Extracting Cerebral Perfusion Signal from BOLD fMRI via Deep Learning

Motivation: Cerebral blood flow (CBF) is a fundamental physiological measure indicating regional brain function and vascular conditions via arterial spin labeling (ASL) perfusion MRI, but ASL sequences is limited. Goal(s): Blood-oxygen-level-dependent (BOLD) fMRI is known to be a function of CBF and other physiological sources. We try to extract CBF information from BOLD signal. Approach: We proposed a convolutional neural network to extract CBF from BOLD fMRI signal. Results: We confirmed the possibility of using supervised deep learning model to extract CBF from BOLD fMRI from independent sequences. Impact: Deep learning enables the estimation of CBF signal directly from the prevalent BOLD fMRI images, offering an alternative to the ASL sequence that is not universally available across research facilities.