Connectome Fingerprinting Predicts Prefrontal Cortical Activation During Abstract Reasoning

Abstract

Connectome fingerprinting (CF) is a machine learning technique that uses an individual’s unique brain connectivity profile to make predictions about their individual brain activity patterns. Previous studies have successfully employed CF using sensory, working memory, and auditory/visual fMRI data. Here, we demonstrate that this technique can be used to examine higher-order cognition and evaluate the effectiveness of the CF method in the prefrontal cortex (PFC), an area where brain function and structure has been shown to be exceptionally variable across individuals. We designed a CF model, using a leave-one-out cross-validation scheme, to predict individually-specific blood-oxygen-level-dependent (BOLD) activation in the lateral PFC. We used an abstract reasoning task designed to probe higher-order cognitive function specific to the lateral PFC. Our results demonstrated that CF better predicts abstract reasoning activation in the lateral PFC compared to the group average. We also tested whether task-evoked BOLD activity could be predicted using a model trained on a subset of the functional connectome. We demonstrate that CF models trained on a partial connectome from associative brain networks achieved similar levels of model performance compared to models trained on the full connectome. We observed that model performance varied across participants, suggesting that within-participant reliability in task activation, and deviation from the group average influence the model. Overall, we used CF to characterize individuals’ task-evoked BOLD activation during abstract reasoning, demonstrating a link between resting-state functional architecture and brain activations and identifying key networks that contribute to higher-order cognition.