Intelligence brain networks of early adolescent: Waxing and waning
Poster No:
971
Submission Type:
Abstract Submission
Authors:
Jiadong Yan1, Yasser Iturria Medina1, Gleb Bezgin2, Alan Evans3, Sherif Karama4
Institutions:
1McGill University, Montreal, Quebec, 2Neuroinformatics for Personalized Medicine lab, Montreal Neurological Institute, McGill University, Montreal, Quebec, 3McGill Centre for Integrative Neuroscience (MCIN), Montreal, Quebec, 4Douglas Institute, McGill University, Montreal, Canada, Montreal, Quebec
First Author:
Co-Author(s):
Gleb Bezgin
Neuroinformatics for Personalized Medicine lab, Montreal Neurological Institute, McGill University
Montreal, Quebec
Neuroinformatics for Personalized Medicine lab, Montreal Neurological Institute, McGill University
Montreal, Quebec
Introduction:
Mounting evidence has established a relationship between intelligence and whole-brain morphometric measures. Studies have indicated that individuals with thicker cortexes in specific brain regions tend to exhibit better performance in certain cognitive tasks. Additionally, subcortical volumes of specific structures have been associated with particular cognitive ability differences. During early adolescence, the brain undergoes significant changes, including variations in cortical thickness and subcortical volume. However, in this developmental stage, the direct relationship between the co-varying patterns of brain geospatial indicators and intelligence remains incompletely understood. To gain deeper insights into this complex association, it becomes essential to identify intelligence brain networks. These intelligence brain networks will provide a valuable understanding of intelligence development and its connection with brain topography. We employ a stable and interpretable machine learning model to fit a large dataset, thereby elucidating the relationship between intelligence development and the co-varying brain networks.
Methods:
Data Preprocessing: The data we utilized is from the Adolescent Brain Cognitive Development (ABCD) dataset. We selected 7910 subjects of 9-14 years old which have both T1 data and cognitive data. According to age, we divided these samples into three groups (9-10, 11-12, and 13-14 years old). Cortical and subcortical surface morphometric measures were calculated with FreeSurfer 7.1.1. Cortical thickness was measured across 148 grey matter brain regions based on the Destrieux anatomical atlas. And subcortical volume measurements were made across 20 subcortical regions based on the ASEG atlas.
Intelligence definition: We used all the available cognitive measurements in the ABCD dataset, such as vocabulary, attention, reading, and short delay recall. Then we performed a PCA on those measurements to obtain the first principal component (Fig. 1). This component is known to be a fair estimate of general intelligence.
Interpretable Linear Regression Model: Our linear regression model has three improvements compared to the basic linear regression model (Fig. 1). First, it removes feature collinearity. Second, it adopts L1 regularization to select important features. Third, it utilizes gradient descent instead of least squares to get better predictions based on big data. We used this model to study the relationship between age and brain topography as well as between intelligence and brain geography.
Intelligence definition: We used all the available cognitive measurements in the ABCD dataset, such as vocabulary, attention, reading, and short delay recall. Then we performed a PCA on those measurements to obtain the first principal component (Fig. 1). This component is known to be a fair estimate of general intelligence.
Interpretable Linear Regression Model: Our linear regression model has three improvements compared to the basic linear regression model (Fig. 1). First, it removes feature collinearity. Second, it adopts L1 regularization to select important features. Third, it utilizes gradient descent instead of least squares to get better predictions based on big data. We used this model to study the relationship between age and brain topography as well as between intelligence and brain geography.
·Fig. 1. An overview of the co-varying Intelligence Brain Network identification.
Results:
Based on the Interpretable Linear Regression Model, we used brain topography data to predict age and intelligence. Both models work well (r>0.3, and pass permutation tests). Moreover, in order to identify important brain regions, we visualized the weights of the linear regression models (Fig. 2). We found that co-varying brain networks corresponding to age are similar to those corresponding to intelligence. And those co-varying brain networks are similar among different age periods.
·Fig. 2. Co-varying Brain Networks corresponding to age and intelligence.
Conclusions:
1. There exists a significant linear relationship between age and brain topographical features, as well as between intelligence and brain geographical features.
2. Brain topographical networks associated with age closely resemble those linked to intelligence. Furthermore, their patterns exhibit considerable similarity across different age groups.
2. Brain topographical networks associated with age closely resemble those linked to intelligence. Furthermore, their patterns exhibit considerable similarity across different age groups.
Higher Cognitive Functions:
Higher Cognitive Functions Other 1
Lifespan Development:
Early life, Adolescence, Aging 2
Modeling and Analysis Methods:
Classification and Predictive Modeling
Keywords:
Other - Intelligence brain networks, Interpretable Linear Regression Model
1|2Indicates the priority used for review
Provide references using author date format
Yan, J. et al. (2022), ‘Modeling spatio-temporal patterns of holistic functional brain networks via multi-head guided attention graph neural networks (Multi-Head GAGNNs)’, Medical Image Analysis, vol. 80, pp. 102518.
Tian, Y. et al. (2021), ‘Machine learning prediction of cognition from functional connectivity: Are feature weights reliable?’, NeuroImage, vol. 245, pp. 118648.