Assessing ASL CBF as a biomarker for early Alzheimer's disease detection and disease progression

摘要。阿尔茨海默病（AD）是一种致命的神经退行性疾病，影响数千万人， 人AD研究的首要任务是找到对早期疾病和疾病进展敏感的生物标志物 因为这可能会为寻找和评估有效的治疗方法提供最好的机会， 目前无法治愈的脑部疾病人类大脑依赖于脑血流量（CBF）为其提供能量， 废物清除先前的研究表明，CBF减少导致神经元不活动和神经毒性。 废物积累和随后的神经元死亡，这可能最终导致AD痴呆。测量 CBF及其纵向过程可能为早期AD及其相关疾病提供高度潜在的生物标志物。 进展动脉自旋标记（ASL）灌注MRI是一种定量CBF的技术， 外源示踪剂因为它相对便宜，可以重复多次，所以非常适合 纵向AD研究。ASL MRI对AD和前驱AD的敏感性已在许多研究中得到证实。 其他组和我们的横断面研究（比较AD与对照组）。但仍有几个重要问题 包括ASL CBF对早期AD和AD进展的预测能力， AD患者脑血流纵向变化规律及性别差异。解决这些问题需要大量的 由于低信噪比（SNR），因此需要大量纵向数据和专业知识来分析ASL MRI数据。的 本项目的目的是通过利用我们在ASL MRI方面的广泛专业知识来解决这些悬而未决的问题 以及来自ADNI（一项大型多部位AD神经影像学研究）的累积纵向ASL数据。的 本项目的新信息或结果将包括ASL CBF对早期AD的预测能力，CBF 疾病进展或逆转时的变化率，AD及其进展中的性别效应，以及 基于深度机器学习（DL）的下一代ASL MRI处理算法，以及基于DL的AD 预测模型我们将首先使用更大的样本确认我们之前的ADNI ASL CBF结果，并更新 方法（以前不可用）。然后，我们将检查ASL CBF用于跟踪和预测疾病的灵敏度 进展或认知能力下降。性别对AD患者脑血流的影响将被明确研究，这可能揭示出 女性AD患病率较高的线索。我们将使用基于DL的ASL去噪重新审视这些研究 算法追求这些目标将有助于建立ASL CBF作为AD生物标志物，并提供一种通用的AD 使用ASL CBF以及ADNI中提供的其他有价值的生物标志物的预测模型。发展中国家和 共享DL ASL MRI去噪方法不仅有利于AD研究， 基于ASL MRI的项目。这一创新但具有临床重要性的项目的可行性由我们的 数十年的经验和研究各方面的大量初步研究，包括ASL MRI， AD ASL研究和机器学习。

ABSTRACT. Alzheimer’s Disease (AD) is a fatal neurodegenerative disease affecting tens of millions of people. A top research priority in AD is to find a biomarker sensitive to early disease and disease progression because that will likely provide the best opportunities for searching and evaluating effective treatments for this currently incurable brain disease. Human brain relies on cerebral blood flow (CBF) for its energy supply and waste removal. Previous research has indicated that CBF reductions cause neuron inactivity and neurotoxic waste accumulation and subsequently neuron death, which may eventually lead to AD dementia. Measuring CBF and following its longitudinal course may then provide a highly potential biomarker for early AD and its progression. Arterial spin labeling (ASL) perfusion MRI is a technique for quantifying CBF without using exogenous tracers. Because it is relatively cheaper and can be repeated many times, it is well suited for longitudinal AD research. Sensitivity of ASL MRI to AD and prodromal AD has been shown in many cross-sectional studies (comparing AD to controls) by other groups and us. But several important questions still remain unanswered including the prediction power of ASL CBF for early AD and AD progression, the longitudinal CBF change patterns, and sex difference of CBF in AD. Addressing those questions needs large size longitudinal data and expertise for analyzing ASL MRI data due to the low signal-to-noise-ratio (SNR). The purpose of this project is to address those open questions by leveraging our extensive expertise on ASL MRI and the accumulating longitudinal ASL data from ADNI (a large ongoing multi-site AD neuroimaging study). The novel information or outcome from this project will include the prediction power of ASL CBF for early AD, CBF change rate when disease progresses or reverts, gender effects in AD and its progression, and a next-generation ASL MRI processing algorithm based on deep-machine learning (DL), and a DL-based AD prediction model. We will first confirm our previous ADNI ASL CBF findings using larger sample and updated methods (not available before). We will then check sensitivity of ASL CBF for tracking and predicting disease progression or cognitive declines. Gender effects on CBF in AD will be explicitly examined, which may reveal a clue for the higher prevalence of AD in females. We will revisit those studies using the DL-based ASL denoising algorithm. Pursuing those aims will help establishing ASL CBF as an AD biomarker and provide a versatile AD prediction model using ASL CBF as well as other valuable biomarkers provided in ADNI. Developing and sharing the DL ASL MRI denoising method will benefit not only AD research but also the various scientific projects based on ASL MRI. The feasibility of this innovative but clinically important project is ensured by our decades of experience and the substantial pilot investigations in each aspect of the study, including ASL MRI, AD ASL study, and machine learning.