The Establishment Of Fetal And Infant Brain Atlases And Their Application In Brain Connectome Study

Atlas-based parcellation is useful in the brain development study.Due to the pronounced differences in overall and regional brain size and shape,and the signal intensity profiles between adult and fetus or infant,the normalization and parcellation of fetal and infant brains using adult templates or atlases can lead to significant offsets of the labelled structures.Atlas can offer atlas coordinate for comparing brain properties between subjects,report the anatomic location of a region of interest(ROI).Moreover,atlas can be used to segment the brain regions as network nodes to investigate brain connectivity in the network studies.In this study,the first part is about the establishment of 1-and 2-year-old atlas.The second part is about structural network development from middle fetal stage to birth.Since no digital mid-fetal stage brain atlas exists,so in this part,template-free algorithm were used to segment the 20 PMW fetal brain as network nodes.Establishing the 20 PMW brain atlas is one of the highly anticipated future research directions.Part 1:1-and 2-year-old human brain DTI atlases with comprehensive gray and white matter labelsIntroduction:Dramatic brain structural changes take place in the first 2 years of life which is also characterized by emergence of neurodevelopmental disorders such as autism.Atlas-based automated structure labeling is useful for analyzing functional and structural neuroimaging data.Due to the relative large and nonlinear neuroanatomical differences between pediatric and adult brains,the pediatric brain parcellation using adult atlases can lead to significant offsets of labeled structures.The brain atlases used for automated structure labeling have not been fully developed in various development stages,especially for 1-and 2-year-old children.Hence,the age-specific 1-and 2-year-old atlases with comprehensive gray matter(GM)and white matter(WM)labels is needed.In this study,with diffusion tensor imaging(DTI)data of 23 healthy 1-year-old and 27 2-year-old children,we created 1-year-old and 2-year-old brain DTI atlases with comprehensive labels of 124 GM and WM structures.Integrated with appropriate registration methods,the atlases can be used for automatic labeling of age-matched children brain images with high accuracy and small offsets of labeling.Applying the atlases to automatically and effectively delineate the deep WM connectome development from 3-38 months was demonstrated.Materials and methods:T1 weighted images and DTI data of 90 children with postnatal age of 3 to 38 months participated in this study.Among these 90 subjects,23 children aged from 9-15 months and 27 toddlers aged from 20-28 month were used to create 1-and 2-year-old templates,respectively.All data were acquired on a 3T Philips Achieva System.To obtain the co-registered T1 weighted and DTI images of each subject,affine and large deformation diffeomorphic metric mapping(LDDMM)transformations were used to wrap b0 images to the contrast-reversed T1 weighted images using DiffeoMap.The resultant registration matrix was applied to transfer diffusion tensor.After these procedures,the co-registered T1 weighted and DTI images were created,with a 1.0mm isotropic resolution.Sngle-subject and population-averaged templates:After the AC-PC alignment,all the T1 weighted images were registered to the representative subject images.One step of averaging of all these registered images,three consecutive steps of averaging of registered images after three affine transformations respectively and one last step of averaging of registered images after LDDMM transformations were conducted.The affine and LDDMM transformation matrices obtained from the scalar images were applied to the tensor field to create normalized tensor fields.The representative single-subject images linearly normalized to the population-averaged-linear template to get the single-subject template.Finally,the single-subject,population-averaged-linear,and population-averaged-nonlinear templates were generated.Computation of Jacobian determinants:Ten 1-year-old children were selected and registered to the established 1-year-old,2-year-old and JHU-DTI-MNI single-subject templates using LDDMM registration.Based on the generated LDDMM transformation matrices,logarithms of the Jacobian determinants of the entire brain were computed.Three maps of Jacobian determinant were generated based on the transformation matrices of these ten 1-year-old brains to the 1-year-old,2-year-old and JHU-DTI-MNI single-subject templates.The histograms of the regional volumetric changes were established with the whole brain Jacobian determinant maps obtained above.Establishment of the comprehensive gray and white matter brain atlases:Comprehensive labels of 1 24 GM and WM structures were obtained with manual delineation on axial planes of the single-subject template with ROIEditor,followed by adjustment in coronal and sagittal planes using available atlases as references.The deep WM tracts were labeled with DTI orientation-encoded colormaps and were 3D reconstructed by DTI tractography.The subcortical GM nuclei were parcellated with T1 weighted images.The cerebral cortex structures were delineated based on the sulcal and gyral patterns on the T1 weighted images followed by the modifications based on 3D reconstructed labeled structures using Amira.Test of the automated labeling using the established atlases:Five randomly selected 1-year-old and five randomly selected 2-year-old subjects in native space were automatically parcellated based on 1-and 2-year-old atlases respectively to test the accuracy of atlas-based automated labeling.For each subject,ten brain structures on five 2D image slices were selected for quantitative evaluation.To investigate inter-rater variability,two raters(L.F.and L.S)manually segmented these ten structures,working as gold standard.The automated labels were tested against manual labels with measurements of Dice ratio and L1 errors.The WM connecome development based on atlas labels:The 1-and 2-year-old atlases were transformed into the subject images in native space,and the labels of WM structures were obtained.1-year-old brain atlas was used to segment the children aged from 3 to 18 month,and 2-year-old atlas was used to segment the children aged from 18 to 38 months.The DTI-derived measurements of the major WM tracts which were categorized into five distinctive track groups were calcalued to illustrate the application of the established atlases.Results:The morphological changes from 1-year-old to adult brains:Higher local volume expansions are apparent in frontal and temporal areas from 1-year-old to adult template as shown in the Jacobian determinant maps.The Jacobian determinant with the transformation from 1-year-old subject brains to the 1-year-old template is more concentrated,and the center of the Jacobian determinant histogram is close to 1.However,the Jacobian determinant with the transformation from 1-year-old subject brains to the adult brain template is more dispersive,and the center of the Jacobian determinant histogram shifts to around 1.3.Comprehensive gray and white matter labels:The established atlases on the T1 weighted images and orientation-encoded colormaps in axial planes includes comprehensive GM and WM labels of 52 cerebral cortical structures,40 deep cerebral WM tracts,10 subcortical GM structures and 22 brainstem and cerebellar structures.Test of the automated labeling using the established atlases:The Dice radio between automated and manual delineation were over 0.8 in most structures,indicating almost perfect automated labeling.The Dice radio was more than 0.75 in cingulate and uncinate fasciculus,indicating registration accuracy even for the irregular anatomical structures.The DTI measurements of the WM tracts and the volumes of brain structure illustrate the application of the established atlases.The development of WM connecome from 3-38 month:The general age-related increases of FA and decrease of MD,A×D,and RD in all tested WM tracts were found.Discussion and Conclusions:We established 1-and 2-year-old DTI atlases with comprehensive 124 GM and WM labels.The test results suggested that the established atlases can be applied to label age-matched brain MRI images automatically and accurately.The calculation of the DTI-derived measurements of deep WM structure connectome demonstrated the application of the atlases.Howerer,the brain atlases used to segment fetal brains do not developed well,especially for the fetal brain in the middle fetal stage.Thus,establishing the fetal brain atlas is urgent for the fetal brain connectome study.The established 1-and 2-year-old DTI atlases may be used for understanding not only normal brain development but also serving as clinical anatomical references for neurodevelopmental disorders.Part 2:Human fetal brain connectome:structural network development from middle fetal stage to birthIntroduction:Complicated molecular and cellular processes take place in a spatiotemporally heterogeneous and precisely regulated pattern in the human fetal brain,yielding not only dramatic morphological and microstructural changes,but also macroscale connectomic transitions.One of the most characteristic fetal brain developmental processes is the migration of neurons from the ventricular zone to the cortical plate along the glial fibers.The neurons begin to grow their axonal,dendritic,and synaptic projections in the cortical plate.In parallel to those maturational processes,human fetal brain white matter(WM)axons appear and structural connections based on axons emerge during this period.As the underlying substrate of the fetal brain structural network,both dynamic neuronal migration pathways and rapidly developing fetal WM fibers could fundamentally reshape early fetal brain connectome.However,both transition of the structural connectome from the mid-fetal stage to birth and contribution of different types of neural fibers to the structural network in the mid-fetal brain are not yet known.Atlases can be used to parcellate the brain as network nodes,but the digital 20 postmentrual week(PMW)atlas does not yet exist.So in this paper,the popular template-free parcellation scheme was applied to parcellate the smooth brain as network nodes.Quantifying structural connectome development can,therefore,not only shed light on brain reconfiguration in this critical yet rarely studied developmental period,but also reveal alterations of the connectome under neuropathological conditions.Materials and methods:The DTI data of 10 postmortem fetal brain samples at around 20 PMW,twelve preterm neonates at 35PMW,and twelve full-term neonates at 40PMW were used.The traced brain fibers filtered at different lengths:Deterministic fiber tracking was performed using TrackVis and all brain fibers were filtered at different lengths.Categorization of the traced brain fibers based on their terminal locations in the cerebral wall for 20PMW:The cerebral wall were manually delineated into three layers,cortical plate with marginal zone,subplate and inner layer.Both terminals of groupl fibers are located in the cortical plate.For these group2 fibers,one of the terminals is located in the inner layers.Cortical parcellation for network node definition:A template-free parcellation scheme was applied to parcellate each brain into 80 different regions with similar size.DTI tractographv for network edge generation:A filtering algorithm was applied to keep only tracts connecting two different nodes and an 80*80 weighted connectivity matrix was established for each subject.The same methods were used to create the connectivity matrix contributed by two different fibers.Network analysis with graph theory:With different edge weight thresholds,network measures were calculated using GRETNA toolbox.Results:Chansins profile of fetal brain fibers during development:The fetal brain fibers become denser from 20 to 40PMW.More long-range fibers connecting between occipital lobe and distal frontal or temporal lobes appear at 35 and 40PMW.20PMW fetal brain cerebral wall and categorization of the traced brain fibers based on their terminal locations in the cerebral wall:The 20PMW fetal brain cerebral wall can be subdivided into three layers:cortical plate with marginal zone,subplate,and inner layer.Based on the fiber terminals,the fibers were categorized into two groups.Network measures contributed by different sroups of fibers and small-world property in the 20PMW fetal brain structure connectome:Small-word property is prominent at 20PMW fetal brain structural connectome.The contribution of group2 fibers to network strength and efficiencies are significantly larger than that of groupl fibers.Fetal structural network development from 20PMW to 4OPMW:The fetal brain structural network gets stronger and more efficient from 20 to 40PMW.Furthermore,network strength and global efficiency increase more rapidly in 20-35PMW than in 35-40PMW.Discussion and Conclusions:Visualization of the whole brain fiber distribution by the lengths suggested that the network reconfiguration in this developmental period could be associated with a significant increase of major long association WM fibers.Non-WM neural fibers could be a major contributor to the structural network configuration at 20PMW and small-world network organization exists as early as 20PMW.The contribution of two distinctive fiber groups to fetal structural connectome at 20PMW may shed light on possible mechanism of critical yet unknown structural architecture maturation in the middle fetal stage.In addition,more rapid increases of network strength and global efficiency in 20-35PMW than in 35-40PMW suggests dramatic structural reconfiguration from the middle-fetal stage to the middle 3rd trimester.Since it is generally difficult to acquire high resolution DTI data for fetal brains at 20 and 35PMW,to our knowledge,this study represents the first record of quantified fetal brain structural connectome development from as early as 20PMW.No digital 20PMW fetal brain atlas which can be used to segment the brain as network nodes exist,so establishing the 20PMW atlas is one of the highly anticipated future research directions.