Spatial development of brain networks during the first six postnatal months

Abstract

The early postnatal period is crucial for brain development and understanding neurodevelopmental disorders. This study examines spatial brain network development in early infancy, a less-explored area. Using independent component analysis on longitudinal resting-state functional magnetic resonance imaging data from 74 neurotypical infants, we examined how the spatial organization of brain networks evolves from birth to 6 months. Our findings show significant age-related changes in spatial characteristics. Network-averaged spatial similarity, reflecting alignment between individual and group-level network maps, increased with age. Concurrently, network engagement range, representing voxel intensity fluctuation within networks, decreased, suggesting a consolidation process where voxel contributions became more uniform. Network strength, calculated as the average of all the significant voxel intensities in the network, indicating the degree of involvement in the specific functional network, increased across age in networks such as the frontal-medial prefrontal cortex and visual networks. We found that network size and network center of mass (illustrating spatial distribution alterations of brain networks) increased in the temporal network. These findings fill a gap in infant neuroimaging by spatially characterizing early functional network development. Quantifying changes in topology, size, and similarity offers a framework for understanding early brain maturation and identifying atypical trajectories.

Longitudinal rs-fMRI (n = 74) charts rapid spatial maturation of infant brain networks over the first 6 months: greater spatial similarity, tighter voxel engagement, and network-specific gains in strength, size, and centroids.

Introduction

The first few months after birth are considered a crucial timeframe in which the fundamental aspects of the brain’s functional and structural framework are established^1,2. Consequently, there is growing interest in understanding how deviations from typical neurodevelopment during this period may play a pivotal role in the emergence of developmental disorders and psychopathology³. Resting-state (rs) functional connectivity (FC)—the measurement of correlated spontaneous low-frequency fluctuations in the blood-oxygen-level dependent functional magnetic resonance imaging (fMRI) signal across functionally related brain regions during rest^{4, 5–6}—provides a window into functional brain organization in early infancy^7,8. It is well-established that most adult functional brain networks are observable from birth (perhaps even in the fetal period) and undergo asynchronous developmental changes over the first postnatal months and years^{9, 10, 11, 12, 13, 14, 15–16}. More recently, studies have begun to identify deviations in these sensitive periods of functional network connectivity (FNC) development may be linked to developmental disabilities, further highlighting the value of charting trajectories of functional brain network maturation^{17, 18, 19, 20–21}.

While prior studies of functional network connectivity (FNC) in infancy have illuminated developmental changes in temporal coupling, spatial organization has primarily been described through group-level visualizations¹⁴. However, longitudinal, within-subject changes in the spatial organization of brain networks remain largely unexamined. Brain networks are known to be spatially dynamic (That is, the spatial configuration of functional regions change over time)²². Spatially dynamic analyses have revealed critical information about brain functional architecture and its association with behavior^23,24 and alterations in conditions such as schizophrenia²⁵ providing new insights into functional brain networks that may be hidden by existing approaches that focus on variations of temporal coupling among spatially fixed functional sources^24,26. In particular, Garcia et al. advanced this direction by examining spatial organization of brain communities using metrics such as community spatial extent, diameter, and radius thus emphasizing the value of spatial characterization alongside traditional community detection²⁷. This highlights the importance of directly characterizing spatial organization as a distinct and informative dimension of network development.

Here, we provide the first longitudinal, individual-level quantification of developmental changes in the spatial organization of functional brain networks during the first six postnatal months, a period characterized by extremely rapid brain growth and development^28,29. Infants completed up to three scans during their first six postnatal months, providing dense longitudinal coverage over this highly dynamic period of development. To capture developmental changes in the spatial organization of brain networks, it is critical to use methods that provide both accurate group-level representations and individualized maps of functional architecture³⁰. To this end, we applied group information-guided ICA (GIG-ICA)³¹, which improves upon conventional ICA approaches by enhancing spatial consistency across subjects while preserving meaningful individual differences. This balance makes GIG-ICA especially well suited for developmental studies, where both common network structures and individual variability are of interest³¹.

As part of our investigation into early brain development, we examined whether functional networks could be grouped based on shared developmental trajectories in their spatial organization. To do so, we applied a spatial modularity analysis to the network maps, identifying clusters of networks that exhibited similar age-related changes in spatial configuration. Our modularity analysis was tailored to the spatial domain—focusing on how the physical configuration of network regions shifts over time³². This approach provides a complementary perspective by revealing coordinated patterns of spatial reorganization across networks during infancy (see Spatial measurements for methodological details).

Secondly, we mapped developmental trajectories across several spatial characteristics to understand how functional networks change during early infancy. Network-averaged spatial similarity provides insights into the consistency of network configurations across individuals, allowing researchers to assess how well a participant’s brain network matches commonly shared spatial patterns. Network engagement range refers to how consistently or variably the voxels contribute to functional networks. Network strength refers to the average intensity of all voxels within a network, indicating the overall level of involvement of those voxels in functional networks. Network size provides a measure of brain growth or pruning processes, key to understanding network expansion or contraction over development. Finally, network center of mass captures the geometric configuration of each network, enabling detection of subtle developmental shifts in network location or shape. Together, these measures provide a comprehensive characterization of the spatial characteristics of functional brain networks across early infancy. While inter-individual variability is an important aspect^{23,33, 34, 35, 36, 37, 38, 39–40}, our primary goal is to understand how these spatial characteristics change with age—a critical but understudied dimension of early brain development. By focusing on spatial organization, we aim to uncover fundamental patterns in how the spatial organization of functional networks evolves during this dynamic postnatal period. Importantly, accounting for spatial development is essential to accurately interpret functional data, as neglecting these changes may obscure or misrepresent the true nature of evolving brain network architecture^23,41.

In line with the rapid neurodevelopmental changes occurring during this period³, we anticipated that spatial maps would undergo substantial reorganization during infants’ first postnatal months. These changes are likely influenced not only by cortical expansion but also by synaptic refinement and myelination processes⁴², which play a crucial role in shaping functional network architecture during early development. By mapping longitudinal change in the spatial organization of functional networks, this study offers insight into a critical aspect of functional brain network maturation that was previously overlooked in studies of early infancy.

Materials and methods

Participants

Participants were N = 74 neurotypical infants (43 males and 31 females) enrolled in prospective longitudinal studies at the Marcus Autism Center in Atlanta, GA, USA (see Table 1 for participant demographics). Infants had a mean gestational age at birth of 39.2 weeks (SD = 1.34) and were considered neurotypical on the basis of having no family history of autism in up to third-degree relatives, no developmental delays in first-degree relatives, no pre- or perinatal complications, no history of seizures, no known medical conditions or genetic disorders, and no hearing loss or visual impairment. Infants with contraindications for MRI were excluded. Participants’ parents provided informed written consent, and the Emory University Institutional Review Board approved the research protocol for this study. All ethical regulations relevant to human research participants were followed.

Table 1. Demographics of participant sample

	(N = 74)
Infant sex	43 M, 31 F
Gestational age at birth, mean (SD) (N = 74)	39.2 weeks (1.34)
Race (N = 66)
Black	7.6%
White	86.3%
More than one race	6.1%
Maternal education (N = 64)
High School	1.6%
Trade/Vocational Training	1.6%
College Courses	6.2%
Associate’s Degree	1.6%
College Degree	34.3%
Graduate Degree	54.7%
Household income (N = 63)
<$40,000	4.8%
$40,000–$80,000	19.0%
$80,001–$100,000	22.2%
$100,001–$150,000	28.6%
>$150,001	25.4%

For information categories with a participant number less than the total sample, the N is specified next to the category title.

Scans were scheduled for each infant at up to 3 pseudorandom timepoints between birth and 6 months (Fig. S8). This approach yielded dense coverage over the 0- to 6-month period (137 scans, each separated by a mean of only 1.4 days (SD = 1.8); See Fig. S8 in Section 1.8 of Supplementary Note for a distribution of participant age at each scan), with 28, 29, and 17 of infants contributing one, two, or three longitudinal scans, respectively. Infant age at each scan was corrected for length of gestation (Corrected age), defined as the age in days minus 7×(40-gestational age in weeks)⁴³.

Data collection

Infant scans were acquired at Emory University’s Center for Systems Imaging Core on a 3 T Siemens Tim Trio (N = 65) or a 3 T Siemens Prisma (N = 72) scanner, using a 32-channel head coil. Infants were swaddled, rocked, and/or fed to encourage natural sleep. Once asleep, infants were positioned on a pediatric scanner bed. To minimize scanner noise to levels below 80 dBA, two measures were implemented: (1) the use of sound attenuating pediatric headphones, featuring MR-safe optical microphones for real-time monitoring of in-ear sound levels, and (2) the integration of a custom-built acoustic hood within the MRI bore⁴⁴.

To mitigate the abrupt onset of scanner noise, white noise—gradually increased in volume—was played through the headphones prior to the first sequence. An MRI-compatible camera (MRC Systems) was affixed to the head coil, facilitating continuous monitoring of the infant during the scan. An experimenter was present in the scanner room throughout the scan and the procedure was stopped if the infant awoke or if an increase in sound level was observed.

Functional MRI data from the Tim Trio scanner were acquired using a multiband Echo planar imaging (EPI) sequence. The acquisition parameters were as follows: repetition time (TR) of 720 ms, echo time (TE) of 33 ms, flip angle of 53°, a multiband factor of 6, a field-of-view (FOV) of 208 × 208 mm, and an image matrix of 84 × 84. The spatial resolution was 2.5 mm isotropic, and 48 axial slices were acquired to cover the whole brain.

Functional MRI data from the Prisma scanner were acquired using a multiband EPI sequence. The key acquisition parameters included a TR of 800 ms, a TE of 37 ms, and a flip angle of 52°. The multiband factor was set to 8, with a FOV of 208 × 208 mm and an image matrix of 104 × 104. The spatial resolution achieved was 2 mm isotropic, encompassing 72 axial slices to cover the entire brain.

Preprocessing

Data preprocessing was conducted as follows: Initially, the first 16 volumes were removed due to significant signal fluctuations required for magnetization equilibrium. This ensured that the remaining data was stable and free from initial transient effects that could interfere with accurate analysis. Head motion correction was performed using the mcflirt function in FSL (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FSL), aligning all volumes to the first one. Importantly, no frame censoring or scrubbing was applied; instead, we accounted for residual motion artifacts by including mean framewise displacement (FD) as a covariate in all statistical models. To correct for distortion in the multi-band rs-fMRI data, single-band data acquired with phase encoding in both the anterior-posterior and posterior-anterior directions were used to estimate the susceptibility-induced off-resonance field. Following distortion correction, slice timing correction was applied.

A two-step normalization process was employed to align the infant datasets into a common Montreal Neurological Institute (MNI) space. The first step involved normalizing the infant fMRI data to the 3-month UNC/UMN baby connectome project (BCP) T1 template⁴⁵. In the second step, the output from the first normalization was further aligned to the common MNI space using the EPI template⁴⁶. To ensure that our developmental findings were not biased by the use of a single (3-month) anatomical template, we performed a validation using age-specific templates. As an illustrative example, we evaluated the developmental trajectory of network-averaged spatial similarity (NASS), a key spatial metric in our study. This validation analysis (Fig. S4 in Section 1.4 of the Supplementary Note) confirmed the robustness of the NASS developmental pattern, revealing nearly identical results when using the age-specific normalization approach. Finally, a 6 mm Gaussian kernel was applied to smooth the spatially normalized fMRI images.

Data analysis

Quality control (QC) was conducted on the preprocessed data by comparing the individual masks with the group mask (see Section 1.1 in the Supplementary Note for detailed QC procedures). In lieu of excluding scans with significant head motion, we chose to retain all scans and apply motion correction (see Section 2.3 for preprocessing details and Section 1.3 of the Supplementary Note for supporting analyses demonstrating the validity and representativeness of including high-motion scans; see also Supplementary Note Figs. S1–S3). Head motion was controlled using framewise displacement (FD) as a covariate in our statistical models (see Section 1.2 of Supplementary Note for the FD formula; see also Section 2.7 for details on statistical analysis). This approach provides several advantages: (1) it preserves statistical power by retaining more data, avoiding the loss of potentially informative scans; (2) it maintains the integrity of the original data distribution, preventing artificial alterations; and (3) it reduces biases associated with data exclusion, which can lead to unrepresentative samples and skew results by disproportionately omitting data from higher-motion participants^47,48 (see Supplementary Note Fig. S9 for the mean FD distribution).

We conducted group ICA on the scans acquired from all infants⁴⁹, using the group ICA of fMRI toolbox (GIFT; http://trendscenter.org/software/gift)^{50, 51–52}. Initially, participant-specific principal components analysis (PCA) was utilized to standardize the data and ensure that each participant’s contribution is comparable within the shared subspace. This participant-specific PCA process also offers advantages in noise reduction and computational efficiency^53,54. Thirty principal components (PCs) capturing the maximum variance for each participant were retained for subsequent analysis. To implement this, all principal components derived from individual participants were combined along the time dimension, and then group-level PCA was performed on this concatenated dataset. Participant-specific PCA emphasizes individual differences at the participant level, whereas group-level PCA emphasizes commonalities among participants⁵⁴.

We utilized the top 20 group-level PCs explaining the maximum variance as input for group ICA. We applied a group ICA model order of 20, as suggested by previous studies^55,56, aimed to capture large-scale brain networks. To ensure the reliability of ICA results, we employed the Infomax optimization algorithm⁵⁷ in ICASSO to decompose the data 100 times with random initialization and bootstrapping of the fMRI time series, allowing for robust estimation of stable spatial components across runs. The independent components most closely aligned with the centroids of stable clusters were deemed the optimal ones and chosen for subsequent analysis. We assessed the reliability and quality of the large-scale brain networks using the ICASSO quality index (IQ), quantifying component stability across runs⁵⁸. In the fMRI large-scale brain networks estimation context, ICASSO IQ is commonly employed to distinguish reliable components suitable for further analysis from unstable ones²⁵. A large-scale brain network was identified if it met specific criteria, including an ICASSO IQ value exceeding 0.80, high spatial overlap with gray matter, peak weight within gray matter, and low spatial similarity to motion, ventricular, and other artifact components. Subsequently, we employed the group information-guided ICA (GIG-ICA) technique to estimate participant-specific large-scale brain networks³¹. GIG-ICA leverages group-level independent components to guide the extraction of subject-specific spatial maps using a multi-objective optimization framework that simultaneously enhances spatial correspondence across participants and preserves individual variability. This approach yields more accurate and reliable networks without artificially reducing inter-individual differences, making it particularly suited for developmental studies³¹. After delineating the large-scale brain networks, we categorized each component by discerning their activation patterns. We ascribed labels to brain networks based on their functional roles within the brain.

Inter-network modularity analysis

To focus on the regions within each brain network where developmental changes are most likely to occur, we explored the relationship between each network and corrected age through voxel-level correlation analyses. This approach involved assessing the correlation between voxel-specific activity in participant-specific spatial maps and corrected age across all scans. The results are presented as spatial maps, highlighting regions within each network where voxel activity shows a significant correlation with age.

After identifying voxels significantly associated with age within each network, we delved into exploring the interrelationship among these voxels across different networks. We first computed the average intensity of all voxels significantly associated with age within each network—a measure we define as the age-correlated voxel mean (ACVM). For each network, this resulted in a single ACVM value per participant, reflecting age-related changes in voxel intensities. We then assessed the similarity of these developmental patterns across networks by correlating the ACVM vectors (Dimension: N participants × 1) between each pair of networks. This yielded an M × M correlation matrix (Where M is the total number of labeled brain networks). A positive correlation between two networks indicates that their ACVMs co-vary with age in the same direction, suggesting coordinated developmental trends. In contrast, a negative correlation suggests opposing age-related patterns, where increased functional engagement in one network coincides with decreased engagement in another. We next employed the modularity metric from the Brain Connectivity Toolbox to segment the network into distinct groups (https://sites.google.com/site/bctnet/)⁵⁹. It is important to note that for this specific analysis, we focused exclusively on the voxels that were significantly associated with age. However, for all other analyses, we conducted our analysis across all voxels or all significant voxels after thresholding within the identified network.

Spatial measurements

We introduced several metrics, including network-averaged spatial similarity, network engagement range, network strength, network size, and network center of mass, to examine developmental changes in the spatial organization of brain networks. It is important to highlight that all of these metrics are derived from voxel intensities in the spatial map, which reflect the degree of engagement of each voxel in contributing to the network.

Network-averaged spatial similarity (NASS)

We introduced the “network-averaged spatial similarity” (NASS) metric to quantify how closely each participant-specific spatial map aligns with the group-level map (Fig. 1A). The group-level map represents the commonalities across all participants, reflecting the shared spatial organization of brain networks that are consistent across individuals. To compute NASS, we calculate the correlation between each participant’s individual spatial map and the group-level map, resulting in a NASS value for each network in every scan. A higher NASS indicates that an individual’s network closely aligns with the shared spatial pattern observed across the group, suggesting that their network structure is more homogenized. Conversely, a lower NASS suggests a greater deviation from the group-level pattern, indicating more individual-specific variation in the network’s spatial map. It is important to note that while increasing NASS with age reflects a convergence toward group-level spatial patterns, this does not imply a reduction in all forms of interindividual variability. For example, temporal dynamics, functional connectivity strength, and individual developmental trajectories may still vary across infants. Therefore, NASS captures a specific, spatial dimension of variability that becomes more consistent with age—likely due to maturational processes leading to stabilized network architecture. Moreover, it’s important to note that smoothing can influence network-averaged spatial similarity. While it increases the signal-to-noise ratio^60,61, enhancing the detection of consistent patterns across participants, it can also affect inter-participant variability by making individual differences less apparent⁶². To mitigate this effect, we carefully selected our smoothing kernel size based on the optimal range identified in prior studies for group-level ICA⁶³, ensuring that the benefits of smoothing are realized while minimizing its impact on inter-participant variability.

Fig. 1 Conceptual overview of spatial network metrics used to quantify individual differences in brain network organization. [Images not available. See PDF.]

A Network-Averaged Spatial Similarity (NASS) measures the similarity between an individual’s spatial map and the group-level map, reflecting how closely each participant’s network aligns with the canonical group pattern. Higher NASS indicates more spatial conformity across individuals. B Network Engagement Range (NER) quantifies the variability of voxel intensity within a network. It is defined as the difference between the highest and lowest voxel intensities, capturing spatial heterogeneity of network engagement. C Network Strength is the average intensity of all voxels exceeding a statistical threshold (Z > 1.96) within a network, reflecting the overall strength of voxel-wise engagement in the network. D Network Size captures the spatial extent of a network by counting the number of suprathreshold voxels. Network size can vary across individuals due to network expansion or shrinkage. Each metric is computed per participant and summarized across participants for group-level analysis.

Network engagement range (NER)

We assessed the intensity range of voxels within each network by introducing a novel metric termed “network engagement range”. This metric is used to quantify the variability of voxel contribution within the network (Fig. 1B). A wider intensity range may indicate greater variability in voxel contribution, suggesting that some areas within the network could highly engaged while others could contribute minimally, thus reflecting a broader spatial distribution of engagement. Conversely, a narrower intensity range might indicate more uniform voxel contributions, indicating that voxel contribution is distributed more consistently across the network. Calculating this range involves identifying the highest and lowest voxel intensities, and the difference between these values constitutes the network engagement range, representing the variance between the maximum positive and minimum negative values of voxels within each network.

Network strength

Network strength is defined as the average intensity of all voxels within a network, reflecting the level of engagement to the functional networks (Representing how strongly voxels are functionally engaged in the network). Elevated network strength values indicate increased voxel involvement within the network. To compute network strength, we first applied a mask to identify voxels exceeding a Z-score threshold of 1.96 (p = 0.05), selecting voxels that significantly contribute to the network. Afterward, we calculated the network strength by averaging the intensities of these selected voxels, resulting in a singular value representing the network strength for each network in every scan (Fig. 1C).

Network size

We quantified the spatial extent of a network using a global metric called network size to assess the shrinkage/expansion of brain networks over time. Changes in the network size within a particular network indicate an alteration in the number of voxels contributing to that network, thereby reflecting fluctuations in its size. In computing network size, we again began by applying a mask to identify voxels exceeding the threshold of Z = 1.96 (p = 0.05) within each network. Then, we counted the number of voxels meeting this criterion to quantify the size of the network (Fig. 1D).

Network center of mass (NCM)

We evaluated both the configuration and extent of a network comprehensively by introducing a novel metric termed the “network center of mass”. This measurement incorporates the concept of the center of mass (COM)⁶⁴, a theoretical point where the total mass of an object is considered concentrated, but in fMRI is defined as the center of a cerebral activation cluster consisting of a defined number of voxels⁶⁴. By leveraging the distance from the COM, the NCM provides insights into the spatial distribution of the network. The computation involves applying a mask to filter voxels surpassing Z = 1.96 (p = 0.05) within each network. Subsequently, we determine the COM for each network using the following formula:

Let x, y, and z represent the indices of the three dimensions (rows, columns, slices), and I(x, y, z) be the value of the voxel at (x, y, z).

1

$$COM\left(x,y,z\right)=\frac{{\sum }_{x=1}^{rows}{\sum }_{y=1}^{cols}{\sum }_{z=1}^{slices}\left(x,y,z\right)\mathrm{\times }I\left(x,y,z\right)}{{\sum }_{x=1}^{rows}{\sum }_{y=1}^{cols}{\sum }_{z=1}^{slices}I\left(x,y,z\right)}$$

Once the COM has been obtained, we calculated the weighted average distance from COM. This involves determining the weighted distance of each voxel from the COM and subsequently calculating the average distance among these weighted distances. The formula for this process is as follows:

2

$$NCM\left(x,y,z\right)=\frac{\sum \sum \sum \sqrt{{\left(x-CX\left(x\right)\right)}^2+{\left(y-CY\left(y\right)\right)}^2+{\left(z-CZ\left(z\right)\right)}^2}\mathrm{\times }I\left(x,y,z\right)}{{\sum }_{x=1}^{rows}{\sum }_{y=1}^{cols}{\sum }_{z=1}^{slices}I\left(x,y,z\right)}$$

In this formula, CX(x), CY(y), and CZ(z) are the coordinates of the center of mass, I(x, y, z) is the value of the voxel at (x, y, z), and $\mathrm{NCM}\left(x,y,z\right)$ is the weighted average distance of all voxels within a network to COM for each network.

Statistics and reproducibility

For the statistical analysis, we employed a generalized additive model (GAM)⁶⁵ to explore the relationships between each metric and the variables of interest. In this model, corrected age, sex, scanner type, and head motion (Mean FD) were included as fixed effects, while individual participants were included as random effects to account for repeated measures ( $\mathrm{Spatial}\mathrm{Metric}~\mathrm{Corrected}\mathrm{Age}+\mathrm{Sex}+\mathrm{Scanner}+\mathrm{Head}\mathrm{Motion}+\left(1\mid \mathrm{Participants}\right)$ ). Specifically, we included only a random intercept for each participant, allowing for individual baseline differences but assuming a common slope (developmental trajectory) across participants (n = 74 biologically independent participants). This approach allows us to account for potential linear and non-linear relationships and adjust for confounding factors, providing more robust and flexible inferences. To explore the potential influence of sociodemographic variables (E.g., maternal education, household income, and race), we conducted a supplementary GAM analysis including these as additional covariates. The results remained consistent with our primary findings and are presented in the Supplementary Note (Fig. S5 in Section 1.5 of Supplementary Note). To manage the multiple comparisons issue, p values were adjusted using the false discovery rate (FDR) correction.

To ensure the robustness of our developmental analyses, we conducted outlier detection for all five spatial network metrics—network-averaged spatial similarity, network engagement range, network strength, network size, and network center of mass—using MATLAB’s isoutlier function. This procedure identified statistical outliers based on each metric’s distribution across participants. We then re-ran all models after excluding these outliers and found that the overall developmental trajectories and statistical inferences remained qualitatively and statistically consistent. Because results were stable and to preserve the longitudinal integrity of the dataset, we elected to retain all data points, including those marked as outliers. For illustrative purposes, this validation process is detailed in Supplementary Note Section 1.7 (Fig. S7) using the network strength metric as a representative example.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Results

Our results are divided into two main sections: Initially, we explore the functional brain networks identified in infants, conducting voxel-level analysis and modularity analysis to examine their association with age. Subsequently, we focus on mapping trajectories of the spatial characteristics of these networks.

Summary of the identified brain networks and their correlations with age

We applied ICA to decompose rs-fMRI data from 74 participants into 20 components. Among these, 13 components were identified as distinct functional brain networks in infants, as shown in Fig. 2. The remaining 7 components were excluded as they primarily reflected non-neuronal noise, including head motion, physiological artifacts, and scanner-related signals.

Fig. 2 Large-Scale Brain Networks. [Images not available. See PDF.]

Sagittal, coronal, and axial views illustrate the z-scored voxel intensity distributions of spatial maps for 13 functional brain networks in infants. The maps are thresholded at Z > 1.96, corresponding to p < 0.05 (one-sided, n = 74 biologically independent participants). Color bars represent Z-scores of voxel intensity values. Frontal networks include: Frontal-mPFC (Frontal medial prefrontal cortex), Frontal-dlPFC (Frontal dorsolateral prefrontal cortex), and Frontal-vlPFC (Frontal ventrolateral prefrontal cortex).

These networks include the primary and secondary visual networks, subcortical network, cerebellum network, primary and secondary motor networks, attention network, default mode network, temporal network, auditory network, frontal-medial prefrontal cortex (mPFC) network, frontal-dorsolateral prefrontal cortex (dlPFC) and frontal-ventrolateral prefrontal cortex (vlPFC) networks. Figure 3 shows the spatial distribution of voxel-wise correlations between voxel intensity and corrected age across various brain networks. Most primary networks, including the primary visual, secondary motor, auditory, cerebellum, and subcortical networks, predominantly exhibit positive correlations, indicating that voxel contribution in these regions increases with age during the first 6 months of life. In contrast, higher-order association networks, such as the secondary visual, attention, default mode, temporal, frontal-mPFC, frontal-dlPFC, and frontal-vlPFC, display a more heterogeneous pattern, with intermingled regions of positive and negative correlations. All results are adjusted for multiple comparisons using FDR correction.

Fig. 3 Correlation between voxel intensities and corrected age. [Images not available. See PDF.]

Sagittal, coronal, and axial views of the correlation between the voxel-level spatial maps of 13 functional infant brain networks and corrected age for all scans (n = 74 biologically independent participants). Analyses were corrected for multiple comparisons using a 5% false discovery rate (FDR).

Another aspect of our analysis focused on examining the age-correlated voxel mean (ACVM) across networks (see 2.5 in the “Methods” for more details). Our modularity analysis of these derived using the Brain Connectivity Toolbox, identified two primary modules within the network set. These modules were defined based on the similarity of age-related spatial change trajectories across networks, using only voxels significantly associated with age. The first module includes the attention, frontal-vlPFC, frontal-mPFC, frontal-dlPFC, default mode, primary motor, and secondary visual networks, while the second module comprises the primary visual, secondary motor, cerebellum, subcortical, auditory, and temporal networks. Figure 4 depicts the correlation matrix among all networks, illustrating the within- and between-module relationships. Positive correlation values (Shown in red hues) indicate that two networks exhibit similar age-related developmental patterns, suggesting coordinated maturation. Conversely, negative correlations (in blue hues) may reflect inverse but still coordinated changes—where one network’s age-related developmental patterns increase with age while another decrease—potentially pointing to complementary or competitive developmental processes. The magnitude of the correlations ranges from approximately −0.6 to 0.6, with stronger colors indicating stronger relationships.

Fig. 4 Correlation between modules. [Images not available. See PDF.]

Correlation between the networks belonging to the first module, comprising attention, frontal-vlPFC, frontal-mPFC, frontal-dlPFC, default mode, primary motor, and secondary visual networks, and the networks belonging to the second module, including primary visual, secondary motor, cerebellum, subcortical, auditory, and temporal networks. We display only those correlation values that remained statistically significant after FDR correction for all p values.

Spatial developmental metrics of brain networks

Age-related changes in spatial characteristics of all 13 networks are shown in Figs. 5–9 (Primary visual, secondary visual, cerebellum, primary motor, secondary motor, attention, subcortical, default mode, auditory, temporal, frontal-mPFC, frontal-dlPFC, and frontal-vlPFC networks).

Fig. 5 Age-related changes in network-averaged spatial similarity (NASS) across functional brain networks. [Images not available. See PDF.]

Scatter plots show the relationship between corrected age (in days) and NASS for 13 functional brain networks. Each red dot represents one scan from an individual infant, and the blue line denotes the fitted generalized additive model with 95% confidence intervals (shaded area). Analyses were adjusted for multiple comparisons by applying FDR correction (n = 74 biologically independent participants).

Fig. 6 Age-related changes in network engagement range (NER) across functional brain networks. [Images not available. See PDF.]

Scatter plots illustrate the relationship between corrected age (in days) and NER across 13 functional brain networks. Each red dot represents one scan from an individual infant, and the blue line depicts the fitted generalized additive model with 95% confidence intervals (shaded area). Analyses were corrected for multiple comparisons using FDR correction (n = 74 biologically independent participants).

Fig. 7 Developmental trajectories of network strength across functional brain networks. [Images not available. See PDF.]

This figure displays the association between corrected age (in days) and network strength for 13 functional brain networks. Each red dot denotes an individual scan, and the blue line represents the fitted generalized additive model with 95% confidence intervals (shaded area). Significant increases in network strength with age were observed in several networks, including secondary visual, secondary motor, frontal-mPFC, and frontal-vlPFC. All statistical results were corrected for multiple comparisons using the FDR method (n = 74 biologically independent participants).

Fig. 8 Longitudinal changes in network size which is reported in voxel count, across early infancy. [Images not available. See PDF.]

The scatter plots display the relationship between corrected age (in days) and network size for 13 functional brain networks. Each red dot corresponds to a single scan, while the blue line indicates the fitted generalized additive model with shaded 95% confidence intervals. Notable expansions in network size were observed in the cerebellum, temporal, attention, frontal-dlPFC, and frontal-vlPFC networks. All significance tests were corrected for multiple comparisons using FDR procedures (n = 74 biologically independent participants).

Fig. 9 Developmental shifts in the network center of mass (NCM) which is reported in voxel distance, during early infancy. [Images not available. See PDF.]

These scatter plots illustrate how NCM—a measure reflecting the spatial centroid of each network—changes with corrected age across 13 brain networks. Red dots represent individual scan values, and blue lines depict the GAM fits with 95% confidence intervals. Several networks, including auditory, temporal, secondary motor, and frontal-mPFC, showed significant spatial shifts over age, indicating age-related reorganization in the spatial distribution of these networks. Multiple comparisons were controlled using FDR correction (n = 74 biologically independent participants).

Network-averaged spatial similarity

The results reveal that network-averaged spatial similarity shows a significant increase across ages for all networks ( $F>5.45,p<0.02$ ; Fig. 5) except for the subcortical network. Notably, the primary visual, secondary motor, frontal-mPFC, and frontal-vlPFC networks showed the steepest rise in spatial similarity within the first 3 months, indicating early developmental consolidation. In contrast, networks such as the default mode and auditory networks exhibited a more gradual increase over the full 6-month period, suggesting a prolonged trajectory of spatial refinement. The subcortical network, however, showed minimal changes in spatial similarity across age, potentially reflecting its early functional maturity and relative stability compared to cortical networks, which undergo more dynamic spatial reorganization.

Network engagement range

A significant decline in network engagement range was observed across multiple networks, including the primary visual, primary motor, secondary motor, temporal, cerebellum, attention, default mode, and frontal-vlPFC networks, indicating that variability in voxel contribution is generally decreasing over time ( $F>6.27,p<0.02$ ; Fig. 6). The default mode network exhibited a noticeable decline in network engagement range during the first 3 months, followed by a more gradual decrease between 3 to 6 months, reflecting reduced variability in voxel-level signal contributions across time. In contrast, the attention network showed an inverse pattern, with a slower decline in the first 3 months that became more pronounced in the latter half of the 6-month period. Additionally, the primary visual, temporal, and cerebellum networks demonstrated a continuous decline in network engagement range throughout the 6-month period, indicating a consistent reduction in voxel-level variability within these networks over time. Conversely, the secondary visual, subcortical, auditory, frontal-mPFC, and frontal-dlPFC networks did not exhibit a significant decline in network engagement range, suggesting that these networks maintain a more stable voxel engagement profile.

Network strength

Network strength shows a significant increase in the primary and secondary visual networks, primary and secondary motor networks, as well as in the temporal, frontal-mPFC, and frontal-vlPFC regions ( $F>5.08,p<0.04$ ; Fig. 7). The frontal-mPFC, frontal-vlPFC, and secondary motor networks exhibited a sharp increase during the first 3 months, followed by a stabilization phase between 3–6 months, suggesting an early period of heightened functional integration. In contrast, the primary visual and primary motor networks showed a gradual but continuous rise over the entire 6-month period, reflecting a steady strengthening of functional engagement. Meanwhile, the cerebellum, attention, subcortical, default mode, auditory, and frontal-dlPFC networks remained largely unchanged, indicating a more stable level of network strength throughout early infancy.

Network size

Network size exhibited notable expansion across several networks, with significant increases observed in the cerebellum, attention, temporal, frontal-dlPFC, and frontal-vlPFC regions ( $F>6.33,p<0.006$ ; Fig. 8). This growth was particularly pronounced in the cerebellum and frontal-dlPFC networks during the first 3 months, suggesting an early phase of structural reorganization, followed by a more gradual increase up to 6 months. In contrast, the temporal, attention, and frontal-vlPFC networks demonstrated a steady upward trajectory over the entire 6-month period, indicating continuous network expansion throughout early development. Notably, no significant reductions in network size were detected in any brain region, highlighting the overall trend of increasing spatial engagement within functional networks during infancy. Meanwhile, networks such as the primary visual, subcortical, and primary motor regions remained relatively stable in size, implying that their structural organization may be established earlier and undergo less pronounced spatial restructuring over this developmental window.

Network center of mass

Lastly, network center of mass exhibited distinct developmental trajectories across different networks. The temporal, auditory, default mode, and secondary motor networks showed a progressive shift in their center of mass ( $F>4.02,p<0.008$ ; Fig. 9). The temporal and auditory networks showed a steady increase in network center of mass across the full 6-month period, suggesting a gradual spatial shift as these networks matured. In contrast, the secondary motor and displayed an initial rise within the first 3 months, followed by stabilization between 3 to 6 months, indicating an early reorganization phase before reaching a more stable state. In addition, the default mode network shows a nonlinear trajectory, with an initial decline in the first few months, a rise around 3 months, and stabilization in the later half of the 6-month period. Conversely, the frontal-mPFC and secondary visual networks experienced a significant decline in network center of mass ( $F>5.52,p<0.009$ ; Fig. 9) during the first 3 months, with little change thereafter, possibly reflecting early consolidation of their spatial organization. Meanwhile, the primary visual, subcortical, cerebellum, primary motor, attention, frontal-dlPFC, and frontal-vlPFC networks maintained relatively stable network center of mass values, indicating minimal spatial displacement throughout the observed period. To address whether differences in overall brain volume may account for the observed age-related changes in network size and center of mass, we performed additional GAM analyses including total brain size as a covariate. Results showed that, across most networks, associations with corrected age remained consistent (Fig. S6 in Section 1.6 of Supplementary Note). All the p values are after FDR-correction.

Discussion

The investigation into the spatial development of large-scale brain networks during the first six postnatal months of life holds significant implications for understanding early brain development and potential markers of neurodevelopmental disorders. The study, conducted with a cohort of 74 infants, utilizes ICA on rsfMRI data. A number of studies have characterized the emergence of functional brain networks during infancy, and our findings generally align with these developmental patterns. Across studies, sensorimotor, auditory, and visual networks consistently emerge earliest, often present at or shortly after birth^12,13. For instance, Fransson et al. used ICA to identify primary visual, sensorimotor, auditory, posterior (Precuneus/lateral parietal/cerebellum), and anterior PFC networks in neonates, supporting the presence of both primary and rudimentary association networks at birth¹³. Gao et al. reported a sequential trajectory, in which sensorimotor and auditory systems develop first, followed by visual networks, then attention and default mode, and finally executive control systems¹². Similarly, Damaraju et al. found increasing coherence within the posterior DMN between 4 and 9 months, and noted that executive and frontoparietal networks showed more prolonged maturation⁶⁶.

Our findings align with the proposed developmental cascade, as we identified 13 distinct functional networks within the first 6 months of life, including both early-appearing primary sensory networks and the presence of higher-order networks such as attention and frontal systems—suggesting these systems are already emerging, though not yet fully mature. However, methodological differences likely account for some cross-study variability. For example, Gao et al. used seed-based functional connectivity and identified salience networks and lateralized frontoparietal networks, which we did not observe. In contrast, our ICA-based analysis identified networks related to cerebellum, subcortical, attention, and three distinct frontal regions, which were not present in Gao’s results^12,67. These discrepancies may be attributed to the approach to motion correction and the use of ICA⁶⁷. ICA is well-suited for infant fMRI because it is more resistant to head motion and can separate overlapping brain signals in a data-driven way. Supporting this, Damaraju et al. reported consistent network structure using ICA at different model orders (30 and 100)⁶⁶, many of which—such as visual, auditory, default mode, and frontal networks—closely resemble ours. Our use of a lower model order (20) likely explains differences in the number and specificity of sub-networks across domains. Finally, Fitzgibbon et al. identified 13 neonatal networks using both ICA and seed-based methods in infants aged 26–45 weeks PMA⁶⁸. While many of their networks visually resemble ours, they did not assign explicit labels, and notably did not identify a distinct frontal network, possibly reflecting the earlier developmental stage of their sample. This discrepancy may also reflect both methodological and age-related factors. Future research should further investigate the presence and characteristics of various networks identified in adults to better understand their developmental trajectories and manifestations in infants.

Inter-network modularity patterns across development

The voxel-level age-related analysis revealed a complex, region-specific pattern of developmental changes in brain networks during the first six postnatal months. Predominantly positive correlations between voxel intensity and corrected age were observed in primary sensory and motor networks—including the visual, motor, and auditory systems—suggesting robust maturation and strengthening of these networks as sensory and motor functions become increasingly refined during early infancy^69,70. These results align with prior work demonstrating that sensory pathways develop rapidly after birth to support critical functions such as vision, movement, and auditory processing^12,71,72. In contrast, higher-order association networks, such as the frontal-mPFC, frontal-dlPFC, and frontal-vlPFC, demonstrated a more complex pattern, with substantial regions of both positive and negative correlations. Rather than attributing this solely to general development or synaptic pruning^{73, 74–75}, these mixed patterns may reflect early processes of functional differentiation, where immature distributed connectivity begins to reorganize into more specialized and efficient networks⁷⁶. The coexistence of positive and negative voxel-age correlations across networks underscores that early brain development is not a uniform process. These findings highlight the importance of considering both region-specific maturation rates and the interplay between growth and refinement when characterizing normative brain development in infancy⁷⁷.

To further examine how these developing networks interact, we conducted a modularity analysis, which identified two primary modules within the infant brain network architecture. This modular organization reflects distinct patterns of network interaction, offering insights into how coordinated systems emerge during early development. The first module, encompassing attention, frontal-vlPFC, frontal-mPFC, frontal-dlPFC, default mode, primary motor, and secondary visual networks, largely aligns with regions involved in higher-order cognitive functions and integrative processes. This module’s composition suggests a coordinated development of cognitive control, attention regulation, and sensory processing during early infancy^78,79. The second module, consisting of primary visual, secondary motor, cerebellum, subcortical, auditory, and temporal networks, indicates a focus on sensory processing, motor control, and basic cognitive functions. This modular organization supports the view that early brain development prioritizes the formation of core functional systems: by clustering these basic systems into a cohesive module, the brain may be optimizing for efficient integration and co-development of sensory-motor pathways essential for survival and early learning⁸⁰. The presence of these tightly linked, low-level networks in a distinct module suggests that early functional architecture is organized to scaffold later-emerging, more complex cognitive networks⁸⁰. Supporting this, previous research examining the functional connectivity of rs-fMRI in infants aged 0–1 year demonstrated that as infants mature, the brain’s functional network becomes increasingly subdivided into a greater number of distinct functional modules⁸¹. This developmental shift reflects the transition from broad, less specialized connectivity patterns to more refined and segregated brain networks as cognitive and sensory processing capacities expand. This finding aligns with our study’s observations, which identify a division of brain networks into two primary categories: higher-order cognitive functions and sensory processing/motor control. The gradual subdivision of functional modules supports the notion that, even from an early age, the brain organizes itself into specialized networks that cater to different cognitive and sensory-motor functions. The segregation into two distinct modules highlights the early specialization and integration of functional brain networks, essential for developing complex behaviors and cognitive abilities in infants^12,82.

Spatial metrics

Across all five spatial metrics examined—network-averaged spatial similarity, network engagement range, network strength, network size, and network center of mass—a consistent pattern of dynamic maturation emerges during the first 6 months of life. Collectively, these metrics reveal a shift toward increased spatial coherence, reduced variability in voxel participation, and expansion or stabilization of functionally relevant territories. This convergence suggests that the infant brain undergoes both refinement and organization of its functional architecture at multiple spatial levels. Importantly, the alignment of these diverse measures supports the idea that early brain development involves coordinated changes in brain structure and function, with networks gradually forming adult-like patterns while still allowing flexibility for continued growth.

Network-averaged spatial similarity

The noted rise in network-averaged spatial similarity, as measured by the correlation between individual spatial maps of each brain network and the group-level spatial map of the same network, reveals intriguing insights into the consistency of network organization. This phenomenon suggests a convergence over age towards a more standardized spatial configuration across various brain networks. The heightened network-averaged spatial similarity implies that the inherent patterns and structures characterizing each network tend to align more closely with the group-level representation over time. The results specifically reveal significant age-related increases in spatial similarity across nearly all networks, particularly steep in primary visual, secondary motor during the first 3 months, indicating growing alignment of individual spatial maps with the group template. This rapid change implies an early “settling” of these networks’ spatial maps. Such a sequence is consistent with prior findings that primary sensorimotor regions mature earlier than higher-order networks⁸⁰. Higher-order networks, such as the default mode network, exhibited a more gradual refinement over the full 6-month period, reflecting prolonged maturation. This observation aligns with developmental models where primary networks stabilize first and associative or default-mode systems refine later⁸³. In contrast, the subcortical network displayed minimal changes, likely reflecting its early functional maturity and inherent stability from birth compared to cortical regions⁸⁴. As the group-level map reflects the common spatial organization of brain networks shared across all participants, an increase in network-averaged spatial similarity with age indicates that an individual’s network is becoming more aligned with the group’s spatial pattern. In contrast, a lower network-averaged spatial similarity value implies a greater divergence from the group-level pattern, indicating more individual-specific variability in the network’s spatial map. This trend aligns with a reduction in individual variability, consistent with earlier research that observed decreased variability in the functional architecture of the prenatal brain as it develops⁸⁵. This reduction likely contributes to the establishment of network foundations, which may later underpin individual behavioral differences⁸⁵. However, our findings diverge from a study in which younger infants demonstrated greater similarity to the group-level map at younger ages⁸⁶. The consistency with Xu et al. may be attributed to their emphasis on spatial patterns of voxel-wise variability, as their metric was spatially focused⁸⁵. In other words, Xu et al.‘s method captures developmental trends in the across-subject variability of voxel values, whereas ours evaluates the alignment of each subject to a standardized group spatial pattern. Although both are spatially oriented, network-averaged spatial similarity is more directly focused on congruence with an age-appropriate reference. In contrast, the discrepancy with Moore et al. may reflect methodological differences: their individualized network maps were derived using a seed-based correlation and template-matching approach, which allows for spatial topography to vary across individuals. Our use of group ICA, by contrast, defines a fixed set of spatial components at the group level, and evaluates subject-level alignment to this common spatial framework⁸⁶.

Network engagement range

We examined the network engagement range, which quantifies the difference between the highest positive and lowest negative values of brain voxels within a specific network. A larger network engagement range signifies a greater disparity in voxel intensity. Put simply, if the network engagement range decreases across age, it suggests network consolidation over time, as voxel intensities become closer, indicating a reduction in random intensity fluctuations and more consistent voxel contribution across the network. In our analysis, a significant decline in network engagement range was observed across most networks, including primary visual, primary and secondary motor, temporal, cerebellum, attention, default mode, and frontal-vlPFC networks, with other networks showing no significant changes. Specifically, the default mode network exhibited a rapid initial decrease in network engagement range within the first 3 months, followed by a more gradual decline, suggesting early stabilization of its core regions, consistent with developmental trajectories reported by Yin et al.⁸⁷. In other words, the default mode network appears to stabilize sooner than many high-order networks. Conversely, the attention network displayed a slower initial decline that became more pronounced later in the 6-month period, aligning with documented delayed maturation patterns of attention-related systems⁸⁷. In addition, most of the other networks are still undergoing refinement over the first postnatal months. For example, Lin et al. and Yin et al.^87,88 have shown that the strength and extent of sensorimotor networks increases more rapidly than those of visual networks throughout early childhood^87,88, consistent with our results showing that the network engagement range decreases in primary and secondary motor networks at a greater rate than in visual networks. On the other hand, the network engagement range of the auditory network remains stable during the first 6 months of life. This stability may be due to the fact that the auditory system is one of the earliest functions to develop in utero. The auditory network is already functioning at birth and is among the first networks to reach an adult-like state^8,10,89. This early maturation enables infants to process sound even in utero, which may explain why no significant changes are observed in the auditory network’s network engagement range during this early developmental period⁸⁹. Fluctuations in the network engagement range across diverse age groups may signify alterations in the range of brain activity. Moreover, changes in the network engagement range may indicate reorganization within brain networks. A widening network engagement range suggests that different regions within the network are engaging differently. This broader range might occur when the brain adapts to new cognitive demands or reorganizes in response to learning, development, or injury³². Conversely, a narrower network engagement range, indicating more consistent voxel contribution, could suggest a refinement of the network. This might occur when the brain consolidates functions into more specialized regions, reducing the need for broad engagement⁹⁰.

Network strength

Trajectories of spatial characteristics, particularly the observed increase in network strength during the first 6 months of life, suggest greater voxel involvement within the network across age. This increased involvement may reflect the dynamic interplay of neurodevelopmental processes. During this period, myelination enhances signal transmission speed and efficiency, supporting the integration of more voxels into coherent network activity. Simultaneously, synaptogenesis leads to the formation of new synapses, expanding the potential for connectivity within networks^76,91,92. Although synaptic pruning reduces the total number of synapses, it selectively strengthens important connections, potentially contributing to more efficient and focused network activation^76,91,92. Thus, the observed greater voxel involvement aligns with the idea that these developmental processes facilitate the maturation of functional networks by balancing broad connectivity with the refinement of specific pathways. These complementary processes integrate more voxels into coherent networks and strengthen the most important pathways, thereby increasing BOLD signal coherence. Indeed, infants show steep rises in cerebral blood flow and metabolism to meet these growing demand⁸³. Our data reveal region-specific trajectories in network strength: primary visual and motor networks exhibited a steady, continuous increase throughout the first 6 months, consistent with early and prolonged maturation of core sensory systems. These regions benefit from early-occurring myelination, aligning with prior findings that visual and parietal areas develop rapidly to support foundational perceptual and motor abilities^12,93. In contrast, the frontal-mPFC, frontal-vlPFC, and secondary motor networks showed a sharp rise in strength during the first 3 months, followed by a plateau— suggesting varied spatial growth patterns, indicative of the complex interplay between neural growth and pruning during early developmental stages⁹⁴, possibly reflecting a burst of synaptic growth and early circuit integration in frontal areas, later constrained by stabilization and pruning^91,92. Other networks such as the cerebellum, attention, subcortical, default mode, auditory, and frontal-dlPFC remained relatively unchanged, reflecting their later-developing roles and the protracted maturation trajectories observed in prior longitudinal studies. Moreover, the heightened network strength could be indicative of increased contribution to the network’s function, and neural activity within these brain networks, which may suggest a crucial developmental phase characterized by the refinement and optimization of neural circuits^71,95.

Network size

Networks exhibiting a significant increase in size may reflect regions undergoing rapid growth, likely influenced by key developmental processes such as synaptogenesis, myelination, and overall gray matter expansion^{96, 97, 98–99}. For example, the observed increase in cerebellar network size (0.08% per day) could be attributed to the cerebellum’s rapid growth in both volume and connectivity during early development¹⁰⁰. This expansion facilitates its integration into larger brain networks involved in motor and cognitive functions. Studies have shown that the cerebellum undergoes significant growth during the first year of life, primarily driven by synaptogenesis, which establishes synaptic connections between neurons, as well as broader neural remodeling processes, including dendritic and axonal growth. These changes contribute to the expansion of neural circuits and may also involve non-synaptic interactions such as gap junctions, further enhancing cerebellar connectivity and function¹⁰⁰. Moreover, Fossella et al. suggest that increased synaptogenesis during infancy and early childhood plays a key role in the maturation of attention networks, which are critical for cognitive functions like working memory and problem-solving¹⁰¹. This biological process is closely tied to the observed developmental expansion of network size, particularly in the attention network (0.04% per day). The pronounced growth in cerebellar and frontal-dlPFC networks during the first 3 months aligns with the documented surge in gray matter and myelination in early infancy. Structural MRI and DTI studies reveal that cortical and subcortical gray-matter volumes more than double within the first year, with subcortical regions—including the cerebellum—showing the steepest early increases^3,102. Additionally, the observed increase in the size of the temporal (0.08%/day), frontal-dlPFC (0.06%/day), and frontal-vlPFC (0.08%/day) networks indicates significant developmental changes that could be related to the enhancing of cognitive and social functions in infants. For instance, one previous study showed, during the first 2 years of life, the brain’s network topology evolves, with an increase in long-distance connections and strengthening of functional connectivity, particularly in the temporal regions¹⁰³. On the other hand, networks such as the primary and secondary visual, primary and secondary motor, subcortical, default mode, auditory, and frontal-mPFC exhibit relatively stable network sizes, suggesting that these regions may be approaching structural stability¹⁰⁴. However, this does not preclude continued developmental refinements, as functional reorganization and microstructural changes may still be occurring. This variation across networks highlights the complexity of early neurodevelopment, emphasizing that different brain regions may mature at distinct rates and through different mechanisms. Given this diversity, a network-specific approach is essential for capturing the nuanced patterns of growth, stabilization, and functional specialization across the infant brain.

Network center of mass

Our analyses revealed distinct developmental trajectories in the spatial distribution of network activity across infancy. Several networks, including the temporal, auditory, default mode, and secondary motor networks, exhibited a progressive shift in their network center of mass, suggesting dynamic spatial reorganization during early postnatal development. Specifically, the temporal and auditory networks demonstrated a steady increase over 6 months, possibly reflecting expanding functional territories as these systems mature and integrate with broader cortical regions¹⁰⁵. In contrast, the secondary motor network showed an early shift within the first 3 months, followed by spatial stabilization, indicating a reorganization phase that may support emerging motor capabilities before settling into a more consistent configuration. The default mode network displayed a nonlinear trajectory—initially shifting posteriorly, then anteriorly, and stabilizing thereafter—possibly reflecting early architectural tuning as the default mode network’s core hubs consolidate. Conversely, the frontal-mPFC and secondary visual networks exhibited a significant decrease in network center of mass during the first 3 months, followed by minimal change, suggesting that their spatial architecture becomes established early in development. Other networks—including the primary visual, subcortical, cerebellum, primary motor, attention, frontal-dlPFC, and frontal-vlPFC—maintained relatively stable center of mass values over time, suggesting limited displacement and potentially earlier maturation or spatial anchoring. These spatial patterns underscore that while some networks undergo dynamic shifts in spatial localization, others achieve a stable configuration early, reflecting a diversity in timing and mechanisms of functional maturation across network systems. This approach builds upon conventional functional connectivity analyses, which have established that the physical distance between nodes plays a crucial role in shaping network connectivity¹⁰⁶. However, rather than focusing solely on temporal correlations, our study examines how voxel contributions within networks are spatially distributed relative to the network’s center of mass. Our findings indicate that spatial distance also influences voxel engagement, suggesting that as brain networks develop, their organization undergoes continuous refinement.

Taken together, these results underscore that the first 6 months of life represent a foundational phase of spatial refinement in brain network organization. The increase in spatial similarity and strength, along with decreasing engagement range and distinct shifts in network center of mass, reflect an evolving balance between broad connectivity and emerging specificity. This phase appears to be guided by regionally distinct neurobiological processes such as myelination, synaptogenesis, and gray matter expansion—processes known to occur at varying rates across sensory, motor, and associative regions⁹³. For example, posterior regions associated with vision and motor functions myelinate earlier and exhibit more rapid stabilization, while frontal and associative regions show more prolonged trajectories⁹³. These findings collectively point to a sensitive window of spatial tuning that likely lays the groundwork for later functional specialization and behavioral development. Future studies should explore how these early spatial transformations relate to long-term cognitive outcomes and whether deviations from these normative patterns signal risk for developmental disorders.

Limitations

This study offers valuable insights into the normative development of brain networks during the crucial first 6 months of life. However, a key limitation of this study is the lack of diversity in the participants. Our participants were predominantly White and from high socioeconomic status backgrounds, and as such are not representative of the racial and sociodemographics of Atlanta, Georgia, United States (Where this study was based). As such, the limited diversity in the sample may restrict the generalizability of the findings. There are also technical constraints that warrant consideration. The spatial resolution of infant fMRI scans is inherently lower than that of adult imaging, partly due to anatomical differences and scanner constraints optimized for pediatric safety. This reduced resolution can limit the precision with which fine-scale network structures are delineated. Furthermore, the relatively short scan durations—necessary to accommodate infant tolerance—may reduce the signal-to-noise ratio and limit the stability of functional network estimates. While our results were robust across multiple metrics and control analyses, future studies leveraging higher-resolution imaging and longer acquisition times in natural sleep or sedated settings may provide additional insights into early brain network development. In addition, another key limitation of scanning infants during sleep is the potential influence of sleep stage variability on functional connectivity measures. Previous studies have shown that sleep states, particularly quiet sleep and active sleep, exhibit distinct patterns of neural activity that can impact functional connectivity estimates¹⁰⁷. Moreover, as infants develop, the proportion of time spent in different sleep stages changes, potentially introducing age-related variability in observed network configurations¹⁰⁸. Given that most infant neuroimaging studies rely on natural sleep to minimize motion artifacts, future research should consider incorporating sleep-stage monitoring, such as EEG-fMRI approaches, to disentangle developmental changes in brain networks from transient fluctuations related to sleep. An additional limitation concerns the characterization of the subcortical network, which showed minimal age-related change in spatial similarity across early development. While this may reflect the early functional maturity and inherent stability of subcortical regions, it is also possible that technical factors, such as lower spatial resolution and reduced signal-to-noise ratio (SNR) in these deep brain areas, limited our sensitivity to detect subtle developmental changes. Subcortical structures are smaller and more compact than cortical networks, making them more vulnerable to partial volume effects and normalization inaccuracies, particularly in infant imaging. Despite applying uniform preprocessing steps and using optimized smoothing parameters, these anatomical and imaging constraints may have masked finer developmental variations. Future work utilizing higher-resolution acquisitions and advanced denoising methods may offer improved insight into the spatial maturation of subcortical networks during infancy.

Future work

A crucial area for future exploration is the study of the 105 intrinsic connectivity networks (ICNs), which are derived from the multiscale functional brain network templates developed through the NeuroMark pipeline^41,109. This method applies a combination of independent component analysis (ICA) and large-scale data integration from projects such as the Human Connectome Project (HCP) and the Genomics Superstruct Project (GSP) to identify reproducible ICNs. These networks span multiple spatial scales, capturing both large-scale global networks and smaller, localized circuits^41,109. Understanding how these ICNs emerge and evolve during early development could provide critical insights into the foundational architecture of brain networks. Additionally, because these templates were derived from diverse, large-scale datasets, they offer a robust framework for comparing infant brain development to patterns seen in adults, enabling a better understanding of how early brain structures mature over time.

In this study, we focused on the global spatial characteristics of functional brain networks, including network-averaged spatial similarity, engagement range, network strength, size, and center of mass. However, another important aspect of spatial network organization is laterality, which refers to differences in functional network properties between the left and right hemispheres. Future research should investigate how the developmental trajectories of these spatial characteristics may vary across hemispheres and whether certain networks exhibit asymmetric maturation patterns. Exploring laterality could provide further insights into functional specialization and its implications for cognitive development. Prior studies have highlighted the role of laterality in brain organization^110,111, and incorporating these insights into developmental network analysis would be an important direction for future work.

Another avenue for future research could involve examining how the spatial maps of brain networks evolve across multiple time scales—ranging from short-term fluctuations linked to sleep stages or arousal states, to long-term developmental transformations occurring over weeks or months. Such an analysis would provide valuable insights into the dynamic neural processes that underpin cognitive and behavioral development²⁵. Additionally, exploring these temporal variations could aid in identifying early biomarkers of neurodevelopmental disorders, paving the way for earlier and more targeted intervention strategies to optimize developmental outcomes. This line of inquiry could significantly enhance our understanding of how transient and enduring changes in network organization impact brain maturation and function over time.

Conclusion

In conclusion, this study provides a comprehensive examination of the spatial development of large-scale brain networks during the first 6 months of life, offering critical insights into the normative processes of early brain maturation. Our findings reveal significant age-related changes in spatial characteristics, highlighting both linear and nonlinear developmental trajectories across different networks. Specifically, network-averaged spatial similarity increased with age, indicating greater alignment between individual and group-level network maps, suggesting a maturation-driven convergence towards more stable spatial configurations. At the same time, the network engagement range decreased, reflecting a consolidation process where voxel contributions became more uniform. Furthermore, network strength increased with age in key networks, such as the frontal-mPFC and visual networks, suggesting enhanced functional engagement. Network size and center of mass exhibited distinct developmental trends, with the secondary visual network expanding over time, while the temporal network showed a reduction in size, possibly reflecting regional specialization. Notably, non-linear growth patterns were particularly evident in the secondary motor and frontal-mPFC networks, underscoring the complexity of early brain development. These findings emphasize that early neurodevelopment is characterized by both the stabilization of core networks and the dynamic reorganization of others. While this study provides valuable insights into normative brain maturation, future research should incorporate at-risk populations and extend longitudinally to explore the long-term implications of these early network changes. By refining our understanding of these early neural trajectories, this research holds promise for informing early detection and intervention strategies in neurodevelopmental disorders.

Acknowledgements

We greatly appreciate the families and their infants who volunteered to participate in this research study. We would also like to thank the research coordinators, assistants, and fellows at the Marcus Autism Center, Brittney Sholar, Carly Reineri, Joannna Beugnon, Lindsey Evans, Jordan Pincus, Jennifer Gutierrez, Tristan Ponzo, and Adriana Mendez, and the MRI Techs at the Emory Center for Systems Imaging Core, Michael White, Sarah Basadre, and Samira Yeboah, for their data collection efforts, as well as Dr. Lei Zhou and Michael Valente for their assistance with equipment development and data acquisition protocols. This work was supported by the National Institutes of Health (NIH) grant number R01 R01MH119251 to S.S. and A.I., R01EB027147 to S.S. and V.C., P50MH100029 to S.S., and National Science Foundation (NSF) grant number 2112455 to V.C.

Author contributions

M.S. led the research project as the first author, contributing substantially to the conception, design, and overall execution of the study, as well as drafting the manuscript. S.S. was instrumental in data collection and writing, ensuring the accuracy and integrity of the collected data. Z.F. and Q.L. played a key role in data preprocessing and contributed significantly to the manuscript drafting. A.I., S.S., and V.C. provided expert supervision throughout the study, offering critical insights and guidance, and contributed to the manuscript’s writing. All authors reviewed and approved the final manuscript, ensuring its intellectual rigor and coherence.

Peer review

Peer review information

Communications Biology thanks Jiaxin Cindy Tu, Wenjiao Lyu and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Primary Handling Editors: Sahar Ahmad and Jasmine Pan.

Data availability

Data collected from National Institute of Mental Health (NIMH) 2P50MH100029, R01MH118285, and R01MH119251 are available from the NIMH Data Archive (NDA). NDA is a collaborative informatics system created by the National Institutes of Health to provide a national resource to support and accelerate research in mental health. Data set identifiers: https://doi.org/10.15154/vjcy-f589. The manuscript reflects the views of the authors and may not reflect the opinions or views of the NIH.

Code availability

Code is available on request from the authors ([email protected]).

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s42003-025-08913-z.

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

References

1. Nielsen, AN et al. Maturation of large-scale brain systems over the first month of life. Cereb. Cortex; 2023; 33, pp. 2788-2803. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/35750056][DOI: https://dx.doi.org/10.1093/cercor/bhac242]

2. Shonkoff, J. P. & Phillips, D. A. The developing. Brain pp. 183–217 (2000).

3. Gilmore, JH; Knickmeyer, RC; Gao, W. Imaging structural and functional brain development in early childhood. Nat. Rev. Neurosci.; 2018; 19, 123.1:CAS:528:DC%2BC1cXivVSiur4%3D [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/29449712][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5987539][DOI: https://dx.doi.org/10.1038/nrn.2018.1]

4. Seraji, M; Mohebbi, M; Safari, A; Krekelberg, B. Multiple sclerosis reduces synchrony of the magnocellular pathway. PLoS ONE; 2021; 16, e0255324.1:CAS:528:DC%2BB3MXhvFCjs7%2FE [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/34437558][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8389379][DOI: https://dx.doi.org/10.1371/journal.pone.0255324]

5. Ebrahimzadeh, E et al. Simultaneous electroencephalography-functional magnetic resonance imaging for assessment of human brain function. Front. Syst. Neurosci.; 2022; 16, 934266. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/35966000][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9371554][DOI: https://dx.doi.org/10.3389/fnsys.2022.934266]

6. Seraji, M., Ellis, C. A., Sendi, M. S. E., Miller, R. L. & Calhoun, V. D. Uncovering effects of schizophrenia upon a maximally significant, minimally complex subset of default mode network connectivity features. In 2024 46th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) 1–4 https://doi.org/10.1109/EMBC53108.2024.10782953 (2024).

7. Duerden, EG; Card, D; Lax, ID; Donner, EJ; Taylor, MJ. Alterations in frontostriatal pathways in children born very preterm. Dev. Med. Child Neurol.; 2013; 55, pp. 952-958. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/23859594][DOI: https://dx.doi.org/10.1111/dmcn.12198]

8. Smyser, CD; Snyder, AZ; Neil, JJ. Functional connectivity MRI in infants: exploration of the functional organization of the developing brain. Neuroimage; 2011; 56, 1437. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/21376813][DOI: https://dx.doi.org/10.1016/j.neuroimage.2011.02.073]

9. Gao, W et al. Evidence on the emergence of the brain’s default network from 2-week-old to 2-year-old healthy pediatric subjects. Proc. Natl. Acad. Sci. USA; 2009; 106, pp. 6790-6795.1:CAS:528:DC%2BD1MXlsV2hu7c%3D [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/19351894][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2672537][DOI: https://dx.doi.org/10.1073/pnas.0811221106]

10. Doria, V et al. Emergence of resting state networks in the preterm human brain. Proc. Natl. Acad. Sci. USA; 2010; 107, pp. 20015-20020.1:CAS:528:DC%2BC3cXhsVyitLrN [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/21041625][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2993415][DOI: https://dx.doi.org/10.1073/pnas.1007921107]

11. Eyre, M et al. The Developing Human Connectome Project: typical and disrupted perinatal functional connectivity. Brain; 2021; 144, pp. 2199-2213. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/33734321][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8370420][DOI: https://dx.doi.org/10.1093/brain/awab118]

12. Gao, W et al. Functional network development during the first year: relative sequence and socioeconomic correlations. Cereb. Cortex; 2015; 25, 2919. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/24812084][DOI: https://dx.doi.org/10.1093/cercor/bhu088]

13. Fransson, P et al. Resting-state networks in the infant brain. Proc. Natl. Acad. Sci. USA; 2007; 104, pp. 15531-15536.1:CAS:528:DC%2BD2sXhtFajt7%2FP [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/17878310][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2000516][DOI: https://dx.doi.org/10.1073/pnas.0704380104]

14. Gao, W; Alcauter, S; Smith, JK; Gilmore, JH; Lin, W. Development of human brain cortical network architecture during infancy. Brain Struct. Funct.; 2015; 220, pp. 1173-1186. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/24469153][DOI: https://dx.doi.org/10.1007/s00429-014-0710-3]

15. Fouladivanda, M. et al. Multi-scale structural rich-club organization of the brain in full-term newborns: a combined DWI and fMRI study. J. Neural Eng.18, https://doi.org/10.1088/1741-2552/abfd46 (2021).

16. Sadeghi, N et al. The interplay between brain and behavior during development: a multisite effort to generate and share simulated datasets. Sci. Data; 2025; 12, pp. 1-8. [DOI: https://dx.doi.org/10.1038/s41597-025-04740-3]

17. Uddin, L. Q., Supekar, K. & Menon, V. Typical and atypical development of functional human brain networks: insights from resting-state FMRI. Front. Syst. Neurosci.4, 21 (2010).

18. Smyser, CD; Neil, JJ. Use of resting-state functional MRI to study brain development and injury in neonates. Semin. Perinatol.; 2015; 39, pp. 130-140. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/25813667][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4379452][DOI: https://dx.doi.org/10.1053/j.semperi.2015.01.006]

19. Smyser, CD et al. Longitudinal analysis of neural network development in preterm infants. Cereb. Cortex; 2010; 20, pp. 2852-2862. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/20237243][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2978240][DOI: https://dx.doi.org/10.1093/cercor/bhq035]

20. Tooley, UA et al. Prenatal environment is associated with the pace of cortical network development over the first three years of life. Nat. Commun.; 2024; 15, pp. 1-16. [DOI: https://dx.doi.org/10.1038/s41467-024-52242-4]

21. Wen, X., Hsu, L., Lin, W., Zhang, H. & Shen, D. Co-evolution of functional brain network at multiple scales during early infancy. ArXivhttps://arxiv.org/abs/2009.06899?utm_source=chatgpt.com (2020).

22. Iraji, A et al. Spatial dynamics within and between brain functional domains: a hierarchical approach to study time-varying brain function. Hum. Brain Mapp.; 2019; 40, pp. 1969-1986. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/30588687][DOI: https://dx.doi.org/10.1002/hbm.24505]

23. Bijsterbosch, J. D. et al. The relationship between spatial configuration and functional connectivity of brain regions. Elife7, e44890 (2018).

24. Iraji, A; Miller, R; Adali, T; Calhoun, VD. Space: a missing piece of the dynamic puzzle. Trends Cogn. Sci.; 2020; 24, pp. 135-149. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/31983607][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7809367][DOI: https://dx.doi.org/10.1016/j.tics.2019.12.004]

25. Iraji, A et al. The spatial chronnectome reveals a dynamic interplay between functional segregation and integration. Hum. Brain Mapp.; 2019; 40, pp. 3058-3077. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/30884018][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6548674][DOI: https://dx.doi.org/10.1002/hbm.24580]

26. Iraji, A et al. Spatial dynamic subspaces encode sex-specific schizophrenia disruptions in transient network overlap and its links to genetic risk. bioRxiv; 2023; [DOI: https://dx.doi.org/10.1101/2023.07.18.548880] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/38014214][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10462123]

27. Garcia, JO; Ashourvan, A; Muldoon, S; Vettel, JM; Bassett, DS. Applications of community detection techniques to brain graphs: algorithmic considerations and implications for neural function. Proc. IEEE; 2018; 106, pp. 846-867. [DOI: https://dx.doi.org/10.1109/JPROC.2017.2786710]

28. Shultz, S; Klin, A; Jones, W. Neonatal transitions in social behavior and their implications for autism. Trends Cogn. Sci.; 2018; 22, pp. 452-469. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/29609895][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6554740][DOI: https://dx.doi.org/10.1016/j.tics.2018.02.012]

29. Holland, D et al. Structural growth trajectories and rates of change in the first 3 months of infant brain development. JAMA Neurol.; 2014; 71, pp. 1266-1274. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/25111045][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4940157][DOI: https://dx.doi.org/10.1001/jamaneurol.2014.1638]

30. Iraji, A et al. Multi-spatial-scale dynamic interactions between functional sources reveal sex-specific changes in schizophrenia. Netw. Neurosci.; 2022; 6, 357. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/35733435][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9208002][DOI: https://dx.doi.org/10.1162/netn_a_00196]

31. Du, Y; Fan, Y. Group information guided ICA for fMRI data analysis. Neuroimage; 2013; 69, pp. 157-197. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/23194820][DOI: https://dx.doi.org/10.1016/j.neuroimage.2012.11.008]

32. Bassett, DS; Sporns, O. Network neuroscience. Nat. Neurosci.; 2017; 20, pp. 353-364.1:CAS:528:DC%2BC2sXjsVSmsrY%3D [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/28230844][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5485642][DOI: https://dx.doi.org/10.1038/nn.4502]

33. Molloy, MF; Saygin, ZM. Individual variability in functional organization of the neonatal brain. Neuroimage; 2022; 253, 119101. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/35304265][DOI: https://dx.doi.org/10.1016/j.neuroimage.2022.119101]

34. Kong, R et al. Spatial topography of individual-specific cortical networks predicts human cognition, personality, and emotion. Cereb. Cortex; 2019; 29, pp. 2533-2551. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/29878084][DOI: https://dx.doi.org/10.1093/cercor/bhy123]

35. Gordon, EM; Laumann, TO; Adeyemo, B; Petersen, SE. Individual variability of the system-level organization of the human brain. Cereb. Cortex; 2017; 27, pp. 386-399. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/26464473]

36. Gordon, EM et al. Precision functional mapping of individual human brains. Neuron; 2017; 95, pp. 791-807.e7.1:CAS:528:DC%2BC2sXht1CmsLjF [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/28757305][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5576360][DOI: https://dx.doi.org/10.1016/j.neuron.2017.07.011]

37. Gratton, C et al. Functional brain networks are dominated by stable group and individual factors, not cognitive or daily variation. Neuron; 2018; 98, pp. 439-452.e5.1:CAS:528:DC%2BC1cXnvFers70%3D [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/29673485][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5912345][DOI: https://dx.doi.org/10.1016/j.neuron.2018.03.035]

38. Labonte, AK et al. Precision functional mapping to advance developmental psychiatry research. Biol. Psychiatry Glob. Open Sci.; 2024; 4, 100370. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/39309212][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11416589][DOI: https://dx.doi.org/10.1016/j.bpsgos.2024.100370]

39. Cui, Z et al. Individual variation in functional topography of association networks in youth. Neuron; 2020; 106, pp. 340-353.e8.1:CAS:528:DC%2BB3cXjsF2ju7o%3D [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/32078800][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7182484][DOI: https://dx.doi.org/10.1016/j.neuron.2020.01.029]

40. Cui, Z et al. Linking individual differences in personalized functional network topography to psychopathology in youth. Biol. Psychiatry; 2022; 92, pp. 973-983. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/35927072][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10040299][DOI: https://dx.doi.org/10.1016/j.biopsych.2022.05.014]

41. Iraji, A et al. Identifying canonical and replicable multi-scale intrinsic connectivity networks in 100k+ resting-state fMRI datasets. Hum. Brain Mapp.; 2023; 44, pp. 5729-5748.1:STN:280:DC%2BB2sfjsVCmtQ%3D%3D [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/37787573][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10619392][DOI: https://dx.doi.org/10.1002/hbm.26472]

42. Bethlehem, RAI et al. Brain charts for the human lifespan. Nature; 2022; 604, pp. 525-533.1:CAS:528:DC%2BB38XptF2hsLY%3D [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/35388223][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9021021][DOI: https://dx.doi.org/10.1038/s41586-022-04554-y]

43. Glass, HC et al. Outcomes for extremely premature infants. Anesth. Analg.; 2015; 120, 1337. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/25988638][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4438860][DOI: https://dx.doi.org/10.1213/ANE.0000000000000705]

44. Valente, M., Shultz, S., Klin, S. & Jones, A. Measuring and reducing acoustic noise in MRI studies of infants: a review of existing guidelines and development of new methods. In 2014 International Meeting for Autism Research.https://www.researchgate.net/publication/268143739_Measuring_and_Reducing_Acoustic_Noise_in_MRI_Studies_of_Infants_A_Review_of_Existing_Guidelines_and_Development_of_New_Methods (2014).

45. Chen, L et al. A 4D infant brain volumetric atlas based on the UNC/UMN baby connectome project (BCP) cohort. Neuroimage; 2022; 253, 119097. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/35301130][DOI: https://dx.doi.org/10.1016/j.neuroimage.2022.119097]

46. Fu, Z; Liu, J; Salman, MS; Sui, J; Calhoun, VD. Functional connectivity uniqueness and variability? Linkages with cognitive and psychiatric problems in children. Nat. Ment. Health; 2023; 1, pp. 956-970. [DOI: https://dx.doi.org/10.1038/s44220-023-00151-8] 2023 1:12

47. Kang, H. The prevention and handling of the missing data. Korean J. Anesthesiol.; 2013; 64, 402. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/23741561][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3668100][DOI: https://dx.doi.org/10.4097/kjae.2013.64.5.402]

48. Peverill, M. et al. Balancing data quality and bias: investigating functional connectivity exclusions in the Adolescent Brain Cognitive Development^SM (ABCD Study) across quality control pathways. Hum. Brain Mapp.46, e70094 (2025).

49. Calhoun, VD; Liu, J; Adali, T. A review of group ICA for fMRI data and ICA for joint inference of imaging, genetic, and ERP data. Neuroimage; 2009; 45, S163. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/19059344][DOI: https://dx.doi.org/10.1016/j.neuroimage.2008.10.057]

50. Calhoun, VD; Adali, T; Pearlson, GD; Pekar, JJ. A method for making group inferences from functional MRI data using independent component analysis. Hum. Brain Mapp.; 2001; 14, pp. 140-151.1:STN:280:DC%2BD3MrhtFyrsQ%3D%3D [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/11559959][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6871952][DOI: https://dx.doi.org/10.1002/hbm.1048]

51. Iraji, A et al. Tools of the trade: estimating time-varying connectivity patterns from fMRI data. Soc. Cogn. Affect Neurosci.; 2021; 16, pp. 849-874. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/32785604][DOI: https://dx.doi.org/10.1093/scan/nsaa114]

52. Calhoun, VD; Adali, T. Multisubject independent component analysis of fMRI: a decade of intrinsic networks, default mode, and neurodiagnostic discovery. IEEE Rev. Biomed. Eng.; 2012; 5, pp. 60-73. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/23231989][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4433055][DOI: https://dx.doi.org/10.1109/RBME.2012.2211076]

53. Rachakonda, S., Silva, R. F., Liu, J. & Calhoun, V. D. Memory efficient PCA methods for large group ICA. Front. Neurosci.10, 17 (2016).

54. Erhardt, EB et al. Comparison of multi-subject ICA methods for analysis of fMRI data. Hum. Brain Mapp.; 2011; 32, pp. 2075-2095. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/21162045][DOI: https://dx.doi.org/10.1002/hbm.21170]

55. Iraji, A et al. The connectivity domain: analyzing resting state fMRI data using feature-based data-driven and model-based methods. Neuroimage; 2016; 134, pp. 494-507. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/27079528][DOI: https://dx.doi.org/10.1016/j.neuroimage.2016.04.006]

56. Ray, KL et al. ICA model order selection of task co-activation networks. Front. Neurosci.; 2013; 7, 51440. [DOI: https://dx.doi.org/10.3389/fnins.2013.00237]

57. Bell, AJ; Sejnowski, TJ. An information-maximization approach to blind separation and blind deconvolution. Neural Comput.; 1995; 7, pp. 1129-1159.1:STN:280:DyaK28%2Fkt1eqtg%3D%3D [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/7584893][DOI: https://dx.doi.org/10.1162/neco.1995.7.6.1129]

58. Himberg, J; Hyvärinen, A; Esposito, F. Validating the independent components of neuroimaging time series via clustering and visualization. Neuroimage; 2004; 22, pp. 1214-1222. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/15219593][DOI: https://dx.doi.org/10.1016/j.neuroimage.2004.03.027]

59. Rubinov, M; Sporns, O. Complex network measures of brain connectivity: uses and interpretations. Neuroimage; 2010; 52, pp. 1059-1069. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/19819337][DOI: https://dx.doi.org/10.1016/j.neuroimage.2009.10.003]

60. Mikl, M et al. Effects of spatial smoothing on fMRI group inferences. Magn. Reson. Imaging; 2008; 26, pp. 490-503. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/18060720][DOI: https://dx.doi.org/10.1016/j.mri.2007.08.006]

61. Candemir, C. Spatial smoothing effect on group-level functional connectivity during resting and task-based fMRI. Sensors23, 5866 (2023).

62. Pajula, J; Tohka, J. Effects of spatial smoothing on inter-subject correlation based analysis of FMRI. Magn. Reson. Imaging; 2014; 32, pp. 1114-1124. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/24970023][DOI: https://dx.doi.org/10.1016/j.mri.2014.06.001]

63. Chen, Z; Calhoun, V. Effect of spatial smoothing on task fMRI ICA and functional connectivity. Front. Neurosci.; 2018; 12, 327739. [DOI: https://dx.doi.org/10.3389/fnins.2018.00015]

64. Fesl, G et al. Is the center of mass (COM) a reliable parameter for the localization of brain function in fMRI?. Eur. Radio.; 2008; 18, pp. 1031-1037.1:STN:280:DC%2BD1c3jvFajtw%3D%3D [DOI: https://dx.doi.org/10.1007/s00330-008-0850-z]

65. Hastie, T; Tibshirani, R. Generalized additive models. Stat. Sci.; 1986; 1, pp. 297-318.

66. Damaraju, E et al. Functional connectivity in the developing brain: a longitudinal study from 4 to 9 months of age. Neuroimage; 2014; 84, pp. 169-180.1:STN:280:DC%2BC3sbkt1Gqtg%3D%3D [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/23994454][DOI: https://dx.doi.org/10.1016/j.neuroimage.2013.08.038]

67. Xu, J; Potenza, MN; Calhoun, VD. Spatial ICA reveals functional activity hidden from traditional fMRI GLM-based analyses. Front. Neurosci.; 2013; 7, 61584. [DOI: https://dx.doi.org/10.3389/fnins.2013.00154]

68. Fitzgibbon, SP et al. The developing Human Connectome Project (dHCP) automated resting-state functional processing framework for newborn infants. Neuroimage; 2020; 223, 117303. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/32866666][DOI: https://dx.doi.org/10.1016/j.neuroimage.2020.117303]

69. Sydnor, VJ et al. Neurodevelopment of the association cortices: patterns, mechanisms, and implications for psychopathology. Neuron; 2021; 109, pp. 2820-2846.1:CAS:528:DC%2BB3MXhsFOktr7F [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/34270921][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8448958][DOI: https://dx.doi.org/10.1016/j.neuron.2021.06.016]

70. Mesulam, MM. From sensation to cognition. Brain; 1998; 121, pp. 1013-1052. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/9648540][DOI: https://dx.doi.org/10.1093/brain/121.6.1013]

71. Kriegeskorte, N; Cusack, R; Bandettini, P. How does an fMRI voxel sample the neuronal activity pattern: compact kernel or complex spatiotemporal filter?. Neuroimage; 2010; 49, 1965. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/19800408][DOI: https://dx.doi.org/10.1016/j.neuroimage.2009.09.059]

72. Tierney, AL; Charles, A; Nelson, I. Brain development and the role of experience in the early years. Zero Three; 2009; 30, 9. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/23894221][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3722610]

73. Ellis, CT; Skalaban, LJ; Yates, TS; Turk-Browne, NB. Attention recruits frontal cortex in human infants. Proc. Natl. Acad. Sci. USA; 2021; 118, e2021474118.1:CAS:528:DC%2BB3MXnt1elsbc%3D [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/33727420][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7999871][DOI: https://dx.doi.org/10.1073/pnas.2021474118]

74. Jung, JY; Lambon Ralph, MA; Jackson, RL. Subregions of DLPFC display graded yet distinct structural and functional connectivity. J. Neurosci.; 2022; 42, pp. 3241-3252.1:CAS:528:DC%2BB38Xht1OktrbO [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/35232759][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8994544][DOI: https://dx.doi.org/10.1523/JNEUROSCI.1216-21.2022]

75. Plakke, B; Romanski, LM. Auditory connections and functions of prefrontal cortex. Front. Neurosci.; 2014; 8, 92988. [DOI: https://dx.doi.org/10.3389/fnins.2014.00199]

76. Tau, GZ; Peterson, BS. Normal development of brain circuits. Neuropsychopharmacology; 2010; 35, pp. 147-168. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/19794405][DOI: https://dx.doi.org/10.1038/npp.2009.115]

77. Wen, X et al. Development of dynamic functional architecture during early infancy. Cereb. Cortex; 2020; 30, pp. 5626-5638. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/32537641][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7898958][DOI: https://dx.doi.org/10.1093/cercor/bhaa128]

78. Posner, M. I. & Petersen, S. E. The attention system of the human brain. Annu. Rev. Neurosci.13, 25–42 (1990).

79. Johnson, M. H. & de Haan, M. Developmental Cognitive Neurosciences 4th edn (Wiley-Blackwell (John Wiley & Sons, Ltd.), 2015).

80. Yu, X et al. Functional connectivity in infancy and toddlerhood predicts long-term language and preliteracy outcomes. Cereb. Cortex; 2021; 32, 725. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/34347052][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10847903][DOI: https://dx.doi.org/10.1093/cercor/bhab230]

81. Wen, X et al. First-year development of modules and hubs in infant brain functional networks. Neuroimage; 2019; 185, 222. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/30315911][DOI: https://dx.doi.org/10.1016/j.neuroimage.2018.10.019]

82. Alcauter, S et al. Development of thalamocortical connectivity during infancy and its cognitive correlations. J. Neurosci.; 2014; 34, pp. 9067-9075.1:CAS:528:DC%2BC2cXhvV2ksrzN [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/24990927][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4078084][DOI: https://dx.doi.org/10.1523/JNEUROSCI.0796-14.2014]

83. Yu, Q et al. Infant brain regional cerebral blood flow increases supporting emergence of the default-mode network. Elife; 2023; 12, e78397.1:CAS:528:DC%2BB2cXjs1Kmt7k%3D [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/36693116][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9873253][DOI: https://dx.doi.org/10.7554/eLife.78397]

84. Hadders-Algra, M. Early human brain development: starring the subplate. Neurosci. Biobehav. Rev.; 2018; 92, pp. 276-290. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/29935204][DOI: https://dx.doi.org/10.1016/j.neubiorev.2018.06.017]

85. Xu, Y et al. Development and emergence of individual variability in the functional connectivity architecture of the preterm human brain. Cereb. Cortex; 2019; 29, pp. 4208-4222. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/30534949][DOI: https://dx.doi.org/10.1093/cercor/bhy302]

86. Moore, LA et al. Towards personalized precision functional mapping in infancy. Imaging Neurosci.; 2024; 2, pp. 1-20. [DOI: https://dx.doi.org/10.1162/imag_a_00165]

87. Yin, W. et al. Charting brain functional development from birth to 6 years of age. Nat. Hum. Behav.9, 1246–1259 (2025).

88. Lin, W et al. Functional connectivity MR imaging reveals cortical functional connectivity in the developing brain. Am. J. Neuroradiol.; 2008; 29, 1883.1:STN:280:DC%2BD1cjjs1yqug%3D%3D [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/18784212][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2583167][DOI: https://dx.doi.org/10.3174/ajnr.A1256]

89. Graven, SN; Browne, JV. Auditory development in the fetus and infant. Newborn Infant Nurs. Rev.; 2008; 8, pp. 187-193. [DOI: https://dx.doi.org/10.1053/j.nainr.2008.10.010]

90. Oldham, S. & Fornito, A. The development of brain network hubs. Dev. Cogn. Neurosci.36, 100607 (2019).

91. Mongerson, CRL; Jennings, RW; Borsook, D; Becerra, L; Bajic, D. Resting-state functional connectivity in the infant brain: methods, pitfalls, and potentiality. Front. Pediatr.; 2017; 5, 14. [DOI: https://dx.doi.org/10.3389/fped.2017.00159]

92. Hodel, AS. Rapid infant prefrontal cortex development and sensitivity to early environmental experience. Dev. Rev.; 2018; 48, 113. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/30270962][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6157748][DOI: https://dx.doi.org/10.1016/j.dr.2018.02.003]

93. Grotheer, M et al. White matter myelination during early infancy is linked to spatial gradients and myelin content at birth. Nat. Commun.; 2022; 13, pp. 1-12. [DOI: https://dx.doi.org/10.1038/s41467-022-28326-4]

94. Cao, M; Huang, H; He, Y. Developmental connectomics from infancy through early childhood. Trends Neurosci.; 2017; 40, pp. 494-506.1:CAS:528:DC%2BC2sXhtVGnsLrN [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/28684174][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5975640][DOI: https://dx.doi.org/10.1016/j.tins.2017.06.003]

95. Goksan, S et al. fMRI reveals neural activity overlap between adult and infant pain. Elife; 2015; 2015, pp. 1-13.

96. Wenger, E; Brozzoli, C; Lindenberger, U; Lövdén, M. Expansion & renormalization of human brain structure during skill acquisition. Trends Cogn. Sci.; 2017; 21, 930. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/29149999][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5697733][DOI: https://dx.doi.org/10.1016/j.tics.2017.09.008]

97. Tooley, UA; Bassett, DS; Mackey, AP. Environmental influences on the pace of brain development. Nat. Rev. Neurosci.; 2021; 22, 372.1:CAS:528:DC%2BB3MXhtVSrur7L [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/33911229][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8081006][DOI: https://dx.doi.org/10.1038/s41583-021-00457-5]

98. Knickmeyer, RC et al. A structural MRI study of human brain development from birth to 2 years. J. Neurosci.; 2008; 28, pp. 12176-12182.1:CAS:528:DC%2BD1cXhsVGmtLvN [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/19020011][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2884385][DOI: https://dx.doi.org/10.1523/JNEUROSCI.3479-08.2008]

99. Gao, W; Lin, W; Grewen, K; Gilmore, JH. Functional connectivity of the infant human brain: plastic and modifiable. Neuroscientist; 2017; 23, 169. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/26929236][DOI: https://dx.doi.org/10.1177/1073858416635986]

100. Butts, T; Green, MJ; Wingate, RJT. Development of the cerebellum: simple steps to make a ‘little brain’. Development; 2014; 141, pp. 4031-4041.1:CAS:528:DC%2BC2cXitFWntrnK [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/25336734][DOI: https://dx.doi.org/10.1242/dev.106559]

101. Fossella, JA; Sommer, T; Fan, J; Pfaff, D; Posner, MI. Synaptogenesis and heritable aspects of executive attention. Ment. Retard Dev. Disabil. Res. Rev.; 2003; 9, pp. 178-183. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/12953297][DOI: https://dx.doi.org/10.1002/mrdd.10078]

102. Muñoz-Moreno, E et al. Structural brain network reorganization and social cognition related to adverse perinatal condition from infancy to early adolescence. Front. Neurosci.; 2016; 10, 208340. [DOI: https://dx.doi.org/10.3389/fnins.2016.00560]

103. Gao, W et al. Temporal and spatial evolution of brain network topology during the first two years of life. PLoS ONE; 2011; 6, e25278.1:CAS:528:DC%2BC3MXhtlajs7jF [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/21966479][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179501][DOI: https://dx.doi.org/10.1371/journal.pone.0025278]

104. Graham, AM et al. The potential of infant fMRI research and the study of early life stress as a promising exemplar. Dev. Cogn. Neurosci.; 2015; 12, pp. 12-39. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/25459874][DOI: https://dx.doi.org/10.1016/j.dcn.2014.09.005]

105. Cui, W., Wang, S., Chen, B. & Fan, G. Altered functional network in infants with profound bilateral congenital sensorineural hearing loss: a graph theory analysis. Front. Neurosci.15, 810833 (2022).

106. Perinelli, A; Tabarelli, D; Miniussi, C; Ricci, L. Dependence of connectivity on geometric distance in brain networks. Sci. Rep.; 2019; 9, pp. 1-9.1:CAS:528:DC%2BC1MXhvVSku7fP [DOI: https://dx.doi.org/10.1038/s41598-019-50106-2]

107. Mitra, A et al. Resting-state fMRI in sleeping infants more closely resembles adult sleep than adult wakefulness. PLoS ONE; 2017; 12, e0188122. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/29149191][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5693436][DOI: https://dx.doi.org/10.1371/journal.pone.0188122]

108. Korom, M et al. Dear reviewers: Responses to common reviewer critiques about infant neuroimaging studies. Dev. Cogn. Neurosci.; 2022; 53, 101055. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/34974250][DOI: https://dx.doi.org/10.1016/j.dcn.2021.101055]

109. Du, Y. et al. NeuroMark: An Automated and adaptive ICA based pipeline to identify reproducible fMRI markers of brain disorders. Neuroimage Clin.28, 102375 (2020).

110. Labache, L; Ge, T; Yeo, BTT; Holmes, AJ. Language network lateralization is reflected throughout the macroscale functional organization of cortex. Nat. Commun.; 2023; 14, pp. 1-13. [DOI: https://dx.doi.org/10.1038/s41467-023-39131-y]

111. Agcaoglu, O; Miller, R; Mayer, AR; Hugdahl, K; Calhoun, VD. Lateralization of resting state networks and relationship to age and gender. Neuroimage; 2015; 104, pp. 310-325.1:STN:280:DC%2BC2M7jt1Okug%3D%3D [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/25241084][DOI: https://dx.doi.org/10.1016/j.neuroimage.2014.09.001]