Executive function-related functional connectomes predict intellectual abilities

• Published in: Intelligence (Norwood) Vol. 85; p. 101527
• Main Authors: He, Li, Liu, Wei, Zhuang, Kaixiang, Meng, Jie, Qiu, Jiang
• Format: Journal Article
• Language: English
• Published: New York Elsevier Inc 01.03.2021; Elsevier Science Ltd
• Subjects: Ability; Cognition; Cognitive ability; Datasets; Executive function; Executive functions; Fluid intelligence; Functional connectivity; Individual differences; Intellectual abilities; Intelligence; Learning algorithms; Machine learning; Magnetic resonance imaging; Measurement; Nervous system; Neural networks; Prediction models; Intellectual abilities; Functional connectivity; Executive functions; Machine learning; Individual differences
• DOI: 10.1016/j.intell.2021.101527
• ISSN: 0160-2896

Executive functions (EFs) refer to a range of cognitive control and regulation processes that coordinate thoughts and actions in a goal-directed way and are regarded as a hallmark of intellectual abilities. However, most studies have used a single measurement to explore the relationship between EFs and intelligence, and there is a lack of robust evidence to demonstrate the link between EF-related neural substrates and intelligence under an integrative framework. To address this issue, we employed a large sample (primary dataset, n = 881; hold-out dataset, n = 181) from the Human Connectome Project, which included high-quality MRI data and multiple EF and intelligence measurements. Based on a machine learning framework, we examined the predictive effect of EF-related functional connectivity (FC) on fluid intelligence (Gf) and crystallized intelligence (Gc) using a connectome-based predictive model. The results showed that all types of EF-related FCs (i.e., EF-common, updating-, shifting-, and inhibition-specific FCs) predicted novel subjects' Gf and Gc in the primary dataset and successfully generalized to the hold-out dataset. Additionally, EF-related FCs appeared to demonstrate better performance in predicting Gc. Identified predictive FCs revealed the domain-general and domain-specific connectivity patterns of EFs, and the network hubs were mainly located in the default mode, cognitive control, salience, and visual networks. These findings facilitate our understanding of the relation between multiple EF domains and intelligence from the perspective of network neuroscience, suggesting that different intellectual abilities and EFs share similar neural bases to some extent, which allows the link between EFs and intelligence to be revisited. •Using EF-related FCs to predict intelligence in a large sample.•EF-related FCs predicted novel subjects' Gf and Gc.•The predictive models built with EF-related FCs generalized to the hold-out dataset.•EF-related FCs appeared to exhibit better performance for predicting Gc.