Generating Synthetic Task-based Brain Fingerprints for Population Neuroscience Using Deep Learning

Abstract

Task-based functional magnetic resonance imaging (tb-fMRI) reveals individual differences in the neural basis of cognitive functions by linking specific tasks to neural responses. However, scaling tb-fMRI to population-level studies is challenging due to its cognitive demands, variations in task design across studies, and the limited scope of tasks in large datasets. To address this, we propose DeepTaskGen, a deep-learning approach that generates non-acquired task-based contrast maps from resting-state fMRI (rs-fMRI) data. Our approach enables generating synthetic task images for non-acquired tasks within the study protocol. We validate this approach using the Human Connectome Project lifespan data, then generate 47 contrast maps from 7 different cognitive tasks for over 20,000 individuals from UK Biobank. DeepTaskGen outperforms several benchmarks in generating synthetic task-contrast maps, exhibiting superior reconstruction performance while retaining inter-individual variation essential for biomarker development. Notably, we further showed that synthetic task contrast maps achieved similar or greater performance compared to actual task contrast maps and resting-state connectomes for predicting a wide range of demographic, cognitive, and clinical variables. This approach will facilitate the study of individual differences and the generation of task-related biomarkers by enabling the generation of arbitrary functional cognitive tasks from readily available rs-fMRI data.