Majed Alfaraj

Artificial Intelligence Specialization · Cohort 2025

Application of Deep-Learning to Enable Quantitative Brain Hemodynamic Mapping Using Resting-State fMRI

Cerebral hemodynamics are critically important for diagnosing and monitoring brain diseases. For this purpose, the potential for resting-state functional magnetic resonance imaging (rs-fMRI) has been explored. rs-fMRI a powerful tool previously used for mapping neural activity, is available in most hospitals. Typically, carbon-dioxide (CO2) modulations and recordings are needed in a deconvolution to obtain hemodynamic metrics such as cerebrovascular reactivity (CVR) and bolus arrival time (BAT). However, deconvolution requires the assumption of linearity, and the interaction between CO2 and resting-state BOLD is highly nonlinear. Therefore, machine learning is an excellent choice for accelerating the CVR quantification process even in the absence of physiological monitoring [37]. Here we propose that rs-fMRI can be used to map cerebral hemodynamic function and delineate impairment. By exploiting time variations in breathing patterns during rs-fMRI, deep learning enables reproducible mapping of CVR and bat of the human brain using resting-state spontaneous fluctuations in CO2 as a natural “contrast medium”. The aim of this project is to use DL methods to estimate quantitative CVR maps directly from the rs-fMRI data, eliminating the need for CO2 recordings. Unlike in previous research, we aim to generate quantitative CVR maps using DL. Both rs-fMRI and hypercapnia-fMRI datasets will be utilized to train this model. We also aim to enhance the quality of rs-CVR maps, as DL models have shown success in improving the signal-to-noise ratio (SNR) in low-field MR images. Inspired

Mentor: Jean Chen

Read the final report (PDF)

See it in the program report (PDF)