MRI

MEG and EEG

MEG and EEG are powerful tools for studying brain development given their capacity to characterize brain function based on neural activity in both time and space and provide electrophysiological information complementary to structural and functional MRI. The hardware, acquisition schemes, and processing of MEG have been specifically adapted to the infant population. Specialized MEG systems (e.g., BabySQUID, Artemis 123, BabyMEG, SARA) for fetus, neonates, and infants have been developed to measure age-specific neural activities (Chen et al., 2019; Roberts et al., 2014). The headsets of the infant MEG systems are specifically tailored for infants. Coil-in-vacuum configurations have been employed in infant-dedicated MEG systems that reduce the gap between detection coils and the helmet surface to as close as 6 mm (Chen et al., 2019). Moreover, during acquisition, strategies like slightly depriving the infant of sleep, feeding, or playing relaxing music in the scanner room can help the infant stay asleep during the scan. Significant progress in EEG, such as 256-channel high-density EEG system, has made it possible to bypass behavioral limitation and identify neural codes for speech in prebabbling infants (Gennari et al., 2021). Elaborated infant MEG and EEG can be found in the literature (Huang & Roberts, 2021, Section 3).

Other Imaging Modalities

NIRS, a noninvasive optical technique, allows quantitative measures of cortical hemoglobin oxygenation (namely oxyhemoglobin and deoxyhemoglobin) and delineates the physiological and metabolic measurement of infant brain. NIRS complements functional measurement from fMRI, EEG, and MEG and physiological measures from perfusion MRI (Franceschini et al., 2007; Lloyd-Fox et al., 2010). NIRS is well suited to infants given their thin skull and scalp. NIRS has moved from point measurements to an array of detectors similar to EEG and MEG systems, providing more holistic, whole-brain system level measurements as well as point measurements (Lloyd-Fox et al., 2010).

CT, PET, and SPECT use ionizing radiation. CT has conventionally been used to measure macroscopic morphology of the infant brain or evaluate head injuries (Barkovich, 1992). PET and SPECT have been used to study the metabolism of the infant brain (Chugani & Phelps, 1986; Chugani et al., 1987). Due to the harmful ionizing radiation induced, PET and SPECT studies have focused on animal models. PET and SPECT are able to assess metabolism under infant brain injury or oxygen deprivation conditions and are still used in clinical studies. CT, PET, and SPECT can offer complementary information to MRI and NIRS and have certain clinical value for assessing head and brain injuries.

Denoising

Like any signal we collect from nature, medical images are corrupted by noise; that is, unwanted disturbances in the image signal that commonly arise from microscopic thermal motion and electric fluctuations. Noise in MRI measurement mainly arises from thermal noise, electronic noise, and dielectric and inductive losses in the subject. SNR is the quantitative way to evaluate noise in an image with respect to the signal. The process of identifying and decreasing or removing noise from images to increase SNR is called denoising. This process is usually included in postprocessing of infant brain images due to low SNR associated with smaller voxel size compatible with the small size of the infant brain. There are three main ways to increase the SNR: increasing the signal without increasing noise during acquisition, decreasing the noise without decreasing the signal during acquisition, and reducing noise in the postprocessing. For instance, use of infant-dedicated coils can increase SNR by increasing the signal amplitude during acquisition because infant-dedicated coils are closer to the infant’s brain and the signal increases with decreasing distance. Controlling the room temperature and the use of less noisy electronics can decrease the noise without decreasing the signal during MRI acquisition. Signals from nearby voxels or signals from similar gradient orientations (for diffusion MRI) are highly correlated and can offer complementary information to each other to remove noise effects (Huang & Roberts, 2021, Section 1) . After data acquisition, methods incorporating random matrix theory and error propagation can be used to reduce noise by leveraging images’ spatial, temporal, and gradient orientation redundancies.

Correction of Distortions and Motion Artifacts

Besides denoising, distortions and artifacts of medical images should also be reduced before further processing. Noise can obscure or blur the images without distorting them, while distortions and artifacts are typically systematic errors that can distort the images. In MRI, image signal arises from spins (microscopic magnetic moments) of molecules that are aligned to a main magnetic field called B₀. The spatial coordinates that track where the signal comes from are encoded by linear magnetic field gradients. In ideal situations, the B₀ field should be homogeneous across the subject’s head and the spatial gradients should be strictly linear. However, in a real-world situation, it is difficult to produce a completely constant magnetic field. Distortions in MR images, particularly those acquired with fast sequences such as EPI, therefore arise from mainly three origins: B₀ field inhomogeneity, gradient nonlinearity, and eddy currents. Moreover, motion artifacts commonly seen in infant imaging arise from subject motion during scanning. Brain motion can usually be described by a rigid transformation consisting of rotation and translation. However, due to the nonlinear nature of distortions, rigid motion in the subject can appear as nonrigid distortions in the images. To mitigate motion artifacts specific to the infant population, advanced motion-tracking hardware and software including optical navigators and respiratory or cardiac gating devices can be applied during the scan (see the section “MRI: Motion Artifacts and Correction in Acquisition”). Combining motion tracking information with measured gradient nonlinearity, B₀ field maps and estimated eddy currents can maximally correct for distortions and motion artifacts in MRI (Dubois, 2021; Dubois et al., 2021; Huang & Roberts, 2021, Section 1). Besides adoption of distortion correction methods widely used for adults, like obtaining a field map characterizing B₀ inhomogeneity, techniques have been developed for data acquired without the field map (e.g., Andersson & Skare, 2002; Ardekani & Sinha, 2005; Huang et al., 2008, Kybic et al., 2000). Two acquisitions with opposing polarities of phase-encoding blips are widely used for correcting distortions induced by susceptibility. In two such acquisitions, distortion going in opposing directions can be utilized to estimate the field by finding the field that will maximize the similarity of the unwarped volumes when applied to the two volumes (Andersson et al., 2002; Smith et al., 2004).

Segmentation, Parcellation, and Registration

Segmentation and parcellation are essential analysis procedures before infant brain images can be used for exploring infant brain structural, functional, and physiological information carried by these images. By extracting well-defined and standardized quantities from different subject groups and age ranges, rich information from infant brain images can be organized into structured measurements for comparison. Such standardization is usually achieved by registering images to a common, well-defined template image space. Rapidly changing morphology and image contrast of infant brain structures necessitate registration to age-specific templates. In very preterm neonate brains, such as those around 30 postmenstrual weeks, some transient neural structures,such as ganglionic eminence (Feng et al., 2019), still exist but will disappear in a few weeks. Age-specific infant and neonate brain atlases (Feng et al., 2019; Oishi et al., 2011, 2019) have been established for this purpose.

Infant brain segmentation is the basis for a suite of analyses, such as brain morphometry analysis (e.g., cortical thickness and gyrification) and network analysis (e.g., node identification). The prerequisite of these analyses is that the infant brain is segmented into different tissue types—such as white matter, cortical and subcortical gray matter, and cerebral spinal fluid—and, further, into different brain regions. Manually segmenting infant brain images is highly labor intensive. Many automatic segmentation algorithms have been developed. In general, reliable automatic segmentation of brain tissues requires high contrast images with consistent contrasts across age. However, as shown in Figure 3, traditional T1w and T2w contrasts are poor in younger infant brains due to the ongoing myelination process. Both T1w and T2w contrasts undergo a contrast inversion between white and gray matter contrasts from 0–1 year old, which causes a so-called isointense stage in 3–9 months. Gray and white matter appear the same in this isointense stage. Some automatic segmentation methods focused on incorporating morphometric constraints, such as regional cortical thickness into automatic segmentation algorithms (e.g., Li et al., 2019; Zollei et al., 2020). Contrasts other than T1w or T2w images have also been incorporated. Furthermore, machine learning and deep learning methods have also been applied to the registration and segmentation of baby brain images and have ushered in a new era for segmenting infant brains (e.g., Wang et al., 2019; Zhang et al., 2015). (Details of methods based on machine learning and deep learning can be found in the section “Machine Learning–Based Image Processing.”) The infant brain cerebral cortex can be further parcellated into multiple cortical regions based on structural or functional parcellation. For structural parcellation, structural morphological information, such as sulcal landmarks, is usually used for parcellating individual cortical regions (see, e.g., Wu et al., 2021, for review). Functional parcellation is based on functional connectivity that divides cortical regions into functionally differentiated regions (e.g., Yeo et al., 2011). Multiple infant-specific brain atlases based on structural parcellation are available (see Oishi et al., 2019, for review). For infant brain functional parcellation (e.g., Peng et al., 2020; Shi et al., 2017), it is important to understand that brain regions constituting higher cognitive function systems are still not distinguishable in infant brains, and infant brains are not just smaller version of adult brains. Thus, parcellation schemes tailored for neonatal and infant brains should be established and used for further regional and brain network analysis. For a given specific infant subject of interest, structural or functional parcellation from those atlases can be transferred to this subject of interest by registration.

White matter tractography can be considered segmentation of specific white matter tracts, when traced fibers are presented as voxels instead of streamlines. With unmyelinated white matter fibers in infant brains, especially in neonate brains, age-specific diffusion MRI tracing parameters (e.g., Huang et al., 2006, 2009; Takahashi et al., 2012) different from those used for tracing adult brain white matter fibers should be adopted.

Open in new tab

Figure 3. High-resolution (0.8 x 0.8 x 0.8 mm³) T1 weighted (T1w, top panel) and T2 weighted (T2w, bottom panel) images of the developing infant brain at the ages of 1.6, 3, 6, 9.3, and 14.6 months.

Source: Generated with the Penn-CHOP developmental connectome data from Huang Laboratory.

Network Analysis

Brain networks can be modeled as graphs with nodes indicating parcellated brain regions and edges indicating connections between nodes. In infant brain imaging, edges in diffusion MRI tractography-based graphs represent white matter structural connections between brain regions, while edges in rs-fMRI-based connectome studies represent functional coactivation between brain regions. These brain network nodes are usually obtained by brain segmentation and parcellation as described above (see the section “Segmentation, Parcellation, and Registration”). Graph theory has empowered network analyses investigating the development of structural and functional connections among parcellated brain regions (Cao & He, 2021). Advances in neuroimaging and neurophysiological techniques together with the connectomic models provide unprecedented opportunities to delineate how the human brain develops from a circuitry or network perspective through noninvasively mapping structural and functional connectome patterns (Cao, Huang, & He, 2017). The complex dynamics of brain disorder or injury progression coupled with rapid brain development make accurately predicting developmental outcomes challenging. Graph theory–based network analysis offers a group of metrics that are sensitive enough to characterize typical brain development and atypical development in brain disorders. New insights into the early development of the brain in both healthy and pathological populations have been obtained through network analysis for a better understanding of the origin of complex neural architecture and dynamics as well as mechanisms underlying developmental neuropsychiatric disorders (Cao, Huang, & He, 2017).

Machine Learning–Based Image Processing

Like other digital medical images loaded into a computer, infant brain images can benefit from machine learning–based image processing. Machine-learning technique gathers information about model parameters from data or patterns. One way to categorize machine-learning algorithms is by how the algorithm learns. As mentioned previously, in each step of infant brain image processing—including denoising, registration, and segmentation—image properties that can be perceived by humans have been mathematically modeled and extracted from images. Machines can learn these properties and perform image-processing tasks. Learning parameters with the so-called training data, namely data with paired relationships between inputs and outputs, is called supervised learning; automatically learning new patterns from data is called unsupervised learning (Litjens et al., 2017). Another way to categorize machine-learning methods is by the assumed model structure. Traditional machine-learning methods assume simple models, such as linear regression, support vector machine, and random forests, without using deep neural networks. Deep learning methods assume multilayer neural networks or convolutional neural networks as the model structure (Litjens et al., 2017; Shen et al., 2017). In processing infant brain images, machine-learning algorithms automatically learn important associations between input images and outcomes by using handcrafted features. We can also let machine learning automatically extract important features from the imaging data. In the latter case, such associations and important features usually cannot be intuitively perceived or handcrafted by humans. Automatically learning associations with machine learning has resulted in advanced methods that predict cognitive and clinical outcomes from imaging data and identify biomarkers extracted from infant brain images (Ouyang et al., 2020; Pagnozzi et al., 2018; Smyser et al., 2016). Using machine learning to automatically extract important features has resulted in machine- and deep-learning–based infant brain segmentation algorithms that greatly boost segmentation accuracies (Li et al., 2019). Developing advanced machine learning and deep-learning techniques to improve denoising and registration for infant brain imaging data still remains an active area of research and often requires computationally demanding simulated ground-truth imaging data with known motion and distortions.

Structural Development of the Infant Brain

Infant brain structural development can be categorized into macrostructural and microstructural development. The former is focused on volume, thickness, and morphology of a specific brain structure or the entire brain (including cortical folding). The latter is focused on microstructures of specific brain structures. The entire brain size as well as gray and white matter volume can be precisely measured with several minutes of anatomical MRI, such as T1w and T2w images. This data shows that brain size increases dramatically during infancy (Figure 3). The brain size reaches 80%–90% of adult volume by 2 years of age (Pfefferbaum et al., 1994), with rapid elaboration of new synapses happening in the first 2 years (Huttenlocher & Dabholkar, 1997). Overall gray matter volume increases to a lifetime maximum around 2 years of (Knickmeyer et al., 2008; Matsuzawa et al., 2001). In addition to volume and cortical thickness, cortical sulcal pattern has been found to be sensitive to brain development and function (Im & Grant, 2019). Tracing white matter tract pathways through DTI-based tractography and measuring white matter tract microstructure through DTI-based metrics have also been applied to study infant brain structural development. Pioneering studies on fetal and infant white matter development (Huang et al., 2006, 2009, 2010; Takahashi et al., 2012) noninvasively and comprehensively delineated major white matter tracts from the fetal stage to early childhood using DTI. It was found that white matter tract groups, including limbic, commissural, association, projection, and brain stem tracts, develop heterogeneously. More primitive brainstem white matter tracts emerge first, followed by projection, limbic, commissural, and association tracts (Figure 4). White matter tract groups mature in the order of tracts associated with more primitive functions, such as breathing (e.g., brainstem tracts), to tracts associated with high-order cognitive functions, such as memory and perception (e.g., association tracts). Measured with DTI-based metrics like FA, infant white matter tract microstructures change significantly with myelination, axonal packing, and axonal growth (e.g., Dubois et al., 2006, 2008, 2014; Kunz et al., 2014; Mishra et al., 2013) and follow the same maturational pattern as that found with DTI tractography. If we zoom out the age period 0–8 years, exponential maturational curve best describes the white matter microstructural (e.g., FA) development with the largest growth rate of white matter FA during the first 2 years of life (Yu et al., 2020). Besides white matter microstructure, the fetal to perinatal period is a unique time window in which DTI-based FA measurement is sensitive to microstructural changes in the cerebral cortex due to disruption of radial glia by dendritic arborization (e.g., Ball et al., 2013; Huang et al., 2013; Huang & Vasung, 2014; McKinstry et al., 2002; Ouyang, Jeon, et al., 2019; Yu et al., 2016). This cortical microstructure measured by FA encodes the “footage” of regional cellular and molecular processes, which can serve as a potential sensitive biomarker in predicting future neurodevelopmental outcomes and identifying individual risks of brain disorders (Ouyang et al., 2020). Sophisticated diffusion MRI models, such as diffusion kurtosis imaging (DKI), have made it possible to delineate the complex and less organized microstructure of the cerebral cortex that cannot be captured by DTI. It has been found that DKI-derived mean kurtosis measurements in the cerebral cortex quantifying cortical microstructural complexity may be indicative of important cellular processes, such as changes of neuronal density, dendritic arborization, synaptic formation, and myelination of intracortical axons in the cerebral cortex (Ouyang, Jeon, et al., 2019; Zhu et al., 2021). Cortical and white matter myelination can also be quantified with measurement of the myelin water fraction based on T1 and T2 relaxometry. A strong increase of myelin water fraction is observed during infancy and follows a nonlinear growth pattern until early childhood in both the cerebral cortex (Deoni et al., 2015) and white matter (Dean, O’Muircheartaigh, et al., 2014; Dean et al., 2016). By adopting both diffusion MRI–based tractography of white matter tracts and diffusion MRI–based white matter microstructural measurement in defining and quantifying graph theory–based network edges, structural connectomic analysis reveals system-level network changes during infancy. Important structural network metrics, such as strength and efficiency, have been found to increase during fetal, neonate, and infant brain development (e.g., Huang et al., 2015; Ouyang, Kang, et al., 2017; Song et al., 2017; Yap et al., 2011; Zhao, Mishra, et al., 2019; Zhao, Xu, et al. 2019). These findings on infant brain structural development set the stage for understanding aberrant brain development in disorders and offer potential for early fetal and infant interventions to avert long-term diseases (Ouyang, Dubois, et al., 2019; Vasung et al., 2019).

Open in new tab

Figure 4. Diffusion MRI tractography of the white matter tracts in developing fetal and infant brain from 16 postmenstrual weeks (pmw) to 2 years. (A) Brainstem tracts including inferior, middle, superior cerebellar peduncle (ICP, MCP, and SCP), and medial lemniscus (ML); (B) projection tracts including internal capsule (IC), corona radiata (CR), corticospinal tract (CST), anterior, superior, and posterior thalamic radiations (ATR, STR and PTR); (C) limbic tracts including cingulum bundle in the cingulate cortex, cingulum bundle projecting to hippocampus (CGH), and fornix (FX); (D) commissural tracts including body, genu, and splenium of corpus callosum (BCC, GCC and SCC); (E) association tracts including external capsule (EC), inferior longitudinal fasciculus (ILF), inferior fronto-occipital fasciculus (IFO), superior longitudinal fasciculus (SLF), and uncinate fasciculus (UNC). Ganglionic eminence (GE), a transient fetal brain structure and well traced with DTI tractography, is also included in this tract group. (F) general timeline of white matter maturation across different tracts and tract groups. Dotted lines indicate that white matter tracts emerge at these ages, though to a relatively minor degree, and arrows indicate that overall white matter tracts are formed with continuous maturational processes, such as myelination and axonal packing thereafter.

Source: Generated with the data from Huang lab.

Functional Development of the Infant Brain

Infant imaging techniques have become a powerful tool to understand functional development and early onset of neuropsychiatric disorder during infancy. Functional imaging studies can be roughly categorized into task-based and resting-state studies. MEG, EEG, and NIRS have been used to investigate the infants’ responses to visual, somatosensory, and auditory responses. “Task-based” BOLD fMRI (see Ellis & Turk-Browne, 2018; Dubois et al., 2021, for review), or fMRI in a stimulus paradigm, has played an important role in understanding the functional development of the infant brain. Practically since infants cannot be instructed to stay still, considerable artifacts result from head motion. Specific MR-related parameters, such as longer T₂* in infant versus adult brains, also need to be considered for infant fMRI sequence design. Despite the challenges of infant fMRI in a stimulus paradigm (Ellis et al., 2020), significant progress has been made in understanding the development of infant sensorimotor (e.g., Dall’Orso et al., 2018), visual (e.g., Biagi et al., 2015), and auditory and language (e.g., Dehaene-Lambertz et al., 2002, 2006; Perani et al., 2010) systems. Besides task-based fMRI, resting-state fMRI has become an effective research tool for infants. Based on resting-state fMRI, coactivation of BOLD signals across brain regions defines functional connectivity and by extension functional connectome. Graph theory analysis can be performed to understand infant functional connectome (Cao, Huang, & He, 2017). At a population level, in infant brains, functional connectome exhibits increasingly efficient information segregation and integration characterized by small-worldness (e.g., Cao, He et al., 2017; Fransson et al., 2011). Global integration of information benefits greatly from increased heterogeneity and hierarchy. The development of functional systems follows the order from primary sensory to higher-order cognitive, with primary functional systems, such as sensorimotor and auditory, already established in neonates (Cao, He, et al., 2017; Doria et al., 2010; Fransson et al., 2011; Smyser et al., 2010, 2011) and higher-order functional systems (e.g., default mode network) gradually emerging after birth (Gao et al., 2009; Yu et al., 2021). At individual levels, pediatric functional networks exhibit remarkable individual variability similar to adults. Growing evidence shows that individual connectomic differences in childhood are correlated to behavior and cognitive capacity in later life (e.g., Alcauter et al., 2014; Graham et al., 2016). Dynamic connectomic growth leads to both high neural plasticity and vulnerability to risk factors in pediatric brains. In general, despite methodological challenges, such as the need for age-specific templates/atlases and increased head motion during the scan (Cao, Huang et al., 2017), the functional connectomic framework provides unique opportunities to comprehensively map infant brain development on population and individual levels. Emerging data also indicate that functional connectivity is sensitive to the onset of major neuropsychiatric disorders, such as autism (Di Martino et al., 2014; Gilmore et al., 2018) and brain injury (Smyser et al., 2019).

Physiological Development of the Infant Brain

Blood perfusion is an essential property of brain physiology, underlying structural, functional, and behavioral development in infancy. Age-related dramatic global CBF increases, with global CBF at 18 months almost 5 times of the CBF around birth, have been reported in the literature using phase-contrast MRI (Liu et al., 2019). Infant rCBF has been conventionally measured with PET (Chugani et al., 1987; Chugani & Phelps, 1986) and SPECT (Chiron et al., 1992), which are not applicable to normal infants given radiation concerns. Given that ASL perfusion MRI offers reliable and comparable infant rCBF measures across platforms, ASL perfusion MRI has become widely used. Regional brain metabolism, including glucose utilization and oxygen consumption, is closely related to local cerebral blood flow that delivers the glucose and oxygen needed to sustain metabolic needs (Raichle et al., 2001; Vaishnavi et al., 2010). A lifespan daily energy expenditure study showed the fastest increase of energy expenditure during infancy across the lifespan (Pontzer et al., 2021). A dramatic increase of energy expenditure occurs in conjunction with brain gray matter volumetric growth. Measured by ASL perfusion MRI, rCBF increases quickly yet heterogeneously across brain regions to meet differential metabolic needs in rapid brain maturational processes during infancy (Yu et al., 2021). Preliminary results on the physiological development of the preterm neonate brain also showed a heterogeneous increase of rCBF across brain regions during preterm development (Ouyang, Liu et al., 2017). In addition to studying physiological development of infants, ASL perfusion MRI has also been used for measuring altered rCBF in infant neurological disorders, such as infant brain injury and ischemic stroke, and for studying its relationship to neurodevelopmental outcomes. For example, a pulsed ASL perfusion MRI study found that CBF predicts language and motor outcomes in neonatal hypoxic-ischemic encephalopathy (Zheng et al., 2020).

Integration of Multimodal Imaging and Its Application to the Infant Brain

As described in the overview, various imaging modalities—including functional, diffusion, structural and perfusion MRI, EEG, MEG, NIRS, PET, and CT—offer unique yet complementary information about the infant brain’s structure, function, and physiology, as well as electrical and magnetic properties of brain circuits. The state-of-the-art analytic approaches on motion correction, brain atlases, computational neuroanatomy, morphology, microstructure, connectivity, and graph theory–based network connectomic analysis have further empowered the ability to generate new insights into the complicated processes in typical infant brain development and infant brain disorders (Huang et al., 2021). For example, for understanding typical infant brain development, unique brain white matter microstructures measured by diffusion MRI can be coupled with functional measurements obtained with EEG. Specifically, an increase in white matter myelination measured by diffusion MRI decreases the visual, somatosensory, and auditory response latencies measured by EEG, as myelination significantly increases the conduction velocity of neural signals (Dubois et al., 2016). Measurements of similar properties from different functional imaging modalities can be used to cross-validate each other and probably offer more robust findings. For example, both fMRI and EEG results suggest that the main components of the adult voice-processing networks are present early and that preterm infants at term-equivalent age have more enhanced processing for voices than full-term newborns (Adam-Darque et al., 2020). Like various applications of multimodal imaging to understand typical infant brain development, multimodal brain imaging has also proven instrumental in investigating brain disorders. For example, a hybrid of diffuse correlation spectroscopy (DCS) and NIRS has been used for measuring changes of oxyhemoglobin, deoxyhemoglobin, and total hemoglobin concentrations as well as rCBF and oxygen metabolism during hypercapnia for neonates with congenital heart disease (Durduran et al., 2010). Multimodal imaging of NIRS-DCS and ASL perfusion MRI demonstrated that NIRS-DCS measurements had good correlation with rCBF measurements from ASL perfusion MRI. With convenience of bedside implementation of NIRS-DCS and through multimodal imaging demonstrating its good correlation with ASL perfusion MRI, this NIRS-DCS method can be potentially used clinically for monitoring critically ill neonates. Another example is adopting multimodal imaging, including MEG and diffusion MRI, to study the auditory systems of children older than infants with autism spectrum disorder (e.g., Berman et al., 2016; Roberts et al., 2009). These studies enabled the finding of auditory response delay (measured from MEG) in children with autism spectrum disorder and white matter microstructural changes (measured from diffusion MRI) likely contributing to the auditory response delay. More studies integrating MRI, EEG, MEG, optical imaging, and other imaging modalities will emerge and harness the multimodal strength for obtaining complementary structural, functional, and physiological information from typically and atypically developmental infant brains.