Early Prediction of Long-term Cognitive Development Using Multimodal MRI in Infants Born Preterm
Poster No:
1382
Submission Type:
Abstract Submission
Authors:
Nehal Parikh1, Mekibib Altaye2, Armin Allahverdy2, Hailong Li2, Beth Kline-Fath2, Weihong Yuan2, Abiot Yenealem Derbie1, Lili He2, Leanne Tamm2
Institutions:
1Cincinnati Children's Hospital, CINCINNATI, OH, 2CINCINNATI CHIILDREN'S HOSPITAL, Cincinnati, OH
First Author:
Co-Author(s):
Introduction:
Cognitive impairment remains the most common long-term adverse outcome following preterm birth. Accurate diagnosis usually takes 3 to 5 years during which time we are missing an early critical window of neuroplasticity. Detection during this window would allow for early targeted interventions to enhance therapeutic efficacy. Our goal was to improve early prediction of cognitive development by term-equivalent age (TEA) by exploiting features from brain morphometry, structural connectivity (SC), and functional connectivity (FC) derived from structural MRI, diffusion MRI (dMRI), and resting state functional MRI (rsfMRI), respectively.
Methods:
We studied a multisite regional cohort of 358 very preterm (VPT) infants born at or below 32 weeks' gestational age from 5 Southwest Ohio NICUs (Cincinnati Infant Neurodevelopment Early Prediction [CINEPS] cohort). All infants were imaged at Cincinnati Children's Hospital between 39 and 44 weeks postmenstrual age on the same 3T Philips scanner and 32-channel receiver head coil with the following identical sequences: dMRI: TE: 88 msec, TR 6972 msec, FA 90°, resolution 2×2×2 mm3, 36 directions; b-value 800 s/mm2, MB factor 2; rsfMRI: TE: 45 msec, TR 893 msec, FA 90°, resolution 2.5×2.5×2.5 mm3 ,400 volumes, MB factor 4; Axial T2w: TE 166 msec, TR 8,300 msec, FA 90°, resolution 1×1×1 mm3. Cognitive development was assessed with the Differential Ability Scales (2nd Edition) General Conceptual Ability (GCA) score at 3 years corrected age. We used established pre- and post-processing pipelines and neonatal brain atlases from the Developing Human Connectome Project (dHCP) to generate brain volumes, cortical maturation metrics, SC and FC as previously described (Bastiani M. 2019; Kline JE, 2020; Kline JE, 2021). Missing data was handled via k-nearest neighbor imputation. We used two unsupervised approaches for feature selection/reduction from the nearly 20,000 MRI predictor variables: CONN, which generated six graph theory measures per modality for each of the 81 dHCP atlas regions of interest and non-negative matrix factorization (NMF), which decomposed these measures to 26 morphometry, 28 FC and 30 SC network components. Global efficiency from SC and FC and postmenstrual age at MRI scan were modeled separately. We selected one conventional MRI (cMRI; global brain abnormality score) and 10 known clinical predictors of cognitive development a priori (Table 1). Last, we applied support vector machine (SVM) to develop a multimodal model that included the above 86 independent variables to predict the GCA score. We created 500 bootstrap datasets to evaluate model performance and correct for over-optimism per the TRIPOD guidelines (Moons KGM, 2015). To assess model fit, we calculated optimism-corrected values for R2, root mean square error (RMSE), and difference between observed and predicted scores.
Results:
The mean (SD) gestational age was 29.3 (2.5) and GCA score was 93.7 (20.0) for the 314 infants (88%) with follow-up data at or after 3 years CA (Table 1A). The 11 a priori selected clinical + cMRI predictors (Table 1) did poorly in predicting the DAS GCA score: R2 24.0%. The combined multimodal model achieved the highest accuracy with an optimism-corrected R2 value of 64.3% (95% CI: 56.3, 72.3) (Table 1B). The predicted cognitive scores of the multimodal model were closely aligned with the observed scores (Fig. 1A); Approximately two-thirds of the predicted GCA scores were within +/- 6 points (1 SD) of the observed scores (Fig. 1B).
·Figure 1. A) Scatter plot of observed versus predicted cognitive scores (mean scores after bootstrap and optimism correction) and B) Histogram of difference between observed and predicted scores.
Conclusions:
In a regional prospective cohort of VPT infants, we used multimodal advanced neuroimaging and machine learning to enable early prediction of long-term cognitive development. Our study generated data that is considerably higher in accuracy than current prognostic models. We aim to use this internally validated model to enable targeted risk stratification for interventions that are designed to enhance cognitive development in high-risk VPT infants.
Disorders of the Nervous System:
Neurodevelopmental/ Early Life (eg. ADHD, autism) 2
Higher Cognitive Functions:
Higher Cognitive Functions Other
Modeling and Analysis Methods:
Classification and Predictive Modeling 1
Connectivity (eg. functional, effective, structural)
Keywords:
Cognition
Development
FUNCTIONAL MRI
Machine Learning
Modeling
Multivariate
PEDIATRIC
STRUCTURAL MRI
WHITE MATTER IMAGING - DTI, HARDI, DSI, ETC
1|2Indicates the priority used for review
Provide references using author date format
Kidokoro H. (2013). 'New MR imaging assessment tool to define brain abnormalities in very preterm infants at term.' American Journal of Neuroradiology, vol. 34, no. 11, pp. 2208-2214
Kline JE. (2020). 'Automated brain morphometric biomarkers from MRI at term predict motor development in very preterm infants.' Neuroimage Clinical. vol 28, no. 102475.
Kline JE. (2021). 'Association Between Brain Structural Network Efficiency at Term-Equivalent Age and Early Development of Cerebral Palsy in Very Preterm Infants.' Neuroimage. vol 245, no.118688
Moons KGM (2015). 'Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD): Explanation and Elaboration'. Annals of Internal Medicine, vol. 162, pp. W1-W73.