Frequency-Domain Machine Learning Estimation of Maximum BOLD Modulation and Grey Matter Oxygen Consumption with Resting-State BOLD-ASL fMRI

Motivation: Calibrated BOLD-arterial spin labelling (ASL) fMRI exploits isometabolic hypercapnic changes of brain physiology to map grey matter maximum BOLD modulation (M) and, through biophysical modelling, estimate the oxygen extraction fraction and the cerebral metabolic rate of oxygen. However, this approach requires a CO2 gas-challenge or breath-holding, limiting its clinical application. Goal(s): It would be ideal to estimate M from low SNR, non-isometabolic resting-state (RS) BOLD-ASL fluctuations. Approach: We investigate the ability of a frequency-domain, data-driven, neural network approach to estimate the physiological parameters of interest from RS data in comparison to a breath-hold approach. Results: The proposed approach can map the desired parameters. Impact: The ability to map oxygen consumption in the grey matter through resting state data using a calibrated fMRI framework would allow a simple implementation of such an approach in research settings paving the way to its utilization in clinical practice.