Neonatal network topography predicts brain maturation and neurodevelopmental outcomes

1720 

Abstract Submission 

Jianlong Zhao^1,2,3, Tengda Zhao^1,2,3, Yuehua Xu^1,2,3, Hongming Li^4,5, Lianglong Sun^1,2,3, Xinyuan Liang^1,2,3, Meizhen Han^1,2,3, Zilong Zeng^1,2,3, Yong He^1,2,3,6

¹State Key Laboratory of Cognitive Neuroscience and Learning, Beijing Normal University, Beijing, China, ²Beijing Key Laboratory of Brain Imaging and Connectomics, Beijing Normal University, Beijing, China, ³IDG/McGovern Institute for Brain Research, Beijing Normal University, Beijing, China, ⁴Center for Biomedical Image Computing and Analytics, University of Pennsylvania, Philadelphia, PA, ⁵Department of Radiology, University of Pennsylvania, Philadelphia, PA, ⁶Chinese Institute for Brain Research, Beijing, China

Jianlong Zhao  
State Key Laboratory of Cognitive Neuroscience and Learning, Beijing Normal University|Beijing Key Laboratory of Brain Imaging and Connectomics, Beijing Normal University|IDG/McGovern Institute for Brain Research, Beijing Normal University
Beijing, China|Beijing, China|Beijing, China

Tengda Zhao  
State Key Laboratory of Cognitive Neuroscience and Learning, Beijing Normal University|Beijing Key Laboratory of Brain Imaging and Connectomics, Beijing Normal University|IDG/McGovern Institute for Brain Research, Beijing Normal University
Beijing, China|Beijing, China|Beijing, China

Yuehua Xu  
State Key Laboratory of Cognitive Neuroscience and Learning, Beijing Normal University|Beijing Key Laboratory of Brain Imaging and Connectomics, Beijing Normal University|IDG/McGovern Institute for Brain Research, Beijing Normal University
Beijing, China|Beijing, China|Beijing, China

Hongming Li  
Center for Biomedical Image Computing and Analytics, University of Pennsylvania|Department of Radiology, University of Pennsylvania
Philadelphia, PA|Philadelphia, PA

Lianglong Sun  
State Key Laboratory of Cognitive Neuroscience and Learning, Beijing Normal University|Beijing Key Laboratory of Brain Imaging and Connectomics, Beijing Normal University|IDG/McGovern Institute for Brain Research, Beijing Normal University
Beijing, China|Beijing, China|Beijing, China

Xinyuan Liang  
State Key Laboratory of Cognitive Neuroscience and Learning, Beijing Normal University|Beijing Key Laboratory of Brain Imaging and Connectomics, Beijing Normal University|IDG/McGovern Institute for Brain Research, Beijing Normal University
Beijing, China|Beijing, China|Beijing, China

Meizhen Han  
State Key Laboratory of Cognitive Neuroscience and Learning, Beijing Normal University|Beijing Key Laboratory of Brain Imaging and Connectomics, Beijing Normal University|IDG/McGovern Institute for Brain Research, Beijing Normal University
Beijing, China|Beijing, China|Beijing, China

Zilong Zeng  
State Key Laboratory of Cognitive Neuroscience and Learning, Beijing Normal University|Beijing Key Laboratory of Brain Imaging and Connectomics, Beijing Normal University|IDG/McGovern Institute for Brain Research, Beijing Normal University
Beijing, China|Beijing, China|Beijing, China

Yong He  
State Key Laboratory of Cognitive Neuroscience and Learning, Beijing Normal University|Beijing Key Laboratory of Brain Imaging and Connectomics, Beijing Normal University|IDG/McGovern Institute for Brain Research, Beijing Normal University|Chinese Institute for Brain Research
Beijing, China|Beijing, China|Beijing, China|Beijing, China

The neonatal human brain lays a critical neural foundation for establishing wide cognitive and behavioral abilities that last a lifetime[1-3]. However, how the individualized functional topography at birth evolves with age and supports later neurodevelopmental outcomes remain largely unclear. Here, we investigated the variability pattern of neonatal brain functional topography by combining individualized functional network and machine learning predictive models in neonatal large sample resting fMRI datasets.

Based on 367 term-born neonates from the Developing Human Connectome Project[4], we firstly employed a regularized NMF approach to derive individualized functional networks for each neonate [5](Fig. 1A). Then we quantify the overall variability of functional topography across subjects based on median absolute deviations (Fig. 1B) and a fine-grained description of variability within each functional network (position, size, overlap, and region homogeneity) (Fig. 1C). Finally, we proposed a network-based SVR approach to delineate how individual functional topography of neonatal brain could predict brain maturation and cognitive, language, and motor function at 18 months (Fig. 1D). The network-wise feature weights were evaluated to understand the network contribution for predicting brain maturity and various behavior abilities. Cognitive, language, and motor development were assessed using the Bayley Scales of Infant and Toddler Development, 3rd Edition (BSID-III).

We identified 11 functional networks for each neonate (Fig. 2A left). The highest between-subject variance is mainly located in association cortex while the lowest variance is located in primary cortex (Fig. 2A middle, right, Pspin < 0.05). The fine-grained description of variability pattern of the size, position, overlap, region homogeneity was similar to the voxel-level pattern (all index: Pspin < 0.05, Fig. 2B). However, there exist some indicator-specific distributions especially within the association system, such as the medial prefrontal cortex and temporal-parietal junction cortex. The network-based SVR model achieved a significant prediction for brain maturity and later cognition, language and motor performance at 18 months (age: r = 0.46; cognition: r = 0.39, language: r = 0.29, motor: r = 0.37; all Pperm < 0.001, Fig. 2C). To assess the effect of random fold assignment, we repeated this procedure 100 times by newly classifying all neonates into 10 subsets each time, which yielded highly consistent results (age: mean r = 0.51; cognition: mean r = 0.34; language: mean r = 0.27; motor: mean r = 0.34; all Pperm < 0.001, Fig. 2D). Finally, for predicting brain maturity and various behavior abilities, we found that high contributing networks are located consistently at primary motor networks and heterogeneous within association networks (Fig. 2E).

The results suggest that the variation pattern of neonatal primary networks is discernible from that of association networks. Critically, we demonstrated that individual variation of neonatal primary networks brain supports the accurate prediction of brain maturity and later cognition, language, and motor performance at 18 months. Individual variation of association networks contributes differentially to brain maturation and behavior prediction. Our results have implications for understanding how functional neuroanatomy at birth matures and supports the early establishment of neurodevelopmental outcomes.

Normal Brain Development: Fetus to Adolescence ²

Classification and Predictive Modeling

fMRI Connectivity and Network Modeling ¹

Brain Atlases

BOLD fMRI

FUNCTIONAL MRI

Machine Learning

Modeling

Other - neonates； development; parcellation; topography;