each subject was in the negative direction consistent with the EEG literature (Bentin and Deouell 2000, Eimer 2000, Itier and Taylor 2004, Rossion 2014).

To compute broad band functional connectivity (4–40 Hz), the Hilbert transform was used to generate instantaneous phase values at each sample across the entire recording at each of the AAL sources. Phase-based functional connectivity was computed pairwise using the cross-trial weighted phase lag index (wPLI) (Vinck et al. 2011). The pairwise timeseries of wPLI values between eight ROIs (bilateral fusiform gyri, amygdalae, insulae, and ACC) and each of the other 90 AAL cortical and subcortical sources were extracted yielding a 90-by-8 connectivity matrix for each participant. The wPLI values were z-scored relative to a baseline period of −200 to 0 ms. A time window of 200–400 ms was selected for statistical analyses; z-scored wPLI values were averaged across this latency window. This latency window was selected based on visual inspection of the z-scored mean connectivity strength across the 90-by-8 connectivity matrix averaged across participants, demonstrating a peak in wPLI between 200 and 400 ms. The latency window was also selected based on previous MEG work from our group demonstrating increased functional connectivity to emotional faces during this latency period (Safar et al. 2021, 2022, 2023).

Statistical analyses

For the M170 component, the peak latency and amplitude were extracted for each participant and analysed using repeated-measures ANOVAs with hemisphere (left and right) as a within-subject factor and age as a continuous between-subject factor.

For functional connectivity, a paired t-test was conducted using the network-based statistic (NBS) to determine a significant difference in connectivity between the active and baseline windows. Additionally, we conducted correlation analyses in NBS to examine age-related changes in functional connectivity. NBS is a non-parametric method that is validated for the analysis of large networks, while accounting for the family-wise error rate (FWER) (Zalesky et al. 2010, 2012). We selected the primary component-forming threshold based on the sparsity of network, such that the network included 5% of total possible pairwise connections.

Results

M170 evoked responses

Children demonstrated a mean z-scored peak amplitude of −4.45 (SD = 2.10) in the left fusiform gyrus and −4.91 (SD = 2.66) in the right fusiform gyrus (Fig. 2a and b). We found a significant main effect of age on amplitude (F(1,43) = 5.27, P = .027), with amplitude becoming larger with age (Fig. 2c). There was no significant main effect of hemisphere, nor was there a hemisphere-by-age interaction.

Figure 2.

Grand averaged M170 evoked responses by age. The grand averaged z-scored evoked responses across emotional faces in the left fusiform gyrus (a) and right fusiform gyrus (b) for each age group. The x-axis represents time in milliseconds and the y-axis represents z-scored amplitude. The significant main effect of age on amplitude is plotted, with amplitude becoming larger with increasing age (c). The x-axis represents age in years and the y-axis represents z-scored amplitude.

For peak latency, children demonstrated a mean peak latency of 165.40 ms (SD = 24.82) and 171.10 ms (SD = 28.91) in the left and right fusiform gyri, respectively (Fig. 2a and b). No significant main effects of age or hemisphere on latency were found, nor was there a hemisphere-by-age interaction (Fig. 2).

Functional connectivity

A network of significantly increased functional connectivity was found 200–400 ms following emotional face stimulus onset (33 edges, 30 nodes, pFWER < .001) compared to baseline (0 to −200 ms). The primary hubs of the network involved the bilateral amygdalae and insulae with most connections to occipital and frontal brain regions, and the left ACC with connections to temporal, parietal, frontal, and visual regions. Most connections were observed between bilateral limbic and right frontal areas and right occipital areas, including the right inferior occipital gyrus and right calcarine. In addition, several orbitofrontal brain regions were recruited in the network, including the right medial orbitofrontal gyrus and right superior orbitofrontal gyrus (Fig. 3). No significant correlations between age and functional connectivity were found (pFWER > .05).

Figure 3.

Increased within-group functional connectivity. A significant within-group network of increased functional connectivity 200 to 400 ms following emotional face onset compared to baseline (0 to −200 ms) in 3- to 5-year-old children. The network is represented in the glass brain; node size is scaled by degree.

Discussion

The current study is the first to investigate the temporal and spatial dynamics of neural mechanisms underlying emotional face processing in 3- to 5-year-old children using OPM-MEG; several findings were observed. First, children demonstrated a robust evoked response following the onset of emotional faces in the bilateral fusiform gyri. Second, we observed a significant negative association between age and mean peak amplitude, with the magnitude of the response becoming larger with increasing age. Lastly, functional connectivity results revealed a network involving primarily connections between limbic, occipital, and frontal regions following the onset of emotional faces, indicating that core and extended brain areas (Haxby et al. 2000, 2002) are recruited by emotion-related circuitry in early development.

Only a handful of studies have investigated face-sensitive neural responses with MEG in young children (Taylor et al. 2010, 2011, He et al. 2015, He and Johnson 2018). Kylliäinen and colleagues (Kylliäinen et al. 2006a) investigated the neural mechanisms underlying face processing in a sample of 8- to 11-year-old children and adults. The authors found a face-sensitive response in adults at approximately 135 ms; however, in children, the same response was less prominent, not right lateralized, and not face sensitive. Consistent with this finding, Taylor and colleagues (2010) examined face responses using MEG in children 6–16 years of age and young adults and did not find an M170 component until the late teens and in adults. Instead, a robust M140 component was found in the children, not seen in the adults. Given that the N170 ERP component is often bifid in children (Taylor et al. 2001b, Taylor et al. 2004), they hypothesized that M140 may reflect the neuromagnetic counterpart of the earlier component of the bifid N170.

Recording neural responses in children using the adult-sized conventional MEG system poses many challenges. The rigid sensor placement within the ‘one size fits all’ dewar designed to fit adults results in a larger gap between the MEG sensors and brain sources in children, leading to a reduction of signal strength and inhomogeneous head coverage. The larger gap between the head and the dewar also allows for increased head motion. Therefore, some studies examining face responses in young children have turned to using child-sized MEG systems to attempt to overcome these challenges (He et al. 2015, He and Johnson 2018, Chen et al. 2021, 2022). Chen and colleagues (2022) investigated bilateral fusiform gyri activity to faces longitudinally from 4 months to 4 years of age using such a system, demonstrating more adult-like face responses by approximately 3 to 4 years of age, with an M170 latency of 168 ms. He and colleagues (2015) also found an M170 response to faces in 3- to 6-year-old children using a paediatric MEG system; when compared with the M170 in adults tested using a conventional system, the responses were larger, broader, and delayed in the children. Using OPM-MEG, the current findings extend these studies, such that we found clear M170 responses to emotional faces in the bilateral fusiform gyri in young children 3, 4, and 5 years of age that resembled the M170 in adults and showed age-related changes. This is likely accounted for by improvements in the fit of the sensors for child-sized heads and reduced motion effects leading to better uniformity of sensor coverage, source modelling, and signal quality.

In our study, young children demonstrated an age-related increase in the amplitude of the M170 component. Previous ERP studies of emotional faces have shown that the development of the N170 amplitude follows an inverted-U-shaped trajectory, increasing over young childhood, decreasing over middle childhood, before increasing again in adolescents through to adulthood (Taylor et al. 1999, 2004, Batty and Taylor 2006, Kuefner 2010). In concordance, the current findings demonstrate a similar pattern of increased M170 amplitude from 3 to 5 years of age, which may reflect the increasing emergence of this component during early childhood from the infant N290 and P400 precursors (Chen et al. 2021). Next steps include examining whether a decrease in amplitude would be observed in mid-childhood using OPM-MEG, as seen in previous ERP work (Taylor et al. 2004, Batty and Taylor 2006, Kuefner 2010).

Contrary to our hypothesis, we did not find age-related differences in latency of the M170. Previous neurophysiological studies have shown decreased latency to faces with increasing age from 4 years to adults, indicating maturational changes in processing speed (Taylor et al. 1999, 2004, Batty and Taylor 2006, Chen et al. 2021). It is possible that the age range in the current study was too narrow to capture these developmental latency changes. Consistent with this hypothesis, a previous ERP study did not find age-related changes in N170 latency to emotional faces between 3- to 4-year-old and 5- to 6-year-old children; however, it did observe differences between 3- to 4-year-olds and 7- to 8-year-olds (Vlamings et al. 2010). As we saw amplitude changes with age, however, it may be that they are more marked in this age range. Future research using a broader age range of children is necessary to delineate these incremental age-related changes.

In addition, we observed a network of increased functional connectivity 200–400 ms following the onset of emotional faces relative to baseline. The network recruited seven out of the eight selected regions known to be important in emotional face processing (Haxby et al. 2000, 2002, Adolphs 2002, Menon and Uddin 2010). Most connections stemmed from the bilateral amygdalae and insulae to the right occipital (i.e. right calcarine and inferior occipital gyrus) and occipitotemporal (i.e. right lingual) and orbitofrontal brain regions (i.e. right medial and superior orbitofrontal gyri). The amygdalae are critical for the detection, evaluation, and allocation of attention to salient emotional faces (Adolphs et al. 1995, Morris et al. 1996, Thomas et al. 2001, Davis and Whalen 2001) and have a top-down influence on occipital and occipitotemporal areas, enhancing detailed perceptual processing of emotional stimuli (Bar et al. 2006, Murray and Izquierdo 2007, Furl et al. 2013). Several connections between orbitofrontal areas and the right amygdala were also observed in the network. Research in adults suggests that bidirectional reciprocal connections between the amygdalae and the orbitofrontal cortex serve to facilitate visual recognition of emotional stimuli (Frank et al. 2019). The bilateral ACC were also highly connected to widespread brain areas. The ACC and the insulae comprise the salience network, which is pertinent to the assessment of salient emotional stimuli, judgement of one’s own and others’ emotional states, and empathy (Di Martino et al. 2009, Menon and Uddin 2010).

Studies using MEG to investigate functional connectivity in preschool aged children to faces are scarce. He and Johnson (2018) used dynamic causal modelling to examine effective connectivity among six bilateral core face ROIs: superior temporal gyri, fusiform gyri, and inferior occipital gyri) in 10 young children (3–6 years), and 11 adults during face repetition processing. The results showed that adults demonstrated intrahemispheric feed-forward connections from the bilateral inferior occipital gyri to the fusiform and superior temporal sulci. Compared to adults, young children had an extra connection between the inferior occipital gyrus and the contralateral fusiform gyrus; the authors interpreted this to suggest continued maturation and fine-tuning of connections from childhood to adulthood. The current study extends these findings by demonstrating the involvement of a whole-brain functional network underpinning emotion processing in preschool-aged children. Consistent with our hypothesis, bilateral connections between core (e.g. occipital inferior gyri), and extended higher-order brain areas (e.g. temporal, limbic, and frontal) were found to support emotional face processing by 3 years of age.

Overall, our results are the first to demonstrate reliable face-sensitive responses and whole-brain functional connectivity in young children using OPM-MEG. These results support the feasibility of recording robust neural responses from young children using OPM-MEG and add to the growing literature examining the early development of face processing. We showed that OPMs are sensitive to the development of face processing, with results demonstrating a significantly evolving response in bilateral fusiform with age. The present findings are foundational for future research characterizing the early neural mechanisms underpinning both typical emotion processing, as well as being a basis for future clinical studies such as face processing and pareidolia in neurodivergent populations. For example, a recent EEG study demonstrated differences in delta and theta frequency bands between children and youth with and without autism during processing of pareidolia face images (Akdeniz 2025). The assessment of neural mechanisms in young children using OPMs could provide insights into the early identification of differences in face processing and inform interventions aimed at improving social perception and cognition.

Author contributions

Kristina Safar (Formal analysis [lead], Resources [supporting], Visualization [lead], Writing—original draft [lead]), Marlee Vandewouw (Formal analysis [supporting], Software [lead], Writing—review & editing [supporting]), Natalie Rhodes (Resources [supporting], Software [supporting], Writing—review & editing [supporting]), Julie Sato (Data curation [supporting], Resources [supporting], Writing—review & editing [supporting]), and Margot Taylor (Conceptualization [lead], Funding acquisition [lead], Resources [supporting], Writing—review & editing [supporting])

Supplementary data

Supplementary data is available at SCAN online.

Conflict of interest: The authors have no conflicts of interest to declare.

Funding

Funding was provided by the Canadian Institutes of Health Research (PJT-178370) and the Simons Foundation Autism Research Initiative (SFARI; 2021 Human Cognitive and Behavioural Science award).

Data availability

The data underlying this article will be shared on reasonable request to the corresponding author.