Multi-atlas based Direct Estimation in Preclinical Alzheimer's Disease

项目概要： 在这个应用中，我们将建立一个新的范例，基于图谱的大脑MRI分析临床前阿尔茨海默氏症 疾病（AD），其特征在于直接估计患者的属性，具有独特的基于知识的信息- 检索技术MRI图谱被广泛用于自动图像分割，特别是最近的进展 在多图谱方法中已经产生了上级分割精度。在传统的基于地图集的分析中， 图谱仅用作模板来分割患者图像，然后 以估计患者的诊断属性，诸如认知状态 或临床评估。相比之下，在提出的基于多图谱的直接估计（MADE）方法中，我们 利用所述多图谱库作为与丰富的临床信息相关联的知识数据库; 患者图像和图谱图像之间的相似性将用于权衡来自多个图谱的信息，并且 将进行加权决策以绕过分割过程直接估计患者属性。 我们的初步数据已经证明了基于MADE的T1加权图像分析的优势， 老年痴呆症患者，与体积分析相比。此外，非图像人口统计学和临床信息 的患者可以很容易地纳入MADE框架，以进一步提高估计的准确性。 我们的目标是开发基于MADE的计算工具，并使用它们来改善诊断和预后。 AD的临床前阶段。本研究将得到BIOCARD队列的支持，这是一个精心设计的 这项纵向研究跟踪了350名参与者超过20年。全面认知评估和MRI 从1995年开始收集这些参与者的测试，当时他们在入学时认知正常。这 独特的数据库使我们能够在长期的临床前阶段研究脑退化，并开发 计算工具，可能有助于诊断决策，在这个关键阶段。在目标1中，我们将开发 并优化基于MADE的脑结构MRI分析图，以估计患者当前的认知能力 状态和疾病阶段，使用ADNI和BIOCARD数据。一旦工具成熟， 他们在一个云计算平台上-MRiCloud，供公众使用。在目标2中，我们使用优化的MADE 预测BIOCARD队列中认知障碍的管道。具体来说，我们将使用MADE管道来 预测基线后1 - 5年患者的认知功能下降，并预测他们从 在20年随访期间，正常至认知障碍（MCI、AD或其他类型的异常），以及 估计诊断时间。拟议项目的成功可能会导致下一代 基于知识的计算机辅助诊断，并可能提高早期诊断和预后的准确性， 临床前AD患者。

Project summary: In this application, we will establish a novel paradigm of atlas-based brain MRI analysis of preclinical Alzheimer’s disease (AD), featuring in direct estimation of the patients’ attributes with a unique knowledge-based information- retrieval technology. MRI atlases are widely used for automated image parcellation, especially, recent advances in multi-atlas methods have yielded superior segmentation accuracy. In conventional atlas-based analysis, atlases are used merely as templates to segment a patient image, and then volumes, shapes, intensities are extracted from the segmented structures to estimate the patient’s diagnostic attributes, such as cognitive status or clinical assessments. In contrast, in the proposed multi-atlas based direct estimation (MADE) approach, we utilize the multi-atlas library as a knowledge database that is associated with rich clinical information; anatomical similarity between the patient and atlas images will be used to weigh the information from multiple atlases, and a weighted decision will be made to directly estimate the patient attributes, bypassing the segmentation process. Our preliminary data have demonstrated the advantages of MADE-based analysis of T1-weighted images in dementia patients, compared to volumetric analysis. In addition, non-image demographic and clinical information of the patients can be readily incorporated into the MADE framework to further enhance the estimation accuracy. Our goals is to develop MADE-based computational tools and use them to improve diagnosis and prognosis in preclinical phase of AD. This study will be supported by the BIOCARD cohort, which is a well-designed longitudinal study that followed 350 participants over 20 years. Comprehensive cognitive assessments and MRI exams have been collected in these participants since 1995 when they were cognitive normal at enrollment. This unique database allows us to investigate brain degeneration in the long preclinical phase, and develop computational tools that can possibly assist diagnostic decisions in this critical phase. In Aim 1, we will develop and optimize MADE-based brain structural MRI analysis diagram to estimate the patients’ current cognitive status and disease stages, using both the ADNI and BIOCARD data. Once the tools become mature, we deploy them on a cloud-computing platform—the MRICloud, for public use. In Aim 2, we use the optimized MADE pipeline to predict cognitive impairment in BIOCARD cohort. Specifically, we will used the MADE pipeline to predict patients’ cognitive decline at 1-5 years after baseline; and also predict their probability of conversion from normal to cognitive impairment (MCI, AD or other types of abnormality) across the 20 years of follow-up, and estimate the time-to-diagnosis. The success of the proposed project could lead to the next generation of knowledge-based computer-aided diagnosis, and potentially improve early diagnosis and prognosis accuracy in preclinical AD patients.