Results from the limited number of fcMRI studies investigating preterm and term infants reflect the complex interplay of evolving structural and functional architecture during early development (). These studies have typically examined RSN development from term age through the first two years of life, though a subset of studies have also included preterm infants. The majority of these studies included healthy subjects born at term, seeking to define normal neural network development in this population. Five studies have also explored longitudinal development. RSNs involving cortical and subcortical structures have been identified as early as 26 weeks PMA. Recognized networks demonstrate a pattern of gradual maturation and dynamic configuration which correlates with subject age. While demonstrating the applicability of the technique in this population, these studies utilized varied acquisition and analysis techniques, at times exhibiting noted variability in results. As expanding numbers of fcMRI investigations of infants further define patterns of normative neural network development, findings can be increasingly correlated with those obtained from complementary modalities, providing greater understanding of the mechanisms of functional cerebral development.
Previous investigations of resting-state network development in infants using fcMRI
Fransson et al. performed the first reported investigation of neural network development in neonates, studying 12 former preterm infants (5 female, 7 male) at term-equivalent PMA (Fransson, Skiold et al. 2007
). Infants were scored for the presence of brain injury, and only infants without parenchymal hemorrhagic infarction or white matter injury were included. Varied perinatal conditions, including bronchopulmonary dysplasia, patent ductus arteriosus ligation, and necrotizing enterocolitis were reported in a subset of infants. Subjects were sedated with chloral hydrate. Images were acquired on a 1.5-T scanner using an 8-channel receive only head coil. Parameters were verified via
reproduction of known RSNs in a single adult subject. Data containing significant motion were removed from the analysis during postprocessing. Spontaneous brain activity was detected using ICA. The majority of the signal variance occurred in the 0.01–0.05 Hz range. A total of 5 RSNs were reported. These were located in the occipital, somatomotor, temporal, parietal, and anterior prefrontal cortices. Networks were bilateral, with strong correlation across hemispheres. Limited intrahemispheric connectivity between disparate regions was found. The authors also reported a ‘proto-default-mode network’ which included the posterior cingulate and lateral parietal cortices. There was no identified network involving anterior and posterior midline regions. The authors concluded that the infant brain hosts resting-state activity, demonstrating commonalities and disparities with patterns previously noted in neural networks identified in adults.
Fransson et al. subsequently sought to identify RSNs in healthy term infants (Fransson, Skiold et al. 2009
). Nineteen healthy, term-born infants (11 female, 8 male) were studied. Studies were typically performed within the first 2 weeks of life. Infants were scanned during natural sleep. fcMRI acquisition parameters and analysis techniques were comparable to those described above. Six RSNs were reported, located in the occipital, sensorimotor, temporal, parietal, and prefrontal cortices in addition to the basal ganglia. Networks demonstrated varied strength based upon location. Results were similar to those reported in their initial study of preterm infants. A comparable ‘proto-default-mode network’ was reported. The authors concluded that multiple resting networks were identifiable in the term-born infant brain, with neither preterm birth nor light sedation interfering with the ability to detect resting-state patterns in the infant brain.
Recently, Fransson et al. applied network analysis using graph-theoretical measures to fcMRI data collected from infants in an effort to identify ‘cortical hubs’ of RSNs in this population (Fransson, Aden et al. 2010
). Eighteen healthy, term-born infants (11 female, 7 male) were studied using similar methodology to prior investigations. Graph theory was applied to correlation matrices containing the pairwise correlation coefficients for target brain voxels with respect to all other voxels. Cortical hubs (i. e
., brain regions demonstrating a disproportionately high level of anatomical connectivity and which presumably play a critical role in information flow) were identified primarily in regions within or adjacent to the primary sensorimotor cortices, with higher association cortices, including those previously identified to compose the mature default mode network and frontoparietal attention network, found to be less substantive candidates. In addition, functional connections associated with identified cortical hubs largely involved primary sensory systems, including the sensorimotor, auditory, and visual systems. Each of these patterns stood in direct contrast to results obtained from adult subjects studied with comparable techniques, for which cortical hubs were identified in higher association cortices such as the insula, precuneus, and ventromedial prefrontal cortex. The networks identified in both infants and older subjects demonstrated properties of a small world network organization, with short path lengths combined with local clustering enabling efficient data processing and information flow at local and global levels. Finally, presumed precursors of the default mode network, predominantly centered on the posterior cingulate cortex, were again reported. The authors concluded their findings supported the concept that neural networks undergo gradual maturation with advancing age, with functional connections necessary for completing tasks related to basal perception and action behavior maturing much earlier than those associated with higher order cognitive functions.
Smyser et al. performed longitudinal resting-state network analysis in a cohort of preterm infants aged 26 weeks PMA through term equivalent age at PMA-specific time points (Smyser, Inder et al. 2010
). A total of 90 data sets were collected from 53 preterm infants. Longitudinal data were available for 28 infants. Infants were excluded if found to have high grade intraventricular hemorrhage, multicystic periventricular leukomalacia, and/or large cerebellar hemorrhage. Ten term control infants were also studied. Subjects were scanned on a 3T scanner using an infant head coil during natural sleep or while resting quietly. Data containing significant motion were removed during processing. SCA was used. Multiple RSNs were identified as early as 26 weeks PMA involving varied cortical regions, in addition to the thalamus and cerebellum (). Identified networks demonstrated characteristic patterns of development based upon location, with distance playing a critical role in rate of RSN development. Early neural networks predominantly consisted of localized connections centered on cortical regions of interest, with nearby voxels demonstrating high levels of correlation that decreased with distance (i.e.
, ‘local bloom’). These networks gradually became more focused (i.e.
, decreasing size of localized activation) with advancing age, likely reflecting the maturing structural and functional localized organization of the developing brain. Further, functional connections between physically more distant regions did not appear until later in development. Significant differences were identified in infants born at term versus those born prematurely. Putative precursors of the default mode network were identified in term control infants. These connections were not detected in preterm infants, including those at term equivalent age. The identified patterns of network development demonstrate the earliest forms of cerebral functional connectivity, and were consistent with prior investigations of functional neurodevelopment in this population.
Longitudinal neural network development in preterm infants
Doria et al. subsequently explored early development of RSNs through longitudinal investigation of prematurely-born infants aged 29 weeks PMA through term-equivalent (Doria, Beckmann et al. 2010
). Sixty-two preterm data sets (28 female, 34 male) were included in the analysis. Diffuse high signal intensity was identified in 35 infants, with no additional overt anatomic lesions reported. Eight term-born infants were also investigated. Studies were performed on a 3T scanner using an 8-channel phased-array head coil. Investigations at term-equivalent were performed under sedation with chloral hydrate, with a subset of infants studied at earlier PMA scanned during natural sleep. Data sets containing pervasive motion were excluded. Both ICA and SCA were employed. Visual, auditory, somatosensory, motor, default mode, frontoparietal, and executive control RSNs comparable to those identified in adult subjects were identified in infants as early as 30 weeks PMA. A subset of networks was not identifiable in infants in the earliest age categories, but all networks were recognizable by term equivalent. Identified RSNs demonstrated network specific rates of development, exhibiting increasingly coherent interhemispheric activity with advancing PMA. Comparable results were obtained using each analysis approach, with increasing thalamic connectivity noted using SCA. No significant discrepancies were identified between infants scanned with and without light sedation. Recognized differences between prematurely-born and term infants were limited by correction for multiple comparisons. These findings further established the importance of this developmental period in the emergence of RSNs, while initiating investigation of important methodological concerns in this population.
Liu et al. performed the initial fcMRI study of older infants in natural sleep by investigating RSNs in the sensorimotor area in a cohort of one-year-old subjects (Liu, Flax et al. 2008
). Eleven healthy, term-born infants (6 female, 5 male) were studied. Studies were performed on a 1.5-T scanner. A standard (adult) RF head coil was utilized to permit use of sound attenuation equipment. Image acquisition parameters were selected in part to maintain lower sound levels for sleeping subjects. Data containing significant motion were removed during processing. ICA was utilized to identify RSNs. Two networks were reported in the sensorimotor cortex, one located in the hand region and the other in the foot/leg region. Peak frequency for signal variance was 0.02–0.04 Hz. Unilateral correlation was identified for 9 individual subjects and in the group analysis, with bilateral correlation between homotopic counterparts identified in the remaining two infants. The authors postulated the asymmetry in networks was in part secondary to differing levels of sleep, hypothesizing that deep sleep reduces inter
hemispheric connectivity, but not intra
hemispheric connectivity. They also suggested identified networks represented more mature forms of those reported in Fransson et al.’s investigation of former preterm infants. However, this predominance of intrahemispheric connectivity in the sensorimotor cortex has not been confirmed in subsequent investigations in this age group, with identified neural networks in this region demonstrating bilateral associations between homotopic counterparts, including during sleep (see below) (Lin, Zhu et al. 2008
Lin et al. performed the first reported investigation of longitudinal development within selected neural networks during infancy and early childhood (Lin, Zhu et al. 2008
). A total of 35 fcMRI data sets were analyzed: 16 collected from late preterm and term infants 2–4 weeks of age, 12 from one-year-old infants and 7 from two-year-old infants. Subjects were healthy infants born between 35 and 42 weeks gestation with normal anatomic images. Subjects were scanned during natural sleep. Studies were performed on a 3T scanner. In a subset of infants and one-year-old subjects, TR time was decreased with constant total acquisition time. Data containing significant motion were removed during processing. Three regions were investigated utilizing SCA, including the primary motor, sensory, and visual cortices. Neural networks were identified in each region for subjects within each age category. For each region, functional correlations between homotopic counterparts demonstrated increasing size and strength with advancing age, with variability in rate of development between the sensorimotor and occipital cortices. Similar results were obtained for images collected with both TR times.
The same group performed a similar investigation of temporal and spatial development within the default mode network. Results from a cohort including 20 infants (11 female, 9 male), 24 one-year-olds (8 female, 16 male), and 27 two-year-olds (17 female, 10 male) were reported (Gao, Zhu et al. 2009
). Inclusion criteria were similar to those described above. Subjects were scanned without sedation. Studies were again performed on a 3T scanner with identical acquisition parameters. In this study, ICA was utilized to identify neural networks, and graph theory was applied to group correlation matrices, with a spring embedding algorithm used to depict connection patterns. The authors report an incomplete, primitive default mode network in infants which increases markedly in number of regions connected and strength by one year of age. This process continues through two years of age, with the network increasingly similar to that identified in older pediatric populations and adults. Functional associations between the medial prefrontal cortex and posterior cingulate cortex were observed in each age group, and connections to the medial prefrontal cortex emerged more strongly after one year of age. Application of graph theory demonstrated regionally-specific patterns of evolving connections between regions composing the mature default mode network. The medial prefrontal cortex and posterior cingulate cortex were most strongly connected with other regions (i.e.
, ‘hubs’), while areas located at greater distances away from the midline demonstrated lower connection strength.
Damaraju et al. investigated the effects of prematurity on the spatial and temporal properties of RSN development during early childhood (Damaraju, Phillips et al. 2010
). fcMRI data were analyzed from 47 infants, including 16 preterm and 9 term infants at 18 months of age and 13 preterm and 9 term infants at 36 months of age. All subjects had normal development and structural MRI scans. Studies were performed on a 3T scanner. Subsets of preterm infants were sedated with chloral hydrate, while term infants were studied without sedation. Both ICA and SCA were employed, with ICA results used to define regions of interest for subsequent application in SCA. Time course spectra were determined using multi-taper spectral estimation. RSNs including visual, temporal, motor, basal ganglia, and default mode networks were identified in each group, with similar group mean patterns obtained using each analysis approach. Spatial properties between preterm and term-born infants were generally comparable in each age category. However, discrepancies in power spectrum and connection strength were identified between RSNs of infants categorized by gestational age at birth studied at 36 months of age. In infants born prematurely, neural networks including the basal ganglia demonstrated increased spectral energy at low frequencies. Additionally, RSNs in term infants were stronger than those in their preterm counterparts, with differences evident in a network specific pattern. The influence of sedation was limited throughout the analyses, with the significance of increased connection strength in non-sedated subjects restricted by small subject numbers. The authors concluded anatomical locations of RSNs are established early across all groups, with differences between neural networks in preterm infants and those born at term more prominent with increasing age.