Developing Advanced CEST MRI Techniques to Noninvasively Detect Neurological Disorders at 3T


Student thesis: Doctoral Thesis

View graph of relations


Related Research Unit(s)


Awarding Institution
Award date20 Jan 2021


Chemical exchange saturation transfer (CEST) magnetic resonance imaging (MRI) is a molecular imaging method which can noninvasively detect proteins/lipids and metabolites via exchangeable protons. It is able to detect the molecular changes in diseases, such as tumor, stroke and Alzheimer’s disease (AD). Currently, CEST MRI studies have been demonstrated mainly on high-field MRI scanners (≥7T), where images with high signal-to-noise ratio (SNR) can be obtained. However, most of clinical MRI scanners are equipped with low field strengths (≤3T). There are two major factors that hamper the translation of CEST MRI techniques to low field scanners (≤3T). First, CEST effects are prone to be compromised by the scaling effects of magnetization transfer contrast (MTC) and direct water saturation (DS), especially when a high saturation power is applied. Second, the T1 relaxation time of tissues at low field is much shorter than that at high field, making CEST studies challenging at low field. To facilitate the translation of CEST MRI to clinical applications, it is necessary to develop new CEST MRI techniques at low field. In this thesis, we designed and established some advanced CEST MRI techniques that had high sensitivity and specificity at clinical field strength 3T and applied them to noninvasively detect neurological disorders, including AD and multiple sclerosis (MS). Our findings would promote the clinical applications of CEST MRI at 3T. This thesis was presented in the following three major aspects:

First, we proposed an advanced dynamic glucose enhanced (DGE) CEST MRI approach, named on-resonance variable delay multi-pulse (onVDMP), to simultaneously monitor natural D-glucose uptake and clearance in brain parenchyma and cerebrospinal fluid (CSF). Conventional DGE MRI only monitors the glucose signal in brain parenchyma, while our proposed method could additionally monitor the glucose signal in CSF. This could provide more information related to glymphatic system of brain. We then applied our DGE MRI method to study the glucose changes in AD mouse brain. We observed significantly higher uptake in parenchyma of young (6 months) transgenic AD mice (APP/PS1) compared to age-matched wild type (WT) mice (P = 0.007). Significantly lower uptakes were observed in parenchyma and CSF of old (16 months) AD mice (P = 0.035). Interestingly, both young and old AD mice had a significantly slower CSF clearance than age-matched WT mice (young: P = 0.017; old: P = 0.031). This resembles recent reports of the hampered CSF clearance that leads to protein accumulation in the brain. These findings suggest that DGE MRI can identify altered glucose uptake and clearance in young AD mice upon the emergency of amyloid plaques at clinical field strength 3T MRI. Based on these findings, we further developed a DGE MRI acquisition and post-processing scheme for sensitive monitoring glucose uptake and clearance in both brain parenchyma and CSF. By investigating Carr-Purcell-Meiboom-Gill sequence (CPMG), onVDMP and on-resonance spin-locking (onSL), a high-sensitivity DGE MRI scheme composed of CPMG method for monitoring parenchyma and onVDMP method for monitoring CSF was developed. We incorporated the multilinear singular value decomposition (MLSVD) based denoising method in post-processing, which enables the detection of DGE signals from the brain parenchyma and CSF at low concentration of D-glucose (12.5% w/w) injection.

Second, a deep neural network based CEST data processing method (CESTNet) was exploited to analyze the CEST data of mouse brains and promising results have been achieved. CEST data processing usually requires expert knowledge and the outcomes are diverse when different processing methods are applied. This could be challenging for non-CEST expert, e.g. clinicians and physicists. Moreover, we first time proposed to utilize deep neural network to generate apparent exchange-dependent relaxation (AREX) results (AREXNet). Previous studies have shown that AREX provides CEST signal with less influence of longitudinal relaxation (R1 = 1/T1). After optimization and training on CEST data of wild type (WT) mouse brains, CESTNet/AREXNet could rapidly (~1s) generate accurate CEST/AREX results for unseen CEST data of AD mouse brains, indicating the generalization of the networks for different CEST data. Significant lower amide proton transfer/magnetization transfer (APT/MT) signals related to amyloid beta-peptide (Aβ) plaque depositions, which were validated by immunohistochemistry results, were detected in AD mouse brains compared to WT mouse brains. The well-established CESTNet/AREXNet have great potential for AD identification at 3T clinical field strength MRI.

Third, a pulsed-CEST MRI imaging scheme with low saturation powers (B1) and long mixing time (tmix) was established to rapidly acquire relayed nuclear Overhauser effect (rNOE) weighted image with MTC suppression at 3T clinical field strength MRI. The MTC contributions were further reduced by subtracting the Z-spectral signals at two or three offsets by assuming the residual MTC is a linear function between -3.5 ppm and -12.5 ppm. Phantom studies of a lactic acid (Lac) solution mixed with cross-linked bovine serum albumin (BSA) show that strong MTC interference has significant impact on the optimum B1 for detecting rNOEs due to lactate binding. MTC could be effectively suppressed using a pulse train with a B1 of 0.8 μT, a pulse duration (tp) of 40 ms, a tmix of 60 ms and a pulse number (N) of 30, while rNOE signal was well maintained. As a proof-of-concept, we applied the method in mouse brain with injected hydrogel and a cell-hydrogel phantom. Results showed that rNOE weighted images could provide good contrast between brain/cell and hydrogel. Finally, we applied our optimized rNOEw imaging scheme to study the pathology changes regarding myelin lipid/protein in human brain with MS and neuromyelitis optica spectrum disorder (NMOSD) on a clinical 3T MRI scanner. We found that rNOE signal of MS brains was significantly lower than that of NMOSD and healthy brains. Our rNOEw imaging scheme has great potential to assist MS diagnosis and specifically differentiate MS patients from NMOSD patients.