Personalized spatiotemporal hemodynamic response models for functional magnetic resonance imaging

Functional Magnetic Resonance Imaging (fMRI) shows great promise in characterizing brain circuits and networks related to human mental function and identifying pathophysiological changes underlying mental health disorders, healthy and pathological aging, substance misuse, and beyond. Great strides are being made in many areas, but the vast majority of fMRI research relies on the simplifying assumption of a canonical (or highly constrained) hemodynamic response function (HRF) that is substantially inaccurate. The HRF varies across brain regions, individuals, and age, but estimating it with sufficient accuracy and precision is problematic in small to medium-sized studies. As a result, over 95% of fMRI studies use a canonical HRF of fixed form. This results in substantial bias, power loss, and confounding. These problems apply to both task-based and connectivity studies, which rely implicitly on the assumption of a constant HRF across regions and individuals. In response to the “Notice of Special Interest (NOSI) regarding the Use of Human Connectome [HCP] Data for Secondary Analysis”, propose to use the Lifespan aged 5-100) combined with advanced statistical modeling t we HCP data (n=~3,600 high-quality datasets from people o address this issue. In Aim 1, we will contrast commonly used HRF models across the lifespan based on reliability and ability to ‘decode’ task state and phenotypic variables (e.g., cognitive function and mood). We develop novel methods for extracting meaningful phenotypic information from HRF shape and population inference, and develop robust software for best-in-class models. In Aim 2, we integrate best-in-class HRF models into a novel Gaussian process model and use it derive a demographic-specific, spatiotemporal HRF atlas, providing customized HRFs based on readily measurable characteristics (age, sex, and body- mass index) and brain region. In Aim 3, we use the HRF atlas to deconvolve rs-fMRI data and construct an HRF-corrected connectome map. We validate the HRF models, atlas, and connectome on two independent HCP Disease Connectomes and the CAM- CAN dataset (n=~700), and share the atlas, connectome, and software integrations with the research community. The development of these large-sample models will provide more accurate and precise estimates of task-related fMRI activity and connectivity in basic and clinical studies of mental health, aging, substance use, and beyond.