UC Davis

Arterial spin labeling (ASL) is a non-invasive magnetic resonance imaging (MRI) technique. ASL is used to quantitatively measure cerebral blood flow (CBF), which represents the blood supply to the brain and provides a useful index of cerebral vascular perfusion and function. This is particularly important given that impaired cerebral vascular perfusion commonly accompanies neurodegenerative diseases or other pathophysiological impairments. A single-post labeling delay (PLD) ASL is widely used in conventional clinical ASL settings. In single-PLD ASL, we assume that the entire labeled ASL signal is delivered to its final location at a single inversion time, which is not ideal in estimating the CBF of each voxel of the brain. This is because the arterial transit time (ATT), the time required for the bolus of blood to travel from the labeled location to its final location (such as a brain tissue), varies by the location of the brain. Implementing multiple PLDs allows not only a more accurate CBF estimation but also ATT mapping. However, acquiring ASL images with multiple PLDs requires a relatively long scan time compared to a single PLD. Recent deep learning-based applications have shown the potential for reducing the number of features to recover an under-sampled dataset. Therefore, in this dissertation, we first proposed a hierarchically structured 3-dimensional convolutional neural network (H-CNN) to estimate the ATT and CBF maps from a reduced number of PLDs. The proposed method was compared to a conventional nonlinear model fitting method. The results showed that the H-CNN successfully estimated ATT and CBF maps using a reduced number of PLDs or averages with higher accuracy than the conventional nonlinear model fitting. The H-CNN also showed that the smaller number of PLDs can be used to estimate both ATT and CBF without significant discrepancy from the reference. Therefore, the proposed machine-learning-based ATT and CBF mapping can substantially reduce the total scan time of multi-PLD PCASL. Secondly, we further developed the H-CNN to estimate partial volume (PV) corrected ATT and CBF maps because the ASL perfusion images are significantly affected by PV effects by different tissue types such as gray matter (GM) and white matter (WM).In this dissertation, we also investigated the potential clinical applications of ASL. ASL provides various perfusion imaging biomarkers such as ATT, CBF, and cerebrovascular reactivity (CVR). CVR is an indicator of the compensatory dilatory capacity of CBF in response to vasoactive stimulation. Using a computer-controlled gas blender, we induced a hypercapnic challenge during the ASL scan. CVR was determined by calculating the percent difference in CBF between the resting state and during the hypercapnic challenge. We then investigated whether CBF or CVR is related to cognitive performance at the brain region-of-interest (ROI) levels. Additionally, we examined the potential correlation between cognitive status, including normal cognition and mild cognitive impairment, and cerebrovascular perfusion imaging parameters derived from ASL, such as CBF, ATT, and CVR on a voxel-by-voxel basis in the brain. Moreover, we further investigated the relationship between impaired cerebrovascular perfusion and changes in WM microstructure.