Network-BrainAGE: A New Metric to Assess Network-Based Brain Aging
Poster No:
1198
Submission Type:
Abstract Submission
Authors:
Shalaila Haas1,2, Yuetong Yu3, Hao-Qi Cui3, Faye New1, RUIYANG GE3, Sophia Frangou3
Institutions:
1Icahn School of Medicine at Mount Sinai, New York, NY, 2Tri-institutional Center for Translational Research in Neuroimaging and Data Science (TReNDS), Atlanta, GA, 3University of British Columbia, Vancouver, British Columbia
First Author:
Shalaila Haas, PhD
Icahn School of Medicine at Mount Sinai|Tri-institutional Center for Translational Research in Neuroimaging and Data Science (TReNDS)
New York, NY|Atlanta, GA
Icahn School of Medicine at Mount Sinai|Tri-institutional Center for Translational Research in Neuroimaging and Data Science (TReNDS)
New York, NY|Atlanta, GA
Co-Author(s):
Introduction:
Extensive research has established the difference between neuroimaging-predicted and actual chronological age (brainAGE), as a robust and biologically meaningful measure of brain health (Franke et al., 2019). We have previously released brainAGE models in development (Modabbernia et al., 2022) and across the lifespan (Yu et al., 2023). However, brainAGE is generally computed as a global index of age-related brain changes, disregarding information about spatial variation. In this study, we expand on our previous work to develop large-scale brainAGE models for specialized brain networks in healthy individuals across the lifespan.
Methods:
The process for generating network-based brainAGE will follow our previous work (Modabbernia, 2022). The T1w structural images will be processed using standard pipelines in the Freesurfer software (version 7.1) in 9473 healthy individuals (aged 3-92 years; mean [SD] = 24.89[20.49]; N [%] female = 5126 [54.11%]) across 9 datasets: Adolescent Brain Cognitive Development (N = 3779); Pediatric Imaging, Neurocognition, and Genetics (N = 736); Brain Genomics Superstruct (N = 1570); 1000 Functional Connectomes (N = 1000); Brain Imaging Consortium (N = 203); Southwest University Adult Lifespan Dataset (N = 494); Information eXtraction from Images (N = 563); Cambridge Centre for Ageing and Neuroscience (650); and Australian Imaging Biomarkers and Lifestyle Study of Ageing datasets (N = 478). Subsequently, the Schaefer atlas was used to generate parcels (200, 400, 600, 800, 1000) of cortical thickness and cortical surface area, each assigned to one of the 7 Yeo networks (Yeo et al., 2011). Sex-specific pooled datasets were split into a training (80%) and testing (20%) subsets. Model performance was assessed by mean-absolute-error, root-mean-squared-error, and the correlation between predicted and chronological age in the test set after age-related bias adjustment.
Results:
For all networks except the visual network, the mean-absolute-error reduced with increasing number of parcels (ranges across all: visual = 6.00 - 6.70; somatomotor = 5.00 - 5.67; dorsal-attention = 6.91 - 7.32; salience = 5.31 - 5.92; limbic = 5.82 - 6.65; control = 5.46 - 6.20; default = 4.89 - 5.51). For all networks, the root-mean-squared-error reduced with increasing number of parcels (ranges across all: visual = 8.03 - 9.32; somatomotor = 6.71 - 7.97; dorsal-attention = 9.22 - 9.86; salience = 7.43 - 8.41; limbic = 8.00 - 9.27; control = 7.63 - 8.68; default = 6.73 - 7.83). The correlation between predicted and chronological age was high across all networks (mean correlation across networks = 0.93) and varied by maximally 0.02 with increasing parcels. The results for the models using the 1000 parcels of the Schaefer atlas are presented in Figure 1. Correspondence between males and females for mean-absolute-error across networks was high (>0.97).
Conclusions:
This study introduces novel network-specific brainAGE models indicating variability in model performance across specialized networks. These models offers potential applications in clinical samples that might demonstrate stronger associations with cognitive deficits, psychopathology, and risk factors.
Lifespan Development:
Aging 1
Early life, Adolescence, Aging
Modeling and Analysis Methods:
Classification and Predictive Modeling 2
Multivariate Approaches
Novel Imaging Acquisition Methods:
Anatomical MRI
Keywords:
Aging
Machine Learning
Modeling
Multivariate
STRUCTURAL MRI
1|2Indicates the priority used for review
Provide references using author date format
Modabbernia, A. (2022), 'Systematic evaluation of machine learning algorithms for neuroanatomically-based age prediction in youth', Human brain mapping, vol. 43, no.17, pp. 5126–5140.
Yeo, B. T. (2011), 'The organization of the human cerebral cortex estimated by intrinsic functional connectivity', Journal of neurophysiology, vol. 106, no. 3, pp. 1125–1165.
Yu, Y. (2023), Brain-Age Prediction: Systematic Evaluation of Site Effects, and Sample Age Range and Size. bioRxiv, 2023.11.06.565917.