The human brain consists of large-scale spatially distributed, but functionally connected, networks that collectively work to perform complex cognitive tasks such as speech perception and language comprehension [1–4]. As a primary sensory function, hearing plays an important role in shaping how children learn to speak and develop language skills. Previous studies have shown that development of language areas in early childhood could be severely altered in hearing impaired infants due to weakened auditory inputs [5–7]. Understanding the underlying functional organization of language areas enhances our knowledge of language processing mechanisms not only in healthy normal hearing population but also in people with different types of hearing impairments [8–10].

Task-based functional magnetic resonance imaging (fMRI) has been widely used to measure the brain activations to speech sounds [11, 12]. The studies conducted on infants using fMRI have reported significant responses in the regions associated with language processing including inferior frontal gyrus (IFG) and superior temporal gyrus (STG) [12, 13]. These results suggest that language networks are capable of processing basic acoustic information in the first few months of life and continue to develop from infancy to childhood enabling more complex and integrated information processing [2, 11–14]. There has been a rapid growth in resting-state fMRI studies in which spontaneous hemodynamic fluctuations are used to characterize functional networks of the brain at rest [15–18]. Resting-state functional connectivity analysis is a promising approach to investigate development of neural systems in early childhood as it is difficult for infants or young children to engage in active or passive task-related experimental paradigms. Despite many advantages, the use of fMRI in developmental studies related to infant hearing is limited by factors such as extremely confined scanning environment requiring high level of tolerance and cooperation from the participant and the acoustic noise of the scanner [19].

Functional near-infrared spectroscopy (fNIRS) is a non-invasive functional neuroimaging technique that can be used for resting-state functional connectivity analyses [20–22]. fNIRS measures the relative changes of oxy-hemoglobin (HbO) and deoxy-hemoglobin (HbR) concentrations similar to the hemodynamic response measured in terms of blood-oxygen-level-dependent (BOLD) signals in fMRI, but offers much richer time resolved information of neural activity with higher temporal resolution [23]. Compared to fMRI, fNIRS has been widely adopted for studies involving infants and young children as a neuroimaging tool that is less expensive, child-friendly, and more robust against motion artifacts [19, 24, 25]. fNIRS is also a silent functional imaging technique, hence preferred for resting-state studies investigating neural correlates of language development. Consistent with the findings from fMRI studies, several fNIRS studies have demonstrated the presence of resting-state networks in sensorimotor, visual, and auditory systems in infants as well as adults [26–28].

Many structural and functional neuroimaging studies have shown that the spatial and temporal evolution of brain networks happens across our entire lifespan [12, 18]. However, a rapid development of the human brain in the first few years of life is expected as a result of the prolonged postnatal maturation of structural connections and exposure to a wide range of novel sensory inputs [15, 17]. The findings of recent fMRI studies suggest that primary networks such as sensorimotor and visual networks are relatively well established by the end of the first year [17]. A previous fMRI study conducted on 2-day-old neonates has revealed the presence of relatively strong inter-hemispheric homologous connectivity between bilateral language areas and weak intra-hemispheric connectivity between prefrontal and temporal regions [12]. A similar study using adults has shown that intra-hemispheric networks form stronger connections in adults compared to infants [15]. Although, evidence from a large body of literature suggests that much of the development of language regions happens in early childhood, it remains largely unclear how language areas of a newborn evolve toward its mature form during infancy. Moreover, having the knowledge of typical developmental trajectory of language areas of normal hearing infants could lead us to better understand the developmental changes in hearing-impaired infants and the effects of altered connectivity on language delays.

The above findings lead to the hypothesis that as long distance structural connections continue to mature in early childhood [12, 15, 16], functional connectivity would increase with age for both inter-hemispheric and intra-hemispheric language areas. In this study, we conducted an experiment to investigate the developmental trajectories of language areas in a cross-sectional population in the first year of life. Functional connectivity strength, calculated using magnitude-squared coherence of channel pairs of resting-state fNIRS data, was used to assess developmental effects in inter-hemispheric and intra-hemispheric language areas. Regions of interest in the temporal and prefrontal lobes were selected for this study. Inter-hemispheric connectivity was measured between homologous regions across hemispheres, and intra-hemispheric connectivity was measured between the two regions of interest in the same hemisphere. A control group was defined using channel pairs located in inter-hemispheric non-homologous regions that have no known direct structural connections. We hypothesized that functional connectivity, characterized by coherence of inter-hemispheric homologous and intra-hemispheric test groups, would increase relative to coherence of the control group as age increased in the first year of life.

2.1. Participants

Twenty-six infants (11 males and 15 females, mean and standard deviation of corrected age: 196.0 ± 98.8 days, age range: 68–392 days) and 12 healthy adults (six males and six females, mean and standard deviation of age: 32.1 ± 4.5 years, age range: 25–40 years) participated in our study. Four babies were born pre-term (gestational age <38 weeks) as reported in table 1. All infants were included in the analysis by using the chronological age for full-term babies and corrected gestational age for pre-term babies. Based on the information provided by the parent/guardian, the infants had no major complications during the pregnancy or delivery and no known developmental or neurological disorder at the time of testing.

Table 1. Demographics of infant participants in ascending order of the corrected age. Details of four pre-term babies are highlighted in bold font. M—Male, F—Female.

Participant #Participant codeGenderChronological age at test (days)Corrected age at test (days)1C038F68682C030M76763C072M93934C036M95955C042F9898 6 C034 F 127 99 7C045M1181188C029F1251259C028F12612610C084M12712711C076F13713712C032F180180 13 C054 M 212 184 14C033F18518515C041F18718716C052F199199 17 C053 M 269 199 18 C031 M 252 224 19C078F22822820C040M28228221C050F29529522C063F30830823C086M33033024C047F36936925C056M37137126C073F392392

All infants had passed either a newborn hearing screening test or diagnostic audiology assessment. Tympanometry (probe tone 226 Hz or 1000 Hz for infants below 6 months) was performed on the day of testing to exclude the possibility of temporary conductive hearing loss. Pure tone audiometry was performed for octave frequencies between 500 Hz and 8000 Hz on adults and all participants had thresholds of 20 dB HL or better on the day of the experiment. All adults declared that they had not been previously diagnosed with hearing loss or cognitive/neurological impairment.

The experimental study was approved by the human research ethics committees of the Royal Victorian Eye and Ear Hospital (16/1261H) and the Royal Children's Hospital (71941) in Melbourne. Written informed consent was obtained before the experiment from the parent or guardian for infants and adult participants directly.

2.2. Experimental procedure and data acquisition

fNIRS recordings were conducted in a dimly lit sound attenuated room while infants were in their natural sleep and adults were in wakeful rest. The parent or the guardian sat on a comfortable armchair cradling the baby and were given time to feed and settle their baby before the start of the experiment. Parents and adult participants were instructed to limit their body movements and remain stationary as much as possible during the testing. Five minutes of resting-state data were recorded for each participant of infant and adult cohorts and the experimenter monitored the movements of the babies throughout the testing session.

Resting-state data were acquired using the NIRScout (NIRx Medical Technologies, LLC, USA) continuous-wave fNIRS system. This system comprises light emitting diode sources to emit near-infrared light using two distinct wavelengths (760 nm and 850 nm) and avalanche photodiodes as detectors to capture the backscattered light from the cerebral cortex. The optode montage had eight sources and eight detectors placed over bilateral temporal and prefrontal regions forming 18 fNIRS measurement channels as shown in figure 1. The sampling rate of each channel was 7.8125 Hz. Sensitivity profile of the fNIRS channels was generated using Monte Carlo simulations in AtlasViewer [29].

Figure 1. Illustration of the optode montage used in this study. The locations of optical sources and detectors are shown in (a) left and (b) right hemisphere of the brain (on Colin27 brain atlas) with respect to the landmarks of international 10–10 standard. Eight sources and eight detectors placed over bilateral temporal and prefrontal regions are marked by red and blue filled circles, respectively. fNIRS channels are shown in solid yellow lines. Sensitivity profile of the fNIRS channels is presented using a heatmap with blue and red colors representing low and high sensitivity, respectively. This figure was generated using Monte Carlo simulations in Matlab-based toolbox; AtlasViewer.

Download figure:

Standard image High-resolution image

The head circumference of the participants was measured before the testing started, and an appropriately sized flexible headcap (EasyCap, Brain Products GmbH, Germany) was fitted for each participant. The headcap was correctly positioned using the anatomical landmarks (i.e. nasion, vertex, inion, left preauricular point and right preauricular point) and the positions of optodes were checked based on the international 10–10 system by an experienced experimenter. Flat tip optodes were used on infants with a source-detector separation of 2–3 cm as per the recommendations of NIRx for enhanced comfort. Dual tip optodes were mounted on the headcap of adult participants with a source-detector separation of 3–4 cm. Registration of fNIRS channel locations based on the 6-month-old infant AAL atlas and the specificity values obtained using devfOLD toolbox [30] are presented in the supplementary materials (table S1).

2.3. Data analysis

Data analysis was performed exclusively in Matlab R2021b (Mathworks Inc., USA). Resting-state data were pre-processed using the NIRS Brain AnalyzIR Toolbox (also known as nirs-toolbox, version 837), an open-source package for fNIRS data processing [31]. A mix of custom scripts and in-built Matlab functions were used for functional connectivity analyses and statistical analyses. The block diagram in figure 2(a) shows the main steps of the data processing pipeline. A common pre-processing pipeline was employed for both infant and adult fNIRS data, although some parameters were different due to the inherent differences between these two cohorts as explained in section 2.3.1. For each participant, resting-state data of 3.5 min (210 s) were used for the analysis after excluding the first 30 s of data. The aim of this exclusion was to limit the possibility of having undesirable artifacts in signals potentially due to movement of cables if the participant was trying to settle down at the start of the experiment. The length of the dataset used in our study is consistent with previous fNIRS resting-state infant studies [21, 22].

Figure 2. (a) Block diagram of the data processing pipeline of resting-state fNIRS data. Signal quality assessment for each participant of (b) infant and (c) adult cohorts. Each dot represents the scalp coupling index of an fNIRS channel. The gray dashed line shows the threshold of the scalp coupling index used; 0.8 and 0.5 for infant and adult cohorts, respectively. (d) Channel pair definition of the connectivity groups; inter-hemispheric homologous group pairing channels in homotopic regions across hemispheres, intra-hemispheric group pairing channels in ipsilateral temporal and prefrontal regions and inter-hemispheric non-homologous group comprising contralateral long distance connections that have no known direct structural links.

Download figure:

Standard image High-resolution image 2.3.1. Data pre-processing

Two signal quality metrics were used to identify fNIRS channels with poor signal quality. The first metric was calculated during the calibration stage using a short segment of resting-state data by considering the gain, signal intensity levels and estimated noise levels at two wavelengths. The noise was defined as the coefficient of variance (standard deviation/mean value×100). Any channel with signal quality marked as critical (gain: 0 or 8, signal intensity level: 0.01–0.03 or >2.5 V and noise: >7.5%) by the signal acquisition software NIRStar (NIRx Medical Technologies, LLC, USA) were noted down and removed from the analysis. Scalp coupling index [32] which makes use of the correlation between time courses of two wavelengths filtered in the cardiac frequency band to infer the level of contact between the skin and the optical interface was used as the second metric. The cardiac frequency band was designated as 1.5–3 Hz and 0.5–2 Hz for infants and adults, respectively. Resting-state data were epoched by including dummy triggers at 20 s intervals and the scalp coupling index was calculated for each epoch. The averaged scalp coupling index value over epochs was used as the representative quality metric for each channel as shown in figures 2(b) and (c). Any channel with scalp coupling index below 0.8 for infants and 0.5 for adults was rejected and excluded from the functional connectivity analysis. Setting these threshold values for scalp coupling index ensured that all participants retained a minimum of 25% of the channel pairs in each connectivity group. However, majority of the participants had good signal quality with 75% or more channel pairs retained; infants—test group 1: 24, test group 2: 23, control group: 26 (out of 26 participants) and adults—test group 1: 7, test group 2: 6, control group: 9 (out of 12 participants) as shown in figure S1 in the supplementary materials.

Raw light intensity data were converted to optical densities followed by motion artifact correction to correct anomalies in signals due to sudden displacements of optodes. The Temporal Derivative Distribution Repair (TDDR) method [33] was applied to correct sudden changes of amplitude in signals. Positive and negative derivatives were processed separately in the TDDR algorithm. After correcting step-like artifacts using TDDR, wavelet denoising was used to correct spike-like motion artifacts [34]. 'Sym8' function was used as the wavelet basis function and wavelet coefficients four standard deviations away from the mean were defined as outliers. Optical density data were then converted to concentration changes of HbO and HbR using the modified Beer–Lambert law [35].

2.3.2. Functional connectivity analysis

Three connectivity groups were defined to measure functional connectivity between different brain regions associated with language processing. Two test groups (inter-hemispheric homologous and intra-hemispheric) and a control group were defined using long distance channel pairs located in temporal and prefrontal regions of interest. The two regions of interest include Wernicke's (Brodmann's area, BA 22 and 40) and Broca's area (BA 44 and 45) residing in the STG and IFG, respectively. These areas are known to be involved in language processing and expected to develop substantially during infancy [13, 15]. Significant stimulus-locked activations were observed in these regions of interest in response to speech sounds in a separate study conducted on the same cohort of infants using the same optode montage. The block-averaged hemodynamic responses using a standard morphological analysis for an individual representative example in a topological plot (figure S2) and participant averaged responses of infants for each region of interest (figure S3) are presented in the supplementary materials.

Figure 2(d) shows the channel pair definition of the connectivity groups used in this study. The homologous group was defined by pairing channels in inter-hemispheric homologous regions to measure connectivity in homotopic regions across hemispheres while the intra-hemispheric group was defined by pairing channels in ipsilateral temporal and prefrontal regions to measure connectivity between two regions of interest in the same hemisphere. The control group comprised eight inter-hemispheric non-homologous connections that do not have known direct structural links. A frequency-domain connectivity measure, magnitude-squared coherence (simply referred to as coherence, here), was used to measure functional connectivity strength. By measuring the differences in connectivity strength for different ages, we aimed to assess the developmental effects of the strength of these functional connections during infancy.

The control group was used to normalize coherence data by taking the difference of coherence between the test and control groups of channel pairs. The benefit of using the difference of coherence instead of absolute coherence of the test group to characterize underlying functional connectivity is two-fold. First, this approach attempts to address the issue of spurious correlations in functional connectivity measures introduced by systemic physiological artifacts (particularly due to Mayer waves around 0.05–0.1 Hz). Consistent with the findings of previous studies [36–38], peaks of coherence were observed in our results in the frequency band associated with Mayer waves (0.05–0.15 Hz) as shown in figures 6(b)–(e), potentially indicating the presence of spurious correlations in connectivity measures. Second, some form of normalization of coherence data is required as coherence measured in some infant participants could be slightly higher than their age-matched peers due to inherent differences in brain structure and anatomy. By taking the difference of coherence between a test group and a control group, we aimed to remove the correlations due to common non-neuronal artifacts and normalize data to account for participant-specific differences likely not related to the development of language areas.

2.3.3. Regression analysis (age versus functional connectivity)

A subject-level analysis investigated within group differences of functional connectivity of infants at different ages. The analysis was performed by averaging coherence over channel pairs for each connectivity group and calculating the mean coherence in the frequency band of 0.01–0.1 Hz for each participant. Due to the slow nature of the hemodynamic responses, functional connectivity was characterized using the low frequency spontaneous fluctuations below 0.1 Hz consistent with the previous fNIRS connectivity studies [20, 26]. First, the effect of age on connectivity in the control channel pairs was tested to confirm the assumption that there is no significant developmental change in those connections. We then examined the developmental trajectory of inter- and intra-hemispheric language regions of the infant cohort by subtracting coherence, for each participant, and using the difference of coherence between the test and control groups of channel pairs. Based on the hypothesis that functional connectivity would increase with age in the first year of life, the developmental trajectories of language areas were established by fitting a linear regression model to functional connectivity data as a function of age for each test group. A linear model was selected because major structural connections of language areas are expected to mature in the first year with a peak occurring around one-year mark as reported in previous studies [12, 16]. The goodness of fit was characterized using a least-squares cost function.

2.3.4. Group-level validation analysis

To validate the findings objectively at group-level, coherence derived from experimental fNIRS data were used in this analysis similar to our previous study [37]. A baseline condition (inter-subject random) was defined using coherence of data from any random channel pairs where the data of each pair are from different subjects. For intra-subject conditions, a pair of fNIRS signals were randomly selected for each participant from each of the two test connectivity groups and the control group. Coherence was calculated on HbO and HbR data for channel pairs in each connectivity group for both infant and adult cohorts separately after rejecting channels that failed the signal quality assessment tests described in section 2.3.1. The above process was repeated 100 times and coherence was averaged over repeats to obtain group-level coherence estimates for all connectivity conditions. The mean coherence between 0.01 and 0.1 Hz was defined as a representative measure of functional connectivity strength. Statistical analysis was conducted using non-parametric version of ANOVA (Kruskal–Wallis test) to determine the significant differences (or lack thereof) between different connectivity conditions (including the baseline condition) for infants and adult cohorts separately.

3.1. Functional connectivity of language regions increases with age in the first year of life

Based on the strong evidence from structural connectivity studies on the maturation of long-range connections in language regions during infancy [12, 18], we hypothesized that functional connectivity strength of inter-hemispheric and intra-hemispheric language areas would also increase with age in the first year of life. To test this hypothesis, we chose inter-hemispheric homologous and intra-hemispheric groups as test groups and inter-hemispheric non-homologous group as a valid control group for within-subject analysis. Functional connectivity was characterized as the difference of mean coherence in 0.01–0.1 Hz frequency band between test group and the control group.

An increase of inter-hemispheric homologous connectivity relative to control connectivity was observed with increasing age for coherence derived from both HbO and HbR signals as illustrated in figure 3. A regression model was fitted to the difference of coherence data of infants with linear age terms. Statistical analysis revealed a significant coefficient of determination with age for both HbO (R2 = 0.216, p = 0.0169) and HbR (R2 = 0.206, p = 0.0198). The results suggest that inter-hemispheric homologous connectivity significantly increases with age in the first year of life. The Kruskal–Wallis test was performed on the difference of coherence averaged in the 0.01–0.1 Hz frequency band to test the differences between infant and adult cohorts. The statistical results revealed that adults have significantly higher inter-hemispheric homologous connectivity than infants for both HbO (F(1,36) = 10.87, p = 9.76 × 10−4) and HbR (F(1,36) = 8.35, p = 0.0039).

Figure 3. Coherence of (a) inter-hemispheric homologous group relative to the inter-hemispheric non-homologous control group derived from (b) HbO and (c) HbR signals of infant participants at different ages. A linear regression model was fitted to establish the developmental trajectory of inter-hemispheric homologous connectivity with age. Black solid line represents the best linear fit and dashed red lines denote 95% confidence bounds of the fitted line. Error bars in blue show one standard error of mean across channel pairs from error propagation analysis. Between-cohort comparison for infants and adults is also presented in the plots and error bars in gray represent 95% confidence interval of difference of coherence across participants.

Download figure:

Standard image High-resolution image

Similarly, to characterize the course of development for language areas between ipsilateral temporal and prefrontal regions, we used the difference of coherence between intra-hemispheric and control groups. Due to the maturation of fibers predominantly in the dorsal tract between temporal and prefrontal regions, an increase of intra-hemispheric connectivity with age was expected. As foreshadowed, an upward trend was observed for both chromophores: HbO and HbR, as shown in figure 4. Fitting a best linear fit to the difference of coherence data with respect to age revealed that there is a significant coefficient of determination for HbO (R2 = 0.237, p = 0.0117) but not for HbR (R2 = 0.111, p = 0.0956). A one-way ANOVA performed to test the differences between infant and adult cohorts did not report a significant difference for both HbO (F(1,36) = 2.74, p = 0.1066) and HbR (F(1,36) = 3.70, p = 0.0623). Although there are discrepancies between the results of HbO and HbR at subject-level for infant cohort, the general trend demonstrated an increase of intra-hemispheric connectivity with age.

Figure 4. Coherence of (a) intra-hemispheric group relative to the inter-hemispheric non-homologous control group derived from (b) HbO and (c) HbR signals of infant participants at different ages. A linear regression model was fitted to establish the developmental trajectory of intra-hemispheric connectivity with age. Black solid line represents the best linear fit and dashed red lines denote 95% confidence bounds of the fitted line. Between-cohort comparison for infants and adults is also presented in the plots.

Download figure:

Standard image High-resolution image 3.2. Inter-hemispheric non-homologous channel pairings are a valid control group to normalize coherence data

Channel pairs for the inter-hemispheric non-homologous control group were defined based on the assumption that direct long distance structural connections do not exist between contralateral temporal and prefrontal regions. We also assumed that connectivity due to indirect connections through multiple hops would not be fully reflected in the typical frequency band associated with resting-state (0.01–0.1 Hz) due to ultra-long distances between those brain regions [39], hence expected a lower coherence for control group compared to test groups. Therefore, we expected that coherence of this control group of channel pairs would not significantly differ between infants and adults and remain stable in infants in the first year of life. To validate this assumption and confirm that this control group can be used to remove inter-subject differences of coherence not related to age, we conducted a subject-level linear regression analysis to characterize the changes of coherence in the control channel pairs during infancy. A group-level comparison was performed to contrast differences between infant and adult cohorts. The results of these analyses are shown in figure 5.

Figure 5. Coherence of inter-hemispheric non-homologous group derived from (a) HbO and (b) HbR signals of infant participants at different ages. A linear regression model was fitted to assess the relationship of inter-hemispheric non-homologous coherence with age. Black solid line represents the best linear fit and dashed red lines denote 95% confidence bounds of the fitted line. Between-cohort comparison for infants and adults is also presented in the plots.

Download figure:

Standard image High-resolution image

The gradient of the best fitted regression line was almost zero for age versus coherence when using both HbO and HbR signals (−1.06 × 10−4 and −1.90 × 10−5, respectively). The results showed no significant coefficient of determination with age for both HbO (R2 = 9.76 × 10−3, p = 0.6311) and HbR (R2 = 2.4 × 10−4, p = 0.9401). A one-way ANOVA revealed that there was no significant difference in mean coherence (in 0.01–0.1 Hz band) between infant and adult cohorts for both HbO (F(1,36) = 1.37, p = 0.2493) and HbR (F(1,36) = 0.05, p = 0.8244). Taken together, these observations suggest that coherence of the inter-hemispheric non-homologous group does not significantly change with age. Therefore, this group of channels can be considered as an acceptable within-subject control group in our functional connectivity analyses to establish the typical developmental trajectories of inter-hemispheric homologous and intra-hemispheric language areas during infancy.

3.3. Control analysis using a frequency band associated with non-neural activity

To further validate that our findings of functional connectivity with age for test groups were not due to noise, we conducted a control regression analysis by using the coherence of a frequency band that is not known to be associated with neural activity or major physiological event. A previous fMRI study that investigated the contribution of frequencies to resting-state functional connectivity in cortical networks found that lower frequencies (<0.1 Hz) contributed to more than 90% of the significant correlations [40]. Based on these findings, we chose the mean coherence of 1.1–1.3 Hz frequency band for this control analysis.

For inter-hemispheric non-homologous control group, significant coefficients of determination with age were not found for both HbO (R2 = 2.10 × 10−6, p = 0.9944) and HbR (R2 = 0.0670, p = 0.2018) as illustrated in figure S4 in the supplementary materials. This result validates our assumption that the inter-hemispheric non-homologous group does not fully reflect connectivity due to neural activity, thus can be considered as a valid control group to remove participant-specific differences not related to the development. A similar analysis was conducted to investigate the relationship of functional connectivity with age using the coherence of test groups relative to coherence of the control group. No significant coefficient of determination was observed for both inter-hemispheric homologous (HbO: R2 = 0.0013, p = 0.8627; HbR: R2 = 2.81 × 10−6, p = 0.9935) and intra-hemispheric (HbO: R2 = 0.0161, p = 0.5370; HbR: R2 = 0.0743, p = 0.1780) test groups as presented in figures S5 and S6, respectively in the supplementary materials. The results in sections 3.1 and 3.3 collectively suggest that significant positive correlations for functional connectivity with age during infancy can be observed only for spontaneous low frequency fluctuations reflecting neural activity (<0.1 Hz) but not for a higher frequency band associated with non-neural activity.

3.4. Comparison of group-level differences of functional connectivity

A large body of previous work on functional connectivity has overlooked the need to compare test conditions of connectivity with acceptable experimental control conditions. Lack of such comparisons would potentially lead to misinterpreting the attributes of functional networks [37, 41]. To address this issue, we compared the coherence of several test conditions against a baseline condition derived from experimental data to objectively validate our findings. The test and control conditions were intra-subject and defined based on the connectivity groups explained in section 2.3.2, namely: inter-hemispheric homologous, intra-hemispheric and inter-hemispheric non-homologous. Inter-subject random was considered as the baseline condition detailed in section 2.3.4. The functional connectivity patterns between infant and adult cohorts were also compared to establish general trends that could be related to the maturation of language areas.

Connectivity due to neural activity is not expected from a condition in which channels were randomly taken from two participants who participated in the experiment on different days. Therefore, inter-subject random condition was expected to have the lowest calculated coherence out of the four conditions in both infant and adult cohorts as shown in figure 6. Functional connectivity of intra-subject test groups had similar patterns in both infant and adult cohorts. The highest coherence was observed for the inter-hemispheric homologous condition in the frequency band associated with resting-state functional connectivity (0.01–0.1 Hz). While inter-hemispheric non-homologous control condition had the lowest coherence in intra-subject conditions for the same frequency band, coherence of intra-hemispheric condition was lower than inter-hemispheric homologous condition. These patterns of connectivity were consistent across coherence derived from both HbO and HbR signals as illustrated in figures 6(b)–(e).

Figure 6. Participant-averaged magnitude-squared coherence of infant and adult cohorts for different connectivity conditions. (a) Channel pair definition of inter-hemispheric homologous, intra-hemispheric and inter-hemispheric non-homologous groups were used for intra-subject connectivity conditions. Comparison of coherence derived from HbO signals of (b) infants and (c) adults for different test and control conditions (test: inter-hemispheric homologous, intra-hemispheric, control: inter-hemispheric non-homologous; baseline: inter-subject random). Comparison of HbR coherence of (d) infants and (e) adults for the same test and control conditions. Dashed vertical gray lines indicate the typical upper bound of resting-state functional connectivity frequency band. Error bars represent standard error of mean of coherence across repeats.

Download figure:

Standard image High-resolution image

The Kruskal–Wallis test was performed considering the mean HbO coherence of 0.01–0.1 Hz frequency band as the response variable to statistically test the differences between connectivity conditions for infant and adult cohorts separately. The statistical results revealed a significant difference in connectivity conditions for infants (F(3,375) = 128.20, p = 1.32 × 10−27) and adults (F(3,253) = 65.73, p = 3.49 × 10−14). Post-hoc test with Bonferroni correction for multiple comparisons confirmed that all intra-subject conditions were significantly different from the corresponding inter-subject random baseline condition for both infants and adults (adjusted p < 0.001). The same statistical analysis was repeated for HbR coherence, and a significant difference was found for infants (F(3,380) = 114.33, p = 1.29 × 10−24) and adults (F(3,248) = 71.02, p = 2.58 × 10−15). Follow-up post-hoc tests showed that all intra-subject conditions were significantly different from the baseline condition for both infant and adult cohorts (adjusted p < 0.001 for all pairwise comparisons).

An increase of mean coherence was observed in adults compared to infants for intra-subject inter-hemispheric homologous and intra-hemispheric conditions particularly in 0.01–0.05 Hz frequency band for both chromophores. The coherence of inter-hemispheric non-homologous control condition in the same frequency band did not differ substantially between infants and adults. The observed changes of coherence (or lack thereof) are consistent with the findings of similar fMRI studies [15, 39] and could reflect the developmental effects of language areas across infants and adults.