Functional Connectome Construction in Very Preterm Infants for Cognitive Biomarker Identification
Poster No:
1818
Submission Type:
Abstract Submission
Authors:
Junqi Wang1, Hailong Li1,2, Kim Cecil1,2, Mekibib Altaye1,2, Nehal Parikh1,2, Lili He1,2
Institutions:
1Cincinnati Children's Hospital Medical Center, Cincinnati, OH, 2University of Cincinnati, Cincinnati, OH
First Author:
Co-Author(s):
Hailong Li, PhD
Cincinnati Children's Hospital Medical Center|University of Cincinnati
Cincinnati, OH|Cincinnati, OH
Cincinnati Children's Hospital Medical Center|University of Cincinnati
Cincinnati, OH|Cincinnati, OH
Kim Cecil
Cincinnati Children's Hospital Medical Center|University of Cincinnati
Cincinnati, OH|Cincinnati, OH
Cincinnati Children's Hospital Medical Center|University of Cincinnati
Cincinnati, OH|Cincinnati, OH
Mekibib Altaye, PhD
Cincinnati Children's Hospital Medical Center|University of Cincinnati
Cincinnati, OH|Cincinnati, OH
Cincinnati Children's Hospital Medical Center|University of Cincinnati
Cincinnati, OH|Cincinnati, OH
Nehal Parikh, DO, MS
Cincinnati Children's Hospital Medical Center|University of Cincinnati
Cincinnati, OH|Cincinnati, OH
Cincinnati Children's Hospital Medical Center|University of Cincinnati
Cincinnati, OH|Cincinnati, OH
Lili He, PhD
Cincinnati Children's Hospital Medical Center|University of Cincinnati
Cincinnati, OH|Cincinnati, OH
Cincinnati Children's Hospital Medical Center|University of Cincinnati
Cincinnati, OH|Cincinnati, OH
Introduction:
Infants born very prematurely (<32 weeks, VPT infants) are at risk for cognitive deficits [1]. Early interventions within the first two years post-birth, when neuroplasticity is highest, can notably enhance cognitive outcomes in at-risk infants [2]. Studies have explored resting-state functional connectome (FC) biomarkers of cognitive deficits [3, 4]. However, most existing methods have difficulties tackling the high level of signal noise present in neonatal blood–oxygen-level-dependent (BOLD) signals. Brain structural connectome (SC) derived from diffusion tensor imaging (DTI) provides a stable fibrous representation of the brain. There are intrinsic relationships between the SC and FC [5]. Our work introduced an advanced graph learning model that uses SC for improved FC construction. The model, tested on the simulated dataset and a VPT infant cohort, is hypothesized to outperform existing methods in FC construction.
Methods:
The IRB-approved study involved 395 VPT infants from five Neonatal Intensive Care Units. All infants were imaged between 39-44 weeks postmenstrual age on a 3T MRI scanner during natural sleep without sedation. We administered standardized Bayley Scales of Infant and Toddler Development (Bayley-III) cognitive tests at the 2-year corrected age, categorizing subjects into low-risk (Bayley-III score > 85) and high-risk (Bayley-III score ≤ 85) for cognitive deficits. We processed rs-fMRI and DTI scans using the developing Human Connectome Project (dHCP) pipeline [6, 7] with an 82-region dHCP atlas [8]. We calculated the fractional anisotropy weighed connectome for each infant as the prior knowledge SC.
The smoothness property of BOLD signals on the FC has been reported and utilized to estimate the brain FC [9]. Based on the intrinsic coupling between SC and FC, we proposed a putative property that the intrinsic BOLD signals should be smooth on the SC as well. The proposed enhanced FC learning model regulated the intrinsic BOLD signals using the SC, which in turn facilitated the estimation of the deterministic FC. As illustrated in Figure 1, our model contains a Construction block and a Denoising block and is optimized following an alternative manner between the two blocks. where we iteratively performed FC estimation and signal denoising, incorporating the SC smoothness constraint in the process. The intermediate FCs were constructed from the intrinsic BOLD signals regulated by the SC and intermediate FC from the previous iteration. After multiple iterations, the converged FC from our model is considered optimized FC.
The smoothness property of BOLD signals on the FC has been reported and utilized to estimate the brain FC [9]. Based on the intrinsic coupling between SC and FC, we proposed a putative property that the intrinsic BOLD signals should be smooth on the SC as well. The proposed enhanced FC learning model regulated the intrinsic BOLD signals using the SC, which in turn facilitated the estimation of the deterministic FC. As illustrated in Figure 1, our model contains a Construction block and a Denoising block and is optimized following an alternative manner between the two blocks. where we iteratively performed FC estimation and signal denoising, incorporating the SC smoothness constraint in the process. The intermediate FCs were constructed from the intrinsic BOLD signals regulated by the SC and intermediate FC from the previous iteration. After multiple iterations, the converged FC from our model is considered optimized FC.
Results:
We simulated the ground truth SC, FC, and BOLD signals and then compared the accuracy of reconstructed FC using our proposed method, the radial basis function kernel (RBF) method, and the graph learning method that does not consider SC (Table 1). Our proposed method achieved superior reconstruction performance over competing methods. We further applied the proposed model to the VPT infant cohort. Based on the enhanced FC, we observed altered functional activation patterns in medial and inferior temporal gyri between low-risk and high-risk groups. These altered regions are not observed in FC constructed using traditional methods.
Conclusions:
We proposed a novel FC construction algorithm, validated with synthetic data, and applied it to a large cohort of VPT infants. This revealed distinct functional networks between low- and high-risk infants for cognitive deficits.
Disorders of the Nervous System:
Neurodevelopmental/ Early Life (eg. ADHD, autism)
Modeling and Analysis Methods:
fMRI Connectivity and Network Modeling 1
Neuroanatomy, Physiology, Metabolism and Neurotransmission:
Anatomy and Functional Systems 2
Neuroinformatics and Data Sharing:
Informatics Other
Keywords:
Design and Analysis
FUNCTIONAL MRI
Modeling
1|2Indicates the priority used for review
Provide references using author date format
2. Morgan, C. (2021) 'Early Intervention for Children Aged 0 to 2 Years With or at High Risk of Cerebral Palsy: International Clinical Practice Guideline Based on Systematic Reviews', JAMA Pediatrics, vol. 175, no. 8, pp. 846-858.
3. Shang, J. (2019), 'A machine learning investigation of volumetric and functional MRI abnormalities in adults born preterm', Humam Brain Mapping, vol. 40, no. 14, pp. 4239-4252.
4. Gozdas, E. (2018), 'Altered functional network connectivity in preterm infants: antecedents of cognitive and motor impairments?', Brain Structure and Function, vol. 223, no.8, pp. 3665-3680.
5. Fjell, A.M. (2017), 'Relationship between structural and functional connectivity change across the adult lifespan: A longitudinal investigation', Human Brain Mapping, vol. 38, no. 1, pp. 561-573.
6. Fitzgibbon, S.P. (2020), 'The developing Human Connectome Project (dHCP) automated resting-state functional processing framework for newborn infants', NeuroImage, vol. 223, pp. 117303.
7. Bastiani, M. (2019), 'Automated processing pipeline for neonatal diffusion MRI in the developing Human Connectome Project'. NeuroImage, vol. 185, pp. 750-763.
8. Makropoulos, A. (2018), 'The developing human connectome project: A minimal processing pipeline for neonatal cortical surface reconstruction', NeuroImage, vol. 173, pp. 88-112.
9. Wang, J. (2023), 'Dynamic weighted hypergraph convolutional network for brain functional connectome analysis', Medical Image Analysis, vol. 87, pp. 102828.