Brain cerebral cortical microstructure underlies neuronal circuit formation and function emergence during brain maturation. Regionally distinctive cortical microstructural architecture profiles around birth result from immensely complicated and spatiotemporally heterogeneous underlying cellular and molecular processes (Silbereis et al., 2016), including neurogenesis, synapse formation, dendritic arborization, axonal growth, pruning, and myelination. Disturbance of such precisely regulated maturational events is associated with mental disorders (Innocenti and Price, 2005). Diffusion magnetic resonance imaging (dMRI) has been widely used for quantifying microstructural changes in white matter (WM) maturation (e.g. Dubois et al., 2008; Mukherjee et al., 2001). Because of its sensitivity to organized cortical tissue (e.g. radial glial scaffold; Rakic, 1995; Sidman and Rakic, 1973) unique in the fetal and infant brain, dMRI also offers insights into maturation of cortical cytoarchitecture. Cortical fractional anisotropy (FA), a dMRI-derived measurement, of infant and fetal brain can effectively quantify local cortical microstructural architecture related to dendritic arborization and synaptic formation. Thus, cortical FA can potentially be used to infer specific brain circuit formation. In early cortical development, most of cortical neurons are generated in the ventricular and subventricular zone. These neurons migrate toward the cortical surface along a radially arranged scaffolding of glial cells where relatively high FA values are usually observed (Huang et al., 2013; McKinstry et al., 2002). During emergence of brain circuits, increasing dendritic arborization (Bystron et al., 2008; Sidman and Rakic, 1973), synaptic formation (Huttenlocher and Dabholkar, 1997), and myelination of intracortical axons (Yakovlev and Lecours, 1967) disrupt the highly organized radial glia in the immature cortex and result in cortical FA decreases. Such reproducible cortical FA change patterns were documented in many studies of perinatal human brain development (Ball et al., 2013; Huang et al., 2006; Huang et al., 2009; Huang et al., 2013; Kroenke et al., 2007; McKinstry et al., 2002; Neil et al., 1998; Ouyang et al., 2019a; Ouyang et al., 2019b; Yu et al., 2016), suggesting sensitivity of cortical FA measures to maturational processes of cortical microstructure. Diffusion-MRI-based regional cortical microstructure at birth, encoding rich ‘footage’ of regional cellular and molecular processes, may provide novel information regarding typical cortical development and biomarkers for neuropsychiatric disorders.

The first 2 years of life is a critical period for behavioral development, with brain development in this period most rapid across the lifespan. In parallel to rapid maturation of cortical architecture and establishment of complex neuronal connections (Hüppi et al., 1998; Ouyang et al., 2019a; Pfefferbaum et al., 1994), babies learn to walk, talk, and build the core capacities for lifetime. Infant behaviors including cognition, language, and motor emerge during this time and become measurable at around 2 years of age. Reliable diagnosis for many neuropsychiatric disorders, such as autism spectrum disorder (ASD), can be made only around 2 years of age or later (Marín, 2016), as diagnoses rely on observing behavioral problems that are difficult to recognize in early infancy (Arpi and Ferrari, 2013; Ozonoff et al., 2010). On the other hand, early intervention for ASD, especially before 2 years of age, has demonstrated potential on improving outcomes (Rogers et al., 2014). Given that infants cannot communicate with language or writing in early infancy, there may be no better way to assess their brain development other than neuroimaging. Prediction of future cognition and behavior at 2 years of age or later based on brain features around birth creates an invaluable time window for individualized biomarker detection and early tailored intervention leading to better outcomes.

Individual differences in brain WM microstructural architectures (Scholz et al., 2009; Yu et al., 2020), behavior, and functions (Braga and Buckner, 2017; Xu et al., 2019) have been well recognized. Individual variability in brain structures and associated individual variability in future behaviors can be harnessed for robust prediction at the single-subject level (Kanai and Rees, 2011; Rosenberg et al., 2018), a step further than group classification. A few studies have been conducted previously to investigate within-sample imaging-outcome correlations (Ball et al., 2015; Counsell et al., 2014; Deoni et al., 2016; Hintz et al., 2015; Keunen et al., 2017; Peyton et al., 2020; Wee et al., 2017; Woodward et al., 2006), while such correlation approaches made it impossible to be applied to new and incoming subjects. Machine learning approaches that can adopt new subjects and yield continuous prediction values have been explored only recently based on WM structural networks (Girault et al., 2019; Kawahara et al., 2017). Compared to association of WM microstructure with cortical regions through WM fiber end point connectivity, association of cortical microstructure with cortical regions is more direct. Thus, FA of a specific cortical region can be used to directly reflect certain functions of the same cortical region. Our previous study Ouyang et al., 2019b demonstrated that cortical FA predicted neonate age with high accuracy. Regionally distinctive cortical microstructure around birth encodes the information that may predict distinctive functions manifested by future behavior and potentially identify the most sensitive regions as imaging markers to detect early behavioral abnormality. However, dMRI-based cortical microstructure has not been evaluated for predicting either discrete or continuous future behavioral measurement so far. And dMRI-based cortical microstructure has not been incorporated into a machine-learning-based prediction model for predicting future behavior, either.

In this study, we leveraged individual variability of cortical microstructure profiles of neonate brains for predicting future behavior. A novel machine-learning-based model using regional cortical microstructure markers from dMRI was developed to predict continuous outcome values. This model is also capable of incorporating new subjects in contrast to within-sample imaging-outcome correlations. We hypothesized that dMRI-based cortical microstructure at birth only (without inclusion of any WM microstructure information) could robustly predict the future neurodevelopmental outcomes. Out of 107 recruited neonates, high-resolution (0.656 × 0.656 × 1.6 mm³) dMRI data were acquired from 87 neonates, of which 46 underwent a follow-up study at their 2 years of age for neurobehavioral assessments of cognitive, language, and motor abilities. Cortical microstructural architectures at birth were quantified by cortical FA on the cortical skeleton to alleviate partial volume effects (Ouyang et al., 2019b; Yu et al., 2016). Regional cortical FA measures were then used to form feature vectors to predict neurodevelopmental outcomes at 2 years of age. We further quantified the contribution of each cortical region in predicting different outcomes, as distinctive behaviors are likely encoded in uniquely distributed pattern across the cerebral cortex.

We leveraged individual variability in the cortical microstructural architecture at birth for a robust prediction of future behavioral outcomes in continuous values. Cortical microstructure at birth, encoding rich ‘footage’ of regional cellular and molecular processes in early human brain development, was evaluated as the baseline measurements for prediction. Our previous studies (Huang et al., 2009; Ouyang et al., 2019b; Yu et al., 2016) found that individual neonate cortical microstructure profile characterized by different levels of dendritic arborization could be reliably quantified with dMRI-based cortical FA that reveals cortical maturation signature. In this study, we further demonstrated that individual variability of cortical microstructure profile around birth can be used to robustly predict the cognitive and language outcomes of individual infant at 2 years of age. By harnessing different encoding patterns of cognitive and language functions across the entire cortex, we quantified distinguishable contributions of each cortical region and highlighted the most sensitive regions for predicting different outcomes. Cortical regions contributing heavily to the prediction models exhibited distinctive functional selectivity for cognition and language. To our knowledge, this is the first study evaluating regional cortical microstructure for predicting future behavior, laying the foundation for future works using cortical microstructure profile as ‘neuromarkers’ to predict the risk of an individual developing health-related behavioral abnormalities (e.g. ASD). The prediction model is also capable of incorporating new and incoming subjects, a step further than previous within-sample imaging-outcome correlation studies (Ball et al., 2015; Counsell et al., 2014; Deoni et al., 2016; Hintz et al., 2015; Keunen et al., 2017; Peyton et al., 2020; Wee et al., 2017; Woodward et al., 2006). Before the presented prediction can be used to identify individual infant at risk of neurodevelopmental disorders (e.g. ASD) at a time when the infant is pre-symptomatic in behavioral assessments, the prediction model needs to be further refined with a larger sample size.

Cerebral cortex plays a central role in human cognition and behaviors. High performance achieved in prediction of cognition and language based on cortical FA measures (Figure 2) is probably due to sensitivity of cortical microstructural changes to maturational processes including synaptic formation, dendritic arborization, and axonal growth. Distinctive maturation processes manifested by differentiated cortical FA changes across cortical regions in fetal and infant brains were reproducibly reported development (Ball et al., 2013; Huang et al., 2006; Huang et al., 2009; Huang et al., 2013; Kroenke et al., 2007; McKinstry et al., 2002; Neil et al., 1998; Ouyang et al., 2019a; Ouyang et al., 2019b; Yu et al., 2016). Although cortical thickness, volume, or surface area from structural MRI scans (i.e. T1- or T2-weighted images) were conventionally primary structural measurements to characterize infant cerebral cortex development (Hazlett et al., 2017; Hill et al., 2010; Lyall et al., 2015), they cannot characterize the complex microstructural processes that take place inside the cortical mantle. Compared to macrostructural changes quantified by these conventional measurements, the underlying microstructural processes quantified by cortical FA may be more sensitive to infants with pathology such as those with risk of ASD. Because WM microstructure (e.g. FA) measurement is more widely used in dMRI studies than cortical microstructure measurement, we also evaluated WM FA at birth for predicting cognitive and language outcome at 2 years of age (Figure 4). Despite the fact that microstructural measures from solely cerebral cortex or WM at birth have similar sensitivity in predicting neurodevelopment outcomes at 2 years old (Figure 4), cortical microstructure is more directly associated with specific cortical regions and thus certain cortical functions, compared to association of WM with the cortical regions through end point connectivity. On the other hand, combined cortical and WM microstructural measures at birth offered more baseline information and resulted in improved prediction (Figure 4b and c).

Human brain development in the first 2 years is most rapid across the lifespan. During first 2 years after birth, overall size of an infant brain increases dramatically, reaching close to 90% of an adult brain volume by 2 years of age (Pfefferbaum et al., 1994). Despite rapid development, infancy period (0–2 years) of human is probably the longest among all mammals with cognitive and language functions unique in human emerging during this critical period. For instance, infants start to learn their mother tongue from babbling to full sentences, during the age of 6 months to 2–3 years (Kuhl, 2004). This lengthy yet extremely dynamic brain development processes make the prediction of 2-year neurodevelopmental outcome with brain information at birth ultimately invaluable. DMRI-based cortical microstructure at birth, before any behavioral tests could be performed, can well predict both cognition and language outcomes at 2 years of age (Figure 2). A regionally heterogeneous distribution pattern across cerebral cortex was displayed (Figure 2). Furthermore, consistent with their functions documented in the literature, the cortical regions contributing heavily to the prediction models exhibited distinguishable functional selectivity for cognition and language (Figure 3 and Figure 3—figure supplement 1). The cognitive scale in Bayley-III estimates cognitive functions including object relatedness, memory, problem solving, and manipulation on the basis of nonverbal activities (Bayley, 2006). Cortical regions with high weight in the prediction model are tightly associated with cognitive functions. Right rectus, precuneus, parahippocampal, and left fusiform gyri were among the top 10 cortical regions (painted red in Figure 3b) where microstructural measures contributed uniquely to the cognition prediction, but not to the language prediction. Parahippocampal gyrus provides poly-sensory input to the hippocampus (Witter et al., 2000) and holds an essential position for mediating memory function (Young et al., 1997). Precuneus is a pivotal hub essential in brain’s default-mode network (Buckner et al., 2008) and involved in various higher-order cognitive functions (Cavanna and Trimble, 2006). Rectus and fusiform gyri are associated with higher-level social cognition processes (Viskontas et al., 2007). For regions contributing heavily to both cognition and language prediction and painted yellow in Figure 3b, bilateral entorhinal gyri likely serve as a pivot junctional region mediating the processes of different types of sensory information during the cortex-hippocampus interplay (Witter et al., 2000). Middle and lateral fronto-orbital gyrus are involved in cognitive processes including learning, memory, and decision-making (Wikenheiser and Schoenbaum, 2016). Left postcentral gyrus with high feature contribution weight in both prediction models is associated with the general motor demands of performing tasks. Besides higher-order cognitive functions, language functions are also unique in human beings. The first 2 years of life is a critical and sensitive period for the speech-perception and speech-production development (Kuhl, 2004; Werker and Hensch, 2015). The language scale from Bayley-III includes two subdomains: receptive communication and expressive communication. Here the term ‘communication’ refers to any way that a child uses to interact with others, and includes communication in prelinguistic stage (e.g. eye gaze, gesture, facial expression, vocalizations, and words), social-emotional skills, and communication in more advanced stage of language emergence (Bayley, 2006). The distinctive regions with high feature contribution weights only in language prediction model included left inferior frontal gyrus (IFG), cingular gyrus, insular cortex, and right angular gyrus, painted green in Figure 3b. These regions are related to receptive and expressive communication. It is striking that left IFG, known as ‘Broca’s area’, was identified by this data-driven prediction model because it is well known that Broca’s area plays a pivotal role in producing language (Poeppel, 2014). Angular gyrus, another crucial language region in the parietal lobe, supports the integration of semantic information into context and transfers visually perceived words to Wernicke’s area. Left insula, a part of the articulatory network in the dual-stream model of speech processing, is involved in translating acoustic speech signals into articulatory representations in the frontal lobe (Hickok and Poeppel, 2007). Since the Bayley language scale also includes a number of items reflecting social-emotional skills, such as how a child responds to his/her name or reacts when interrupted in play, the high feature contribution weight of insular cortex may be due to its important role in social emotions (Lamm and Singer, 2010). The high feature contribution weight of cingulate cortex might be related to its key involvement in emotion and social behavior (Bush et al., 2000). Taken together, identifying these regions with highest feature contribution weights sheds light on understanding local brain structural basis underlying emergence of distinctive functions manifested by daily behavior, enhancing our knowledge of brain-behavior relationships.

Motor scores from Bayley-III were not predicted reliably in this study possibly due to low variability and low signal-to-noise ratio (SNR) of cortical FA measurement in primary sensorimotor cortical regions associated with motor function. Cortical FA measurements at primary sensorimotor cortex are relatively low compared to those at other cortices (Ball et al., 2013; Ouyang et al., 2019b) and are barely above the noise floor, as primary sensorimotor cortex develops earlier compared to cortical regions associated with higher-order brain functions. Individual variability of cortical microstructure at primary sensorimotor cortex cannot be well captured with relatively low SNR for the cortical FA measurements at these regions. High individual variability enables reliable prediction. Low functional variability at primary sensorimotor cortex was found in a largely overlapped cohort in a separate study (Xu et al., 2019) from our group, and was reproducibly found in another cohort (http://developingconnectome.org). With strict exclusion criteria of participating cohort around birth, higher average and lower variance of motor scores than those of cognition or language scores play an important role in poor prediction of motor scores. Larger group variability in motor scores and larger sample size can offset the limitations elaborated above and enhance the prediction of motor scores.

Technical considerations, limitations, and future directions are discussed below. The capacity of cortical microstructural profile at birth to predict later behavior is substantial (Figure 2). We trained models to classify the low and normal outcomes. The ROC curves and accuracy measurements in Figure 2—figure supplement 2 demonstrated high accuracy of the classification of low- and normal-outcome groups. Robustness of the prediction model was further tested against various factors. These factors included different cortical parcellation schemes (random and finer parcellations versus parcellation based on an atlas) for measuring feature vectors and individual age adjustment (Figure 2—figure supplement 1). Importantly, high performance of the prediction models is reproducible after taking above-mentioned factors into consideration. We used MAE as a metric to measure prediction errors as MAE is more robust to extreme or outlier values compared to other prediction errors such as mean square error or root mean squared error (Willmott and Matsuura, 2005). Despite relatively high dMRI resolution (0.656 × 0.656 × 1.6 mm³) being used, partial volume effects (Jeon et al., 2012) cannot be ignored for measuring cortical FA. The partial volume effects are different across brain with thinner cortical regions more severely affected. To maximally alleviate the partial volume effects and enhance the measurement accuracy, we adopted a ‘cortical skeleton’ approach (Ouyang et al., 2019b; Yu et al., 2016), demonstrated in the Figure 1—figure supplement 4, to measure cortical microstructure at the center or ‘core’ of the cortical plate. Although both preterm and term-born infants are included, none of them was clinically referred. All infants had been recruited solely for brain research and rigorously screened by a neonatologist and a pediatric neuroradiologist to exclude any infants with signs of brain injury (see Materials and methods for more details). To limit the effects of exposure to the extrauterine environment, this study was designed to make the interval between birth and scan age as short as possible. As a result, we did not find any significant correlation between birth age or MRI scan age and neurodevelopmental outcomes (see Supplementary file 2). The literature (Bonifacio et al., 2010) also indicated that the effects of premature birth on brain development are considered to be relatively trivial compared with the effects of brain injury and co-morbid condition which was not presented in any recruited infant due to strict exclusion criteria. However, including preterm infants is still considered a limitation as more pronounced neurodevelopmental deviation of children with preterm birth yet without any brain injury is possible after 2 years of age. Despite this limitation, this later deviation may not significantly affect conclusion of this study focused on 0–2 years of age. The established prediction here incorporating MRI of preterm and term born neonates scanned across 32–42 PMW could benefit future outcome prediction studies based on in utero MRI of the fetuses in the third trimester with recent advances in utero MRI techniques (e.g. Khan et al., 2019; Thomason et al., 2013; Vasung et al., 2020). With potential long-term effects on neurodevelopment in preterm neonates, in utero MRI may be a better choice for future studies on normal brain development in the third trimester than preterm neonates. Although we have taken many precautions to extract cortical FA measures and tested internal validity of our prediction analysis, several limitations will need to be addressed in future research. Despite the fact that relatively high performance of behavioral prediction at a group level was achieved with current cohort of infants, the prediction model will benefit from validation (e.g. k-fold cross-validation) and replication with an independent infant cohort of a larger sample size for generalization. Thus, prediction model with individual variability representing a general population from a much larger cohort is warranted in future research before this approach is effective for meaningfully predicting the outcome for a single individual. As indicated in Figure 4b and c, more baseline information resulted in improved prediction. Future research will also benefit from incorporating distinctive and complementary measurements from multimodal neuroimaging (e.g. Kwon et al., 2014; Smyser et al., 2016), including structural (i.e. T1- or T2-weighted), functional, and diffusion MRI. Genetic factors could also be incorporated. With these multimodal measurements, more advanced machine learning algorithms such as multi-kernel and deep learning need to be adopted and developed to further improve prediction. This study included a healthy cohort of infants for evaluating cortical microstructure for predicting future behavior. Such evaluation in the setting of pathology needs to be further validated. The observed neurodevelopmental outcomes were also contributed by unmeasured factors such as maternal age and tertiary educational level as well as other home environment variable following discharge from the hospital, all of which should be taken into consideration in the future prediction model.

In conclusion, whole-brain cortical FA at birth, encoding rich information of dendritic arborization and synaptic formation, could be reliably used for predicting neurodevelopmental outcomes of 2-year-old infants by leveraging individual variability of these measures. Feature contribution weight in cognitive or language prediction is heterogeneous across brain regions. The cortical regions contributing heavily to the prediction models exhibited distinguishable functional selectivity for cognition and language. Identifying regions with highest feature contribution weights offers preliminary findings on understanding local brain microstructural basis underlying emergence of future behavior, enhancing our knowledge of brain-behavior relationships. These findings also suggest that cortical microstructural information at birth may be potentially used for prediction of behavioral abnormality in infants with high risk for brain disorders early at a time when infants are pre-symptomatic in behavioral assessments and intervention may be most effective.

Neonate MRI datasets are publicly available and can be freely downloaded from brainmrimap.org (a public website maintained by Huang lab). Behavioral datasets are available in the supplemental information of this publication. Source codes used for prediction are available from first author’s github repository (https://github.com/MHouyang/Prediction-of-neurodevelopmental-outcome ; copy archived at https://archive.softwareheritage.org/swh:1:rev:e3506bfbfa03db27651b1803e5cb662b623f5360/).

1. 1. Arpi E
   2. Ferrari F

   (2013) Preterm birth and behaviour problems in infants and preschool-age children: a review of the recent literature

   Developmental Medicine & Child Neurology 55:788–796.

   https://doi.org/10.1111/dmcn.12142

   • Google Scholar
2. 1. Ball G
   2. Srinivasan L
   3. Aljabar P
   4. Counsell SJ
   5. Durighel G
   6. Hajnal JV
   7. Rutherford MA
   8. Edwards AD

   (2013) Development of cortical microstructure in the preterm human brain

   PNAS 110:9541–9546.

   https://doi.org/10.1073/pnas.1301652110

   • PubMed
   • Google Scholar
3. 1. Ball G
   2. Pazderova L
   3. Chew A
   4. Tusor N
   5. Merchant N
   6. Arichi T
   7. Allsop JM
   8. Cowan FM
   9. Edwards AD
   10. Counsell SJ

   (2015) Thalamocortical connectivity predicts cognition in children born preterm

   Cerebral Cortex 25:4310–4318.

   https://doi.org/10.1093/cercor/bhu331

   • PubMed
   • Google Scholar
4. Book

   1. Bayley N

   (2006)

   BSID-III: Bayley Scales of Infant Development

   Harcourt Assessment, San Antonio.

   • Google Scholar
5. 1. Bonifacio SL
   2. Glass HC
   3. Chau V
   4. Berman JI
   5. Xu D
   6. Brant R
   7. Barkovich AJ
   8. Poskitt KJ
   9. Miller SP
   10. Ferriero DM

   (2010) Extreme premature birth is not associated with impaired development of brain microstructure

   The Journal of Pediatrics 157:726–732.

   https://doi.org/10.1016/j.jpeds.2010.05.026

   • PubMed
   • Google Scholar
6. 1. Braga RM
   2. Buckner RL

   (2017) Parallel interdigitated distributed networks within the individual estimated by intrinsic functional connectivity

   Neuron 95:457–471.

   https://doi.org/10.1016/j.neuron.2017.06.038

   • PubMed
   • Google Scholar
7. Book

   1. Buckner RL
   2. Andrews-Hanna JR
   3. Schacter DL

   (2008) The brain's default network: Anatomy, function, and relevance to disease

   In: Kingstone A, Miller M. B, editors. Annals of the New York Academy of Sciences: Vol. 1124. the Year in Cognitive Neuroscience. Blackwell Publishing. pp. 1–38.

   https://doi.org/10.1196/annals.auindex_1

   • Google Scholar
8. 1. Bush G
   2. Luu P
   3. Posner MI

   (2000) Cognitive and emotional influences in anterior cingulate cortex

   Trends in Cognitive Sciences 4:215–222.

   https://doi.org/10.1016/S1364-6613(00)01483-2

   • PubMed
   • Google Scholar
9. 1. Bystron I
   2. Blakemore C
   3. Rakic P

   (2008) Development of the human cerebral cortex: boulder committee revisited

   Nature Reviews Neuroscience 9:110–122.

   https://doi.org/10.1038/nrn2252

   • PubMed
   • Google Scholar
10. 1. Cavanna AE
    2. Trimble MR

    (2006) The precuneus: a review of its functional anatomy and behavioural correlates

    Brain 129:564–583.

    https://doi.org/10.1093/brain/awl004

    • PubMed
    • Google Scholar
11. 1. Chang CC
    2. Lin CJ

    (2011) LIBSVM: a library for support vector machines

    ACM Transactions on Intelligent Systems and Technology 2:27.

    https://doi.org/10.1145/1961189.1961199

    • Google Scholar
12. 1. Counsell SJ
    2. Ball G
    3. Edwards AD

    (2014) New imaging approaches to evaluate newborn brain injury and their role in predicting developmental disorders

    Current Opinion in Neurology 27:168–175.

    https://doi.org/10.1097/WCO.0000000000000073

    • PubMed
    • Google Scholar
13. 1. Deoni SC
    2. O'Muircheartaigh J
    3. Elison JT
    4. Walker L
    5. Doernberg E
    6. Waskiewicz N
    7. Dirks H
    8. Piryatinsky I
    9. Dean DC
    10. Jumbe NL

    (2016) White matter maturation profiles through early childhood predict general cognitive ability

    Brain Structure and Function 221:1189–1203.

    https://doi.org/10.1007/s00429-014-0947-x

    • PubMed
    • Google Scholar
14. 1. Dubois J
    2. Dehaene-Lambertz G
    3. Perrin M
    4. Mangin JF
    5. Cointepas Y
    6. Duchesnay E
    7. Le Bihan D
    8. Hertz-Pannier L

    (2008) Asynchrony of the early maturation of white matter bundles in healthy infants: quantitative landmarks revealed noninvasively by diffusion tensor imaging

    Human Brain Mapping 29:14–27.

    https://doi.org/10.1002/hbm.20363

    • PubMed
    • Google Scholar
15. 1. Dukart J
    2. Schroeter ML
    3. Mueller K
    4. Alzheimer's Disease Neuroimaging Initiative

    (2011) Age correction in dementia--matching to a healthy brain

    PLOS ONE 6:e22193.

    https://doi.org/10.1371/journal.pone.0022193

    • PubMed
    • Google Scholar
16. 1. Feng L
    2. Li H
    3. Oishi K
    4. Mishra V
    5. Song L
    6. Peng Q
    7. Ouyang M
    8. Wang J
    9. Slinger M
    10. Jeon T
    11. Lee L
    12. Heyne R
    13. Chalak L
    14. Peng Y
    15. Liu S
    16. Huang H

    (2019) Age-specific gray and white matter DTI atlas for human brain at 33, 36 and 39 postmenstrual weeks

    NeuroImage 185:685–698.

    https://doi.org/10.1016/j.neuroimage.2018.06.069

    • PubMed
    • Google Scholar
17. 1. Girault JB
    2. Munsell BC
    3. Puechmaille D
    4. Goldman BD
    5. Prieto JC
    6. Styner M
    7. Gilmore JH

    (2019) White matter connectomes at birth accurately predict cognitive abilities at age 2

    NeuroImage 192:145–155.

    https://doi.org/10.1016/j.neuroimage.2019.02.060

    • PubMed
    • Google Scholar
18. 1. Hazlett HC
    2. Gu H
    3. Munsell BC
    4. Kim SH
    5. Styner M
    6. Wolff JJ
    7. Elison JT
    8. Swanson MR
    9. Zhu H
    10. Botteron KN
    11. Collins DL
    12. Constantino JN
    13. Dager SR
    14. Estes AM
    15. Evans AC
    16. Fonov VS
    17. Gerig G
    18. Kostopoulos P
    19. McKinstry RC
    20. Pandey J
    21. Paterson S
    22. Pruett JR
    23. Schultz RT
    24. Shaw DW
    25. Zwaigenbaum L
    26. Piven J
    27. IBIS Network
    28. Clinical Sites
    29. Data Coordinating Center
    30. Image Processing Core
    31. Statistical Analysis

    (2017) Early brain development in infants at high risk for autism spectrum disorder

    Nature 542:348–351.

    https://doi.org/10.1038/nature21369

    • PubMed
    • Google Scholar
19. 1. Hickok G
    2. Poeppel D

    (2007) The cortical organization of speech processing

    Nature Reviews Neuroscience 8:393–402.

    https://doi.org/10.1038/nrn2113

    • PubMed
    • Google Scholar
20. 1. Hill J
    2. Dierker D
    3. Neil J
    4. Inder T
    5. Knutsen A
    6. Harwell J
    7. Coalson T
    8. Van Essen D

    (2010) A surface-based analysis of hemispheric asymmetries and folding of cerebral cortex in term-born human infants

    Journal of Neuroscience 30:2268–2276.

    https://doi.org/10.1523/JNEUROSCI.4682-09.2010

    • PubMed
    • Google Scholar
21. 1. Hintz SR
    2. Barnes PD
    3. Bulas D
    4. Slovis TL
    5. Finer NN
    6. Wrage LA
    7. Das A
    8. Tyson JE
    9. Stevenson DK
    10. Carlo WA
    11. Walsh MC
    12. Laptook AR
    13. Yoder BA
    14. Van Meurs KP
    15. Faix RG
    16. Rich W
    17. Newman NS
    18. Cheng H
    19. Heyne RJ
    20. Vohr BR
    21. Acarregui MJ
    22. Vaucher YE
    23. Pappas A
    24. Peralta-Carcelen M
    25. Wilson-Costello DE
    26. Evans PW
    27. Goldstein RF
    28. Myers GJ
    29. Poindexter BB
    30. McGowan EC
    31. Adams-Chapman I
    32. Fuller J
    33. Higgins RD
    34. SUPPORT Study Group of the Eunice Kennedy Shriver National Institute of Child Health and Human Development Neonatal Research Network

    (2015) Neuroimaging and neurodevelopmental outcome in extremely preterm infants

    Pediatrics 135:e32–e42.

    https://doi.org/10.1542/peds.2014-0898

    • PubMed
    • Google Scholar
22. 1. Huang H
    2. Zhang J
    3. Wakana S
    4. Zhang W
    5. Ren T
    6. Richards LJ
    7. Yarowsky P
    8. Donohue P
    9. Graham E
    10. van Zijl PC
    11. Mori S

    (2006) White and gray matter development in human fetal, newborn and pediatric brains

    NeuroImage 33:27–38.

    https://doi.org/10.1016/j.neuroimage.2006.06.009

    • PubMed
    • Google Scholar
23. 1. Huang H
    2. Xue R
    3. Zhang J
    4. Ren T
    5. Richards LJ
    6. Yarowsky P
    7. Miller MI
    8. Mori S

    (2009) Anatomical characterization of human fetal brain development with diffusion tensor magnetic resonance imaging

    Journal of Neuroscience 29:4263–4273.

    https://doi.org/10.1523/JNEUROSCI.2769-08.2009

    • PubMed
    • Google Scholar
24. 1. Huang H
    2. Fan X
    3. Weiner M
    4. Martin-Cook K
    5. Xiao G
    6. Davis J
    7. Devous M
    8. Rosenberg R
    9. Diaz-Arrastia R

    (2012) Distinctive disruption patterns of white matter tracts in Alzheimer's disease with full diffusion tensor characterization

    Neurobiology of Aging 33:2029–2045.

    https://doi.org/10.1016/j.neurobiolaging.2011.06.027

    • PubMed
    • Google Scholar
25. 1. Huang H
    2. Jeon T
    3. Sedmak G
    4. Pletikos M
    5. Vasung L
    6. Xu X
    7. Yarowsky P
    8. Richards LJ
    9. Kostovic I
    10. Sestan N
    11. Mori S

    (2013) Coupling diffusion imaging with histological and gene expression analysis to examine the dynamics of cortical Areas across the fetal period of human brain development

    Cerebral Cortex 23:2620–2631.

    https://doi.org/10.1093/cercor/bhs241

    • PubMed
    • Google Scholar
26. 1. Huang H
    2. Shu N
    3. Mishra V
    4. Jeon T
    5. Chalak L
    6. Wang ZJ
    7. Rollins N
    8. Gong G
    9. Cheng H
    10. Peng Y
    11. Dong Q
    12. He Y

    (2015) Development of human brain structural networks through infancy and childhood

    Cerebral Cortex 25:1389–1404.

    https://doi.org/10.1093/cercor/bht335

    • PubMed
    • Google Scholar
27. 1. Hüppi PS
    2. Maier SE
    3. Peled S
    4. Zientara GP
    5. Barnes PD
    6. Jolesz FA
    7. Volpe JJ

    (1998) Microstructural development of human newborn cerebral white matter assessed in vivo by diffusion tensor magnetic resonance imaging

    Pediatric Research 44:584–590.

    https://doi.org/10.1203/00006450-199810000-00019

    • PubMed
    • Google Scholar
28. 1. Huttenlocher PR
    2. Dabholkar AS

    (1997) Regional differences in synaptogenesis in human cerebral cortex

    The Journal of Comparative Neurology 387:167–178.

    https://doi.org/10.1002/(SICI)1096-9861(19971020)387:2<167::AID-CNE1>3.0.CO;2-Z

    • PubMed
    • Google Scholar
29. 1. Innocenti GM
    2. Price DJ

    (2005) Exuberance in the development of cortical networks

    Nature Reviews Neuroscience 6:955–965.

    https://doi.org/10.1038/nrn1790

    • PubMed
    • Google Scholar
30. 1. Jeon T
    2. Mishra V
    3. Uh J
    4. Weiner M
    5. Hatanpaa KJ
    6. White CL
    7. Zhao YD
    8. Lu H
    9. Diaz-Arrastia R
    10. Huang H

    (2012) Regional changes of cortical mean diffusivities with aging after correction of partial volume effects

    NeuroImage 62:1705–1716.

    https://doi.org/10.1016/j.neuroimage.2012.05.082

    • PubMed
    • Google Scholar
31. 1. Jiang H
    2. van Zijl PC
    3. Kim J
    4. Pearlson GD
    5. Mori S

    (2006) DtiStudio: resource program for diffusion tensor computation and fiber bundle tracking

    Computer Methods and Programs in Biomedicine 81:106–116.

    https://doi.org/10.1016/j.cmpb.2005.08.004

    • PubMed
    • Google Scholar
32. 1. Kanai R
    2. Rees G

    (2011) The structural basis of inter-individual differences in human behaviour and cognition

    Nature Reviews Neuroscience 12:231–242.

    https://doi.org/10.1038/nrn3000

    • PubMed
    • Google Scholar
33. 1. Kawahara J
    2. Brown CJ
    3. Miller SP
    4. Booth BG
    5. Chau V
    6. Grunau RE
    7. Zwicker JG
    8. Hamarneh G

    (2017) BrainNetCNN: convolutional neural networks for brain networks; towards predicting neurodevelopment

    NeuroImage 146:1038–1049.

    https://doi.org/10.1016/j.neuroimage.2016.09.046

    • PubMed
    • Google Scholar
34. 1. Keunen K
    2. Benders MJ
    3. Leemans A
    4. Fieret-Van Stam PC
    5. Scholtens LH
    6. Viergever MA
    7. Kahn RS
    8. Groenendaal F
    9. de Vries LS
    10. van den Heuvel MP

    (2017) White matter maturation in the neonatal brain is predictive of school age cognitive capacities in children born very preterm

    Developmental Medicine & Child Neurology 59:939–946.

    https://doi.org/10.1111/dmcn.13487

    • PubMed
    • Google Scholar
35. 1. Khan S
    2. Vasung L
    3. Marami B
    4. Rollins CK
    5. Afacan O
    6. Ortinau CM
    7. Yang E
    8. Warfield SK
    9. Gholipour A

    (2019) Fetal brain growth portrayed by a spatiotemporal diffusion tensor MRI atlas computed from in utero images

    NeuroImage 185:593–608.

    https://doi.org/10.1016/j.neuroimage.2018.08.030

    • PubMed
    • Google Scholar
36. 1. Kroenke CD
    2. Van Essen DC
    3. Inder TE
    4. Rees S
    5. Bretthorst GL
    6. Neil JJ

    (2007) Microstructural changes of the baboon cerebral cortex during gestational development reflected in magnetic resonance imaging diffusion anisotropy

    Journal of Neuroscience 27:12506–12515.

    https://doi.org/10.1523/JNEUROSCI.3063-07.2007

    • PubMed
    • Google Scholar
37. 1. Kuhl PK

    (2004) Early language acquisition: cracking the speech code

    Nature Reviews Neuroscience 5:831–843.

    https://doi.org/10.1038/nrn1533

    • PubMed
    • Google Scholar
38. 1. Kwon SH
    2. Vasung L
    3. Ment LR
    4. Huppi PS

    (2014) The role of neuroimaging in predicting neurodevelopmental outcomes of preterm neonates

    Clinics in Perinatology 41:257–283.

    https://doi.org/10.1016/j.clp.2013.10.003

    • PubMed
    • Google Scholar
39. 1. Lamm C
    2. Singer T

    (2010) The role of anterior insular cortex in social emotions

    Brain Structure and Function 214:579–591.

    https://doi.org/10.1007/s00429-010-0251-3

    • PubMed
    • Google Scholar
40. 1. Lyall AE
    2. Shi F
    3. Geng X
    4. Woolson S
    5. Li G
    6. Wang L
    7. Hamer RM
    8. Shen D
    9. Gilmore JH

    (2015) Dynamic development of regional cortical thickness and surface area in early childhood

    Cerebral Cortex 25:2204–2212.

    https://doi.org/10.1093/cercor/bhu027

    • PubMed
    • Google Scholar
41. 1. Marín O

    (2016) Developmental timing and critical windows for the treatment of psychiatric disorders

    Nature Medicine 22:1229–1238.

    https://doi.org/10.1038/nm.4225

    • PubMed
    • Google Scholar
42. 1. McKinstry RC
    2. Mathur A
    3. Miller JH
    4. Ozcan A
    5. Snyder AZ
    6. Schefft GL
    7. Almli CR
    8. Shiran SI
    9. Conturo TE
    10. Neil JJ

    (2002) Radial organization of developing preterm human cerebral cortex revealed by non-invasive water diffusion anisotropy MRI

    Cerebral Cortex 12:1237–1243.

    https://doi.org/10.1093/cercor/12.12.1237

    • PubMed
    • Google Scholar
43. 1. Mukherjee P
    2. Miller JH
    3. Shimony JS
    4. Conturo TE
    5. Lee BC
    6. Almli CR
    7. McKinstry RC

    (2001) Normal brain maturation during childhood: developmental trends characterized with diffusion-tensor MR imaging

    Radiology 221:349–358.

    https://doi.org/10.1148/radiol.2212001702

    • PubMed
    • Google Scholar
44. 1. Neil JJ
    2. Shiran SI
    3. McKinstry RC
    4. Schefft GL
    5. Snyder AZ
    6. Almli CR
    7. Akbudak E
    8. Aronovitz JA
    9. Miller JP
    10. Lee BC
    11. Conturo TE

    (1998) Normal brain in human newborns: apparent diffusion coefficient and diffusion anisotropy measured by using diffusion tensor MR imaging

    Radiology 209:57–66.

    https://doi.org/10.1148/radiology.209.1.9769812

    • PubMed
    • Google Scholar
45. 1. Oishi K
    2. Mori S
    3. Donohue PK
    4. Ernst T
    5. Anderson L
    6. Buchthal S
    7. Faria A
    8. Jiang H
    9. Li X
    10. Miller MI
    11. van Zijl PC
    12. Chang L

    (2011) Multi-contrast human neonatal brain atlas: application to normal neonate development analysis

    NeuroImage 56:8–20.

    https://doi.org/10.1016/j.neuroimage.2011.01.051

    • PubMed
    • Google Scholar
46. 1. Ouyang M
    2. Cheng H
    3. Mishra V
    4. Gong G
    5. Mosconi MW
    6. Sweeney J
    7. Peng Y
    8. Huang H

    (2016) Atypical age-dependent effects of autism on white matter microstructure in children of 2-7 years

    Human Brain Mapping 37:819–832.

    https://doi.org/10.1002/hbm.23073

    • PubMed
    • Google Scholar
47. 1. Ouyang M
    2. Dubois J
    3. Yu Q
    4. Mukherjee P
    5. Huang H

    (2019a) Delineation of early brain development from fetuses to infants with diffusion MRI and beyond

    NeuroImage 185:836–850.

    https://doi.org/10.1016/j.neuroimage.2018.04.017

    • PubMed
    • Google Scholar
48. 1. Ouyang M
    2. Jeon T
    3. Sotiras A
    4. Peng Q
    5. Mishra V
    6. Halovanic C
    7. Chen M
    8. Chalak L
    9. Rollins N
    10. Roberts TPL
    11. Davatzikos C
    12. Huang H

    (2019b) Differential cortical microstructural maturation in the preterm human brain with diffusion kurtosis and tensor imaging

    PNAS 116:4681–4688.

    https://doi.org/10.1073/pnas.1812156116

    • PubMed
    • Google Scholar
49. 1. Ozonoff S
    2. Iosif AM
    3. Baguio F
    4. Cook IC
    5. Hill MM
    6. Hutman T
    7. Rogers SJ
    8. Rozga A
    9. Sangha S
    10. Sigman M
    11. Steinfeld MB
    12. Young GS

    (2010) A prospective study of the emergence of early behavioral signs of autism

    Journal of the American Academy of Child and Adolescent Psychiatry 49:256–266.

    https://doi.org/10.1097/00004583-201003000-00009

    • PubMed
    • Google Scholar
50. 1. Peyton C
    2. Einspieler C
    3. Fjørtoft T
    4. Adde L
    5. Schreiber MD
    6. Drobyshevsky A
    7. Marks JD

    (2020) Correlates of normal and abnormal general movements in infancy and Long-Term neurodevelopment of preterm infants: insights from functional connectivity studies at term equivalence

    Journal of Clinical Medicine 9:834.

    https://doi.org/10.3390/jcm9030834

    • Google Scholar
51. 1. Pfefferbaum A
    2. Mathalon DH
    3. Sullivan EV
    4. Rawles JM
    5. Zipursky RB
    6. Lim KO

    (1994) A quantitative magnetic resonance imaging study of changes in brain morphology from infancy to late adulthood

    Archives of Neurology 51:874–887.

    https://doi.org/10.1001/archneur.1994.00540210046012

    • PubMed
    • Google Scholar
52. 1. Poeppel D

    (2014) The neuroanatomic and neurophysiological infrastructure for speech and language

    Current Opinion in Neurobiology 28:142–149.

    https://doi.org/10.1016/j.conb.2014.07.005

    • PubMed
    • Google Scholar
53. 1. Rakic P

    (1995) Radial versus tangential migration of neuronal clones in the developing cerebral cortex

    PNAS 92:11323–11327.

    https://doi.org/10.1073/pnas.92.25.11323

    • PubMed
    • Google Scholar
54. 1. Rogers SJ
    2. Vismara L
    3. Wagner AL
    4. McCormick C
    5. Young G
    6. Ozonoff S

    (2014) Autism treatment in the first year of life: a pilot study of infant start, a parent-implemented intervention for symptomatic infants

    Journal of Autism and Developmental Disorders 44:2981–2995.

    https://doi.org/10.1007/s10803-014-2202-y

    • PubMed
    • Google Scholar
55. 1. Rosenberg MD
    2. Casey BJ
    3. Holmes AJ

    (2018) Prediction complements explanation in understanding the developing brain

    Nature Communications 9:589.

    https://doi.org/10.1038/s41467-018-02887-9

    • PubMed
    • Google Scholar
56. Book

    1. Scholz J
    2. Tomassini V
    3. Johansen-Berg H

    (2009) Individual differences in white matter microstructure in the healthy brain

    In: Johansen-Berg H, Behrens T. E. J, editors. Diffusion MRI. London: Academic Press. pp. 237–250.

    https://doi.org/10.1016/B978-0-12-374709-9.00011-0

    • Google Scholar
57. 1. Sidman RL
    2. Rakic P

    (1973) Neuronal migration, with special reference to developing human brain: a review

    Brain Research 62:1–35.

    https://doi.org/10.1016/0006-8993(73)90617-3

    • PubMed
    • Google Scholar
58. 1. Silbereis JC
    2. Pochareddy S
    3. Zhu Y
    4. Li M
    5. Sestan N

    (2016) The cellular and molecular landscapes of the developing human central nervous system

    Neuron 89:248–268.

    https://doi.org/10.1016/j.neuron.2015.12.008

    • PubMed
    • Google Scholar
59. 1. Smyser CD
    2. Dosenbach NU
    3. Smyser TA
    4. Snyder AZ
    5. Rogers CE
    6. Inder TE
    7. Schlaggar BL
    8. Neil JJ

    (2016) Prediction of brain maturity in infants using machine-learning algorithms

    NeuroImage 136:1–9.

    https://doi.org/10.1016/j.neuroimage.2016.05.029

    • PubMed
    • Google Scholar
60. 1. Thomason ME
    2. Dassanayake MT
    3. Shen S
    4. Katkuri Y
    5. Alexis M
    6. Anderson AL
    7. Yeo L
    8. Mody S
    9. Hernandez-Andrade E
    10. Hassan SS
    11. Studholme C
    12. Jeong JW
    13. Romero R

    (2013) Cross-hemispheric functional connectivity in the human fetal brain

    Science Translational Medicine 5:173ra24.

    https://doi.org/10.1126/scitranslmed.3004978

    • PubMed
    • Google Scholar
61. 1. Vasung L
    2. Rollins CK
    3. Yun HJ
    4. Velasco-Annis C
    5. Zhang J
    6. Wagstyl K
    7. Evans A
    8. Warfield SK
    9. Feldman HA
    10. Grant PE
    11. Gholipour A

    (2020) Quantitative in vivo MRI assessment of structural asymmetries and sexual dimorphism of transient fetal compartments in the human brain

    Cerebral Cortex 30:1752–1767.

    https://doi.org/10.1093/cercor/bhz200

    • PubMed
    • Google Scholar
62. 1. Viskontas IV
    2. Possin KL
    3. Miller BL

    (2007) Symptoms of frontotemporal dementia provide insights into orbitofrontal cortex function and social behavior

    Annals of the New York Academy of Sciences 1121:528–545.

    https://doi.org/10.1196/annals.1401.025

    • PubMed
    • Google Scholar
63. 1. Wee CY
    2. Tuan TA
    3. Broekman BF
    4. Ong MY
    5. Chong YS
    6. Kwek K
    7. Shek LP
    8. Saw SM
    9. Gluckman PD
    10. Fortier MV
    11. Meaney MJ
    12. Qiu A

    (2017) Neonatal neural networks predict children behavioral profiles later in life

    Human Brain Mapping 38:1362–1373.

    https://doi.org/10.1002/hbm.23459

    • PubMed
    • Google Scholar
64. 1. Werker JF
    2. Hensch TK

    (2015) Critical periods in speech perception: new directions

    Annual Review of Psychology 66:173–196.

    https://doi.org/10.1146/annurev-psych-010814-015104

    • PubMed
    • Google Scholar
65. 1. Wikenheiser AM
    2. Schoenbaum G

    (2016) Over the river, through the woods: cognitive maps in the Hippocampus and orbitofrontal cortex

    Nature Reviews Neuroscience 17:513–523.

    https://doi.org/10.1038/nrn.2016.56

    • PubMed
    • Google Scholar
66. 1. Willmott CJ
    2. Matsuura K

    (2005) Advantages of the mean absolute error (MAE) over the root mean square error (RMSE) in assessing average model performance

    Climate Research 30:79–82.

    https://doi.org/10.3354/cr030079

    • Google Scholar
67. 1. Witter MP
    2. Wouterlood FG
    3. Naber PA
    4. Van Haeften T

    (2000) Anatomical organization of the parahippocampal-hippocampal network

    Annals of the New York Academy of Sciences 911:1–24.

    https://doi.org/10.1111/j.1749-6632.2000.tb06716.x

    • PubMed
    • Google Scholar
68. 1. Woodward LJ
    2. Anderson PJ
    3. Austin NC
    4. Howard K
    5. Inder TE

    (2006) Neonatal MRI to predict neurodevelopmental outcomes in preterm infants

    New England Journal of Medicine 355:685–694.

    https://doi.org/10.1056/NEJMoa053792

    • PubMed
    • Google Scholar
69. 1. Xu Y
    2. Cao M
    3. Liao X
    4. Xia M
    5. Wang X
    6. Jeon T
    7. Ouyang M
    8. Chalak L
    9. Rollins N
    10. Huang H
    11. He Y

    (2019) Development and emergence of individual variability in the functional connectivity architecture of the preterm human brain

    Cerebral Cortex 29:4208–4222.

    https://doi.org/10.1093/cercor/bhy302

    • PubMed
    • Google Scholar
70. Book

    1. Yakovlev PI
    2. Lecours AR

    (1967)

    The myelogenetic cycles of regional maturation of the brain

    In: Minkowski A, editors. Regional Development of the Brain in Early Life. Oxford: Blackwell Science. pp. 3–70.

    • Google Scholar
71. 1. Young BJ
    2. Otto T
    3. Fox GD
    4. Eichenbaum H

    (1997) Memory representation within the parahippocampal region

    The Journal of Neuroscience 17:5183–5195.

    https://doi.org/10.1523/JNEUROSCI.17-13-05183.1997

    • PubMed
    • Google Scholar
72. 1. Yu Q
    2. Ouyang A
    3. Chalak L
    4. Jeon T
    5. Chia J
    6. Mishra V
    7. Sivarajan M
    8. Jackson G
    9. Rollins N
    10. Liu S
    11. Huang H

    (2016) Structural development of human fetal and preterm brain cortical plate based on Population-Averaged templates

    Cerebral Cortex 26:4381–4391.

    https://doi.org/10.1093/cercor/bhv201

    • PubMed
    • Google Scholar
73. 1. Yu Q
    2. Peng Y
    3. Kang H
    4. Peng Q
    5. Ouyang M
    6. Slinger M
    7. Hu D
    8. Shou H
    9. Fang F
    10. Huang H

    (2020) Differential white matter maturation from birth to 8 Years of Age

    Cerebral Cortex 30:2674–2690.

    https://doi.org/10.1093/cercor/bhz268

    • Google Scholar
74. 1. Zalesky A
    2. Fornito A
    3. Bullmore ET

    (2010) Network-based statistic: identifying differences in brain networks

    NeuroImage 53:1197–1207.

    https://doi.org/10.1016/j.neuroimage.2010.06.041

    • PubMed
    • Google Scholar

Acceptance summary:

This paper uses brain imaging in neonates to show that brain structure at birth can predict behaviour up to 2 years later. This work provides an important link between brain health at birth and later cognitive development.

Decision letter after peer review:

Thank you for submitting your article "Diffusion-MRI-based regional cortical microstructure at birth for predicting neurodevelopmental outcomes of 2-year-olds" for consideration by eLife. Your article has been reviewed by three peer reviewers, one of whom is a member of our Board of Reviewing Editors, and the evaluation has been overseen by Richard Ivry as the Senior Editor. The reviewers have opted to remain anonymous.

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

As the editors have judged that your manuscript is of interest, but as described below that additional experiments are required before it is published, we would like to draw your attention to changes in our revision policy that we have made in response to COVID-19 (https://elifesciences.org/articles/57162). First, because many researchers have temporarily lost access to the labs, we will give authors as much time as they need to submit revised manuscripts. We are also offering, if you choose, to post the manuscript to bioRxiv (if it is not already there) along with this decision letter and a formal designation that the manuscript is "in revision at eLife". Please let us know if you would like to pursue this option. (If your work is more suitable for medRxiv, you will need to post the preprint yourself, as the mechanisms for us to do so are still in development.)

Summary

This manuscript focuses on trying to predict cognitive assessment scores, collected at 18 months of age, from neonatal cortical structure. In 46 neonates, they use support vector regression with leave-one-out cross-validation to model the multivariate relationship between cortical microstructure and later scores. The authors are able to calculate predicted individual cognitive and language scores that correlate with actual scores.

All reviewers appreciate the challenge in collecting these kinds of data and the cross-validation procedure used for outcome prediction. However, we shared a series of concerns that must be addressed before we can consider publication.

Essential revisions

– We are concerned about the representativeness and size of the sample, and therefore how generalisable these results might be to the larger neonatal population. The infants are both born and scanned across a very wide age range, 32-42 postnatal weeks. This means that some of the infants were both very preterm at birth and at scan, with very different cortical microstructure (seen in the supplementary figures). There is also very different partial voluming in these two ages simply due to brain volume differences. Moreover, the study population is not described very well. We cannot discern what the relative proportions of preterm and term infants are. The authors only report that the gestational age ranged from 26 to 41 weeks and that images were obtained at postmenstrual ages ranging from 31 to 41 weeks. More detail is needed regarding the study population and who was imaged when. In addition, more than half of the recruited infants didn't finish the study, and it is important to know if there were differences between those subjects who were apparently lost to follow up and those that weren't.

– The mixing of term and preterm groups in the analysis raises an interpretation issue. These two populations aren't equivalent, and few would argue that preterm infants are "normal" despite having normal conventional neuroimaging studies. It is conceivable that the findings of the study are driven by abnormalities in the preterm infants, and are thereby an indication of areas which are most commonly injured in preterm infants. We suggest the authors consider i) comparing findings of preterm with term infants and ii) evaluating outcome correlations for these populations separately. An alternative would be to examine term-equivalent scans for all participants.

– Please clarify the extent to which the correlations between real and predicted scores driven by extreme scorers.

– The authors have demonstrated the quite dramatic changes that occur over this period (Ouyang et al., 2019). It's difficult to see how you could combine datasets that have the transient anisotropic features in cortex (early) and more isotropic mature features (late) without testing on a hold out. Linear age adjustment doesn't really help in this context as the changes are nonlinear. As an example, the positive and negative scores at both ends of the tails of the Bailey-III cognitive scores are preterm and term born infants respectively which again makes me worry about the confound of age in the results of the regression.

– The average assessment score is nearly a full standard deviation below what you'd expect in a typical population at the same age, averaging 85-90 in the three subscales. Is there a reason the average is so low?

– The predicted scores are also in a very restricted range – although the raw scores range from 65-110 (cognitive) and 55-110 (language) the predictions seem to only range from 80-90. As you move away from the mean, the absolute error increases substantially.

– More generally, the analysis provides a useful proof-of-principle that early FA measures can predict later outcomes, but the predictions are not sufficiently accurate for clinical use. The MAE for the models is around 1 SD of the cognitive scores, and the accuracy using performance cut-offs was 61%/76%. Although sensitivity/specificity are not reported, FA does not seem to be a highly sensitive marker of these outcomes (as suggested in text). This approach may work at a population level but is not particularly effective for meaningfully predicting the outcome for a single individual. A more measured tone in describing the findings and their limitations would provide a more balanced account of the findings.

– With samples sizes this small there is substantial risk of overestimation of the accuracy of any machine learning techniques. The sample size is small, but 5-fold or 10-fold CV is possible here – see Poldrack et al. 10.1001/jamapsychiatry.2019.3671

– White matter FA showed a similar prediction accuracy to cortical FA. This raises questions about specificity. Could simpler measures (e.g., T1 or T2 signal) provide comparable prediction? If so, this may challenge the biological interpretation that the prediction derives from FA's capacity to measure cortical microstructure. Perhaps the authors could further examine this issue by comparing the relative prediction efficacy of simpler and FA-derived measures and looking at how strongly cortical and white matter measures correlate with each other. Is there anything to be gained by combining them?

– Please clarify how motion corruption of the DWI data was assessed and determined?

– Please explain reasons for participant dropout over the 2 year follow-up.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

Thank you for resubmitting your work entitled "Diffusion-MRI-based regional cortical microstructure at birth for predicting neurodevelopmental outcomes of 2-year-olds" for further consideration by eLife. Your revised article has been evaluated by Richard Ivry (Senior Editor) and a Reviewing Editor.

The manuscript has been improved but there are some remaining issues that need to be addressed before acceptance, as outlined below. We emphasise that these issues must be addressed to the satisfaction of the reviewers before the manuscript can be accepted.

The first major concern is that the motion threshold is very large (>3 times voxel dimensions) and the replacement scan seems to be taken from a different session. The eddy correction approach is also out of date and seems to just be an affine registration (no outlier rejection like in tortoise, eddy, shard). The authors need to comprehensively demonstrate that the results are not driven by motion-related artefact.

Second, it may not be appropriate to classify prematurely-born infants as "normal." Although the MRI may appear normal at early neurodevelopmental follow up, many cognitive issues aren't detectable until these children reach school age. This should be identified as a weakness in the Discussion.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

Thank you for sending your article entitled "Diffusion-MRI-based regional cortical microstructure at birth for predicting neurodevelopmental outcomes of 2-year-olds" for peer review at eLife. Your article is being evaluated by three peer reviewers, and the evaluation is being overseen by a Reviewing Editor and Richard Ivry as the Senior Editor.

The reviewers feel that the issue of head motion has not been fully addressed. The reviewers requested that you comprehensively show that motion cannot explain the findings. Showing the distribution of motion estimates in the sample is not sufficient. For publication, we would require a stronger demonstration that motion does not contaminate or confound the predictions by, for example, examining correlations between motion estimates (e.g., framewise displacement) and the outcome/connectivity measures used in the analysis.

Essential revisions

– We are concerned about the representativeness and size of the sample, and therefore how generalisable these results might be to the larger neonatal population. The infants are both born and scanned across a very wide age range, 32-42 postnatal weeks. This means that some of the infants were both very preterm at birth and at scan, with very different cortical microstructure (seen in the supplementary figures). There is also very different partial voluming in these two ages simply due to brain volume differences. Moreover, the study population is not described very well. We cannot discern what the relative proportions of preterm and term infants are. The authors only report that the gestational age ranged from 26 to 41 weeks and that images were obtained at postmenstrual ages ranging from 31 to 41 weeks. More detail is needed regarding the study population and who was imaged when. In addition, more than half of the recruited infants didn't finish the study, and it is important to know if there were differences between those subjects who were apparently lost to follow up and those that weren't.

We thank the reviewer for this comment. We acknowledge that a study with larger sample size may further improve generalization of the proposed model. On the other hand, as elaborated in the response to general comment, we would think a very wide age range will benefit generalization given that these neonates are free from perinatal brain injuries. The model can be applied to future in utero MRI where wide age range could be more likely. In the present study, the statistical analyses demonstrated no significant difference in any of the outcome scores, namely cognitive, language and motor scores, between the preterm and term infants (all p>0.3 in the panel (b) of updated Figure 1—figure supplement 3).

We are very aware of the partial volume effects in measuring cortical microstructure and have developed robust method to measure the cortical fractional anisotropy (FA) at the core of the cortical mantle, as demonstrated in added Figure 1—figure supplement 4. This method has been successfully and consistently implemented in our previous studies (e.g. Ouyang et al., 2019; Yu et al., 2016) too. Details of measuring cortical FA at the core (or skeleton) of the cortical ribbon can be found in the subsection “Measurement of cortical microstructure with brain MRI at birth” in the Materials and methods section. Briefly, skeleton voxels or “core” of the cerebral cortical mantle were obtained. Irregularly small yet significant offsets between the cortical skeleton and subject’s cerebral cortex were widespread, due to imperfect inter-subject registration from transforming cortical skeletons to individual brains. Such offsets were addressed by using a fast marching technique (details in our publication, Jeon et al., 2012; Ouyang et al., 2019). With fast marching, FA at the voxels with highest gray matter tissue probability from segmentation of an individual subject, namely, “core” cortical voxels, will be assigned to skeleton voxels sometimes deviating from “core” (Figure 1—figure supplement 4-a5, b5). As can be appreciated from Figure 1—figure supplement 4 using same scale to display a preterm brain at 33 postmenstrual weeks (PMW) and a term born brain at 41PMW, there are almost no significant differences of partial volume effects on cortical FA measurements between preterm and term-born brains.

Supplementary file 1 has been updated to describe subdivided preterm and term born populations. Supplementary file 2 has also been updated to demonstrate no significant correlation between the neurodevelopmental outcome and age (birth age or scan age) in subdivided preterm or term born populations. We intended to follow up all subjects undergoing MRI around birth and conduct neurodevelopmental assessment at their 2 years of age. Study staff worked diligently for that. They made biannual phone calls and sent birthday cards to stay in contact with families who showed interests to be contacted about follow-up study to ensure contact information remained current. All subjects who did not participate 2-year follow-up were those lost to follow up. Relatively moderate follow-up rate might be related that neither recruited preterm nor term born neonates had any perinatal brain injury or were clinically indicated. Motivation of participating 2-year follow-up for these parents may not be as high as parents of infants scanned due to clinical indication in other literature.

– The mixing of term and preterm groups in the analysis raises an interpretation issue. These two populations aren't equivalent, and few would argue that preterm infants are "normal" despite having normal conventional neuroimaging studies. It is conceivable that the findings of the study are driven by abnormalities in the preterm infants, and are thereby an indication of areas which are most commonly injured in preterm infants. We suggest the authors consider i) comparing findings of preterm with term infants and ii) evaluating outcome correlations for these populations separately. An alternative would be to examine term-equivalent scans for all participants.

Please see responses to the general comments. No injury was found in any preterm infants. None of the preterm infants were recruited due to clinical indication like most of relevant literature. Except preterm, the preterm infants were considered healthy and recruited solely for research purpose. All participating neonates were carefully screened by a neonatologist and a pediatric neuroradiologist to ensure “normality”. This study was also designed to make the interval between the birth and scan time as short as possible to limit the effects of exposure to the extrauterine environment. Furthermore, following reviewers’ suggestion, we compared the neurodevelopmental outcomes between preterm and term infants and found no statistically significant difference between these two populations (all p>0.3), shown in Figure 1—figure supplement 3B. We also evaluated outcome correlations with different age indices (i.e. birth age, MRI scan age and Bayley-III exam age) for preterm and term born infants separately and found no significant correlations (all p>0.1) in updated Supplementary file 2. Considering all factors above, it does not seem necessary to divide the neonates into preterm and term born populations for prediction analysis. Limited total sample size also prevented us from further subdividing the sample into two groups for prediction analysis. As we do not have term-equivalent scans for preterm neonates (exactly for limiting the effects of exposure to the extrauterine environment), we cannot exam term-equivalent scans for all participants.

– Please clarify the extent to which the correlations between real and predicted scores driven by extreme scorers.

We thank the reviewer for this comment. The correlations between real and predicted scores should not be driven by extreme scorers. In the present study we used mean absolute error (MAE) as a metric to measure prediction errors of our method. MAE is more robust to extreme/outlier values in the data compared to other prediction errors such as mean square error and root mean squared error (Willmott and Matsuura, 2005). This clarification has been added to the second to the last paragraph in the Discussion section.

To further confirm our results were not significantly affected by extreme scorers, we removed the smallest score values in the cognitive and language and re-ran correlation analysis between real and predicted scores. We found that all correlations remain significant (MAE = 5.74, r=0.496, p = 5x10^-4 for cognitive and MAE = 6.95, r = 0.461, p = 1.5x10^-3 for language).

– The authors have demonstrated the quite dramatic changes that occur over this period (Ouyang et al., 2019). It's difficult to see how you could combine datasets that have the transient anisotropic features in cortex (early) and more isotropic mature features (late) without testing on a hold out. Linear age adjustment doesn't really help in this context as the changes are nonlinear. As an example, the positive and negative scores at both ends of the tails of the Bailey-III cognitive scores are preterm and term born infants respectively which again makes me worry about the confound of age in the results of the regression.

We thank the reviewer for this comment. We agree that changes of cortical FA are not linear across the entire studied period, but rather biphasic piecewise linear with an inflection point at 36PMW (Ouyang et al., 2019). To better correct the age effect in cortical FA features in this revised manuscript, we adopted the same method to quantify cortical FA time courses (more details in Ouyang et al., 2019). The biphasic piecewise linear model below was used to fit the cortical FA of each gyrus or each parcellated region of interest (ROI) and age during 31-36PMW and 36-42PMW:

Equation 1 31-36 week ${\mathrm{\text{FA}}}_{i,j}={\beta }_{1,i}+{\beta }_{2,i}{t}_j+{\delta }_{i,j}$

Equation 2 36-42 week ${\mathrm{\text{FA}}}_{i,j}={\beta }_{3,i}+{\beta }_{4,i}{t}_j+{\gamma }_{i,j}$

Where t_j is the post-menstrual age of the jth subject; β_1,i, β_3,i are the intercepts and β_2,i, and β_4,I are the slopes for FA of the $i$th cortical ROI; $\delta$_i,j and $\gamma$_i,j are the error term for cortical FA measurement from 31-36PMW and from 36-42PMW, respectively. We corrected the age effect of cortical FA features of each gyrus or each parcellated ROI, ${{\mathrm{\text{FA}}}_{i,j}}_{\mathrm{\text{correct}}}$, based on above biphasic piecewise linear model, following the age correction methods described in the literature (Dukart et al., 2011):

31-36 week ${{\mathrm{\text{FA}}}_{i,j}}_{\mathrm{\text{correct}}}={\mathrm{\text{FA}}}_{i,j}+{\beta }_{2,i}*(36-t_j)$

36-42 week ${{\mathrm{\text{FA}}}_{i,j}}_{\mathrm{\text{correct}}}={\mathrm{\text{FA}}}_{i,j}+{\beta }_{4,i}*(36-t_j)$

After above-mentioned adjustment for the age effect, correlation between the predicted and actual cognitive or language scores is still significant (p<0.05) with original parcellation of 52 cortical regions and across other tested cortical parcellation schemes (128, 256, 512 and 1024 cortical parcels). Prediction performances before and after adjustment for the age effect are demonstrated in the updated Figure 2—figure supplement 1B-1C.

– The average assessment score is nearly a full standard deviation below what you'd expect in a typical population at the same age, averaging 85-90 in the three subscales. Is there a reason the average is so low?

The slightly lower average score may be related that samples are not diversified or large enough. All subjects are from a single county’s public hospital and recruitment was within 3 years. We would also make it clear that in this study average composite scores 85.7 – 91.2 in the three subscales are completely within 1 standard deviation from Bayley-III norms (mean = 100; standard deviation = 15). These subjects are still considered normal (Bayley, 2006).

– The predicted scores are also in a very restricted range – although the raw scores range from 65-110 (cognitive) and 55-110 (language) the predictions seem to only range from 80-90. As you move away from the mean, the absolute error increases substantially.

Restricted range of predicted score probably is mainly driven by restricted range of actual score. Although the actual score range is relatively wide, most of actual scores are clustered in 80 to 100 (Figure 1—figure supplement 3). Restricted predicted score range is also related to cortical parcellation scheme (Figure 2—figure supplement 1A). Especially using parcellation scheme with 128 random cortical parcels yields wide range of predicted scores. In addition, combining regional cortical FA and white matter FA measures in prediction (suggested by Essential revisions #9 below) also yields a wider range in predicted value. All these results suggest that larger individual variability of neurodevelopmental scores from a larger sample and more comprehensive information from multimodal neuroimaging (also see response to Essential revisions #9) will improve prediction performance and reduce prediction errors. Such discussion has already been included in the Discussion section.

– More generally, the analysis provides a useful proof-of-principle that early FA measures can predict later outcomes, but the predictions are not sufficiently accurate for clinical use. The MAE for the models is around 1 SD of the cognitive scores, and the accuracy using performance cut-offs was 61%/76%. Although sensitivity/specificity are not reported, FA does not seem to be a highly sensitive marker of these outcomes (as suggested in text). This approach may work at a population level but is not particularly effective for meaningfully predicting the outcome for a single individual. A more measured tone in describing the findings and their limitations would provide a more balanced account of the findings.

We agree with the reviewer that before the predictions can be sufficiently accurate for clinical use, testing with a larger sample is needed. Sensitivity/specificity has been provided in Figure 2—figure supplement 2. We also agree that the approach shows effectiveness of prediction at the level of certain subgroups, but is not effective enough for accurate prediction of the outcome at the level of a single individual. As suggested by the reviewer, to provide a more balanced account of the finding, we have toned down the discussion significantly in the first and second to the last paragraph in the Discussion section.

– With samples sizes this small there is substantial risk of overestimation of the accuracy of any machine learning techniques. The sample size is small, but 5-fold or 10-fold CV is possible here – see Poldrack et al. 10.1001/jamapsychiatry.2019.3671

We thank the reviewer for this comment. Although k-fold cross-validation (CV) is ideal and can be applied to this prediction model, leave-one-out cross validation approach has also been widely used in studies reviewed in Poldrack et al., 2020. From Figure 3A in Polrack et al., 2020, number of leave-one-out studies was similar to number of k-fold studies.

As elaborated in response to Essential revision #6, #7 and #9, this manuscript emphasizes the capability of cortical microstructure measures of predicting future neurodevelopmental outcomes, not perfect prediction itself, as cortical microstructure measures have been rarely used for prediction. Following reviewer’s suggestion, we conducted a 3-fold cross validation by using cortical FA only, white matter (WM) FA only, and combined cortical and WM FA. As shown in Author response table 1, with combined cortical and WM FA, the predicted cognitive and language scores remained significantly correlated with the actual scores (all p<0.05) for all 3 folds, consistent to better prediction of combined cortical and WM FA in revised Figure 4B and 4C. Although using cortical FA or WM FA only did not pass the 3-fold CV, cortical FA features performed slightly better than WM FA features in terms of correlation coefficients. We have added these findings in the last paragraph of the Results section and the second paragraph of the Discussion section.

– White matter FA showed a similar prediction accuracy to cortical FA. This raises questions about specificity. Could simpler measures (e.g., T1 or T2 signal) provide comparable prediction? If so, this may challenge the biological interpretation that the prediction derives from FA's capacity to measure cortical microstructure. Perhaps the authors could further examine this issue by comparing the relative prediction efficacy of simpler and FA-derived measures and looking at how strongly cortical and white matter measures correlate with each other. Is there anything to be gained by combining them?

We thank the reviewer for raising this question. It is our understanding that any measurement reflecting individual brain maturational level including T1 or T2 signal, white matter microstructure, could possibly be used for predicting neurodevelopmental outcome. The main purpose of this manuscript is not to claim that cortical FA is the “only” index for predicting neurodevelopmental outcome. Instead, we consider revealing capability of cortical FA of predicting neurodevelopmental outcome (probably for the first time) the main and novel contribution of this manuscript. These different brain measures reflect distinct brain developmental properties. For example, simpler measures from structural T1-weighted (T1w) or T2-weighted (T2w) images usually indicate macrostructural measurements such as cortical volume or thickness. White matter FA measures indicate white matter (not directly cerebral cortical) microstructure, although they might be related to cortical FA measures.

Following reviewer’s comment, we combined cortical and white matter FA measures to predict outcomes with results shown in Figure 4B and 4C. Higher performances were achieved in predicting outcomes by using combined features, demonstrated by higher correlation coefficient values between the predicted and actual scores in Figure 4B and 4C. We also added discussion about combining features to improve prediction performance in the Results and Discussion sections.

– Please clarify how motion corruption of the DWI data was assessed and determined?

To quantify head motion in each dMRI scan, all diffusion weighted image (DWI) volumes were aligned to the first stable image volume in the scan using automatic image registration (AIR) in DTIStudio (Jiang et al., 2006). The volume-by-volume translation and rotation from the registration were calculated. DWI volumes with translation measurement larger than 5 mm or rotation measurement larger than 5 degree was determined as corrupted volumes. With 30 scanned diffusion weighted image (DWI) volumes and 2 repetitions, we accepted those scanned diffusion MRI datasets with less than 5 DWI volumes affected by motion. The affected volumes were replaced by the good volumes of another DTI repetition during postprocessing. We have added these details in the Materials and methods section.

– Please explain reasons for participant dropout over the 2 year follow-up.

As elaborated in the response to general comment and Essential revision 2, we intended to follow up all subjects undergoing MRI around birth and conduct neurodevelopmental assessment at their 2 years of age. All subjects who did not participate 2-year follow-up were those lost to follow up. Relatively moderate follow-up rate might be related that neither recruited preterm nor term born neonates had any perinatal brain injury or were clinically indicated. Motivation of participating 2-year follow-up for these parents may not be as high as parents of infants scanned due to clinical indication in other literature.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

The first major concern is that the motion threshold is very large (>3 times voxel dimensions) and the replacement scan seems to be taken from a different session. The eddy correction approach is also out of date and seems to just be an affine registration (no outlier rejection like in tortoise, eddy, shard). The authors need to comprehensively demonstrate that the results are not driven by motion-related artefact.

We thank the reviewer for this comment. The second diffusion MRI (dMRI) scan is immediately after the first dMRI scan. The affected dMRI volumes were replaced by the good volumes from another dMRI scan in the same session. As all dMRI scans were conducted during neonates’ natural sleep, in general the motion was very small during scans. Motion threshold of 5 mm or 5 degrees was used only for discarding diffusion volumes with large motions due to occasional abrupt movements during sleep. The distribution of motion measurements including translation and rotation of dMRI scans in this cohort of 46 neonates who had a neurodevelopmental assessment at 2 years of age was demonstrated in Figure 1—figure supplement 5. The volume-by-volume translation range is 0 to 1.6mm with average 0.78mm and most of translations less than 1mm. The volume-by-volume rotation range is 0 to 0.7 degrees with average of 0.18 degrees and most of rotations less than 0.3 degrees. The information above was added to Materials and methods section.

In Author response image 1, axial b0 image from raw dMRI of a representative neonate brain in this study and that of a 3-year-old child brain scanned with the same imaging protocol are shown. Yellow arrows highlight relatively mild distortion of b0 image (left) of the neonate brain and severe distortion of b0 image (right) of the 3-year-old brain. Both b0 images are before eddy current correction and at the level of eyeball. To correct for the small head motion and eddy current distortions, we only used the 12-parameter (affine) registration by registering all the diffusion weighted images to the b0 image. Although we did not use automatic outlier rejection methods (e.g. Tortoise, eddy mentioned by the reviewer), we manually inspected all image slices and volumes and found no outliers due to severe eddy current distortions. Meanwhile, since no dMRI with opposite phase encoding directions was acquired, we could not apply advanced eddy current distortion correction technique requiring dMRI of both AP (anterior-posterior) and PA (posterior-anterior) directions.

Author response image 1

Download asset Open asset

Second, it may not be appropriate to classify prematurely-born infants as "normal." Although the MRI may appear normal at early neurodevelopmental follow up, many cognitive issues aren't detectable until these children reach school age. This should be identified as a weakness in the Discussion.

Following reviewer’s suggestion, we have removed “normal” in the first paragraph in the Results and Materials and methods section and added discussion to identify inclusion of prematurely-born infants as a limitation in the Discussion section.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

The reviewers feel that the issue of head motion has not been fully addressed. The reviewers requested that you comprehensively show that motion cannot explain the findings. Showing the distribution of motion estimates in the sample is not sufficient. For publication, we would require a stronger demonstration that motion does not contaminate or confound the predictions by, for example, examining correlations between motion estimates (e.g., framewise displacement) and the outcome/connectivity measures used in the analysis.

To comprehensively demonstrate our prediction results were not driven by motion-related artefact, we examined the correlation between motion (quantified by mean framewise displacement suggested in the Editor’s letter) and regional fractional anisotropy (FA) values from cerebral cortex and white matter (Figure 4—figure supplement 1a). After false discovery rate (FDR) correction, no significant correlation between motion estimates and FA values of any of 92 brain regions was found. We also investigated motion effects on prediction results and found that high performance of prediction models was not affected after statistically controlling for motion estimates in cortical and white matter FA measures in the predication models, as shown in Figure 4—figure supplement 1b. Figure 4—figure supplement 1A and 1B collectively demonstrated that motion did not contaminate or confound the significant predication found in this study. The response above has been added at the end of Results section.

Author details

1. Minhui Ouyang

   Radiology Research, Children’s Hospital of Philadelphia, Philadelphia, United States

   Contribution

   Conceptualization, Resources, Data curation, Software, Formal analysis, Validation, Investigation, Visualization, Methodology, Writing - original draft, Writing - review and editing

   Competing interests

   No competing interests declared

2. Qinmu Peng

   1. Radiology Research, Children’s Hospital of Philadelphia, Philadelphia, United States
   2. Department of Radiology, Perelman School of Medicine, University of Pennsylvania, Philadelphia, United States

   Contribution

   Software, Validation, Methodology, Writing - review and editing

   Competing interests

   No competing interests declared

3. Tina Jeon

   Radiology Research, Children’s Hospital of Philadelphia, Philadelphia, United States

   Contribution

   Data curation, Formal analysis, Investigation, Project administration, Writing - review and editing

   Competing interests

   No competing interests declared

4. Roy Heyne

   Department of Pediatrics, University of Texas Southwestern Medical Center, Dallas, United States

   Contribution

   Resources, Data curation, Project administration, Writing - review and editing

   Competing interests

   No competing interests declared

5. Lina Chalak

   Department of Pediatrics, University of Texas Southwestern Medical Center, Dallas, United States

   Contribution

   Resources, Data curation, Project administration, Writing - review and editing

   Competing interests

   No competing interests declared

6. Hao Huang

   1. Radiology Research, Children’s Hospital of Philadelphia, Philadelphia, United States
   2. Department of Radiology, Perelman School of Medicine, University of Pennsylvania, Philadelphia, United States

   Contribution

   Conceptualization, Resources, Data curation, Formal analysis, Supervision, Funding acquisition, Validation, Investigation, Visualization, Methodology, Writing - original draft, Project administration, Writing - review and editing

   For correspondence

   huangh6@email.chop.edu

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0002-9103-4382

• 1,296

  Page views

• 171

  Downloads

• 12

  Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.