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

Area of Science:

• Neuroimaging
• Cognitive Neuroscience
• Artificial Intelligence

Background:

• Task-based functional magnetic resonance imaging (tb-fMRI) is crucial for understanding cognitive functions and individual neural differences.
• Scaling tb-fMRI to population studies is hindered by task demands, design variability, and limited task scope in large datasets.

Purpose of the Study:

• To develop a deep-learning method (DeepTaskGen) for generating task-based contrast maps from resting-state fMRI (rs-fMRI) data.
• To enable the generation of synthetic task images for non-acquired tasks within existing study protocols.

Main Methods:

• DeepTaskGen, a deep-learning model, was proposed to generate task-contrast maps from rs-fMRI data.
• The approach was validated on the Human Connectome Project lifespan data.
• Synthetic contrast maps were generated for 7 cognitive tasks across over 20,000 UK Biobank participants.

Main Results:

• DeepTaskGen demonstrated superior reconstruction performance compared to benchmarks in generating synthetic task-contrast maps.
• The generated maps preserved essential inter-individual variations for biomarker development.
• Synthetic task contrast maps showed comparable or superior performance to actual task maps and rs-fMRI connectomes in predicting demographic, cognitive, and clinical variables.

Conclusions:

• DeepTaskGen facilitates large-scale studies of individual differences in cognitive functions.
• The approach enables the generation of task-related biomarkers from readily available rs-fMRI data.
• This method allows for the creation of arbitrary functional cognitive tasks from resting-state scans.