Computational imaging biomarkers of multiple sclerosis

摘要 多发性硬化症（MS）是一种中枢神经系统的慢性炎症性疾病， 认知和运动缺陷，影响近50万美国人和全球250万人。在 体内MRI可以检测疾病的标志性白色病变及其随时间的变化， 灵敏度高于疾病活动性的临床评估。此外，许多研究表明， 根据连续MRI评估，多发性硬化症患者各种大脑结构的萎缩累积速度比正常人快。 健康对照，并与残疾的措施。因此，能够可靠和高效地 表征白色病变、各种神经解剖结构及其变化的形态计量学特征 随着时间的推移，直接从体内MRI将具有巨大的潜在价值，用于诊断疾病，跟踪 进展和评估治疗。 虽然已经开发了许多用于从MS患者的MR扫描中分割白色病变的自动工具， 开发，这些通常只针对特定的研究协议进行调整，并不解决 表征MS患者的脑萎缩模式，其中已知病变的存在会干扰 萎缩估计。此外，MS中的计算神经成像工作几乎集中在 专门用于展示人口水平的统计关联，而不是预测模型， 联合收割机同时结合所有信息来源，以计算个体中最敏感的生物标志物 患者 为了解决这些局限性，本项目旨在（1）开发和验证扫描仪的自动化工具， 神经解剖学背景下白色病变的自适应分割;（2）开发和部署 用于预测个体患者水平的残疾的空间正则化模型;以及（3）概括， 验证，并在纵向设置中应用所提出的分割和预测工具。成功 该项目的完成将产生一组MS中的计算成像生物标志物， 临床观察比目前可用的方法;公开可用的软件工具， 跨广泛的成像硬件和协议分割MS患者的纵向扫描;以及 更详细地描述疾病进展基础的形态学和时间动力学 和MS中残疾的累积。

Abstract Multiple sclerosis (MS) is a chronic inflammatory disorder of the central nervous system that causes significant cognitive and motor deficits and affects nearly half a million Americans and 2.5 million individuals worldwide. In vivo MRI can detect the disease’s hallmark white matter lesions and their changes over time with a significantly higher sensitivity than clinical assessment of disease activity. Furthermore, numerous studies have shown that the atrophy accrual in various brain structures, assessed from serial MRI, is faster in patients with MS than in healthy controls, and correlates with measures of disability. Therefore, the ability to reliably and efficiently characterize the morphometry of white matter lesions, various neuroanatomical structures, and their changes over time directly from in vivo MRI would be of great potential value for diagnosing disease, tracking progression, and evaluating treatment. While many automatic tools for segmenting white matter lesions from MR scans of MS patients have been developed, these are typically tuned for specific research protocols only, and do not address the problem of characterizing brain atrophy patterns in MS patients, where the presence of lesions is known to interfere with atrophy estimation. Furthermore, computational neuroimaging efforts in MS have been focused almost exclusively on demonstrating statistical associations on population levels, rather than on prediction models that combine all sources of information simultaneously to compute the most sensitive biomarker in individual patients. In order to address these limitations, this project aims to (1) develop and validate automated tools for scanner- adaptive segmentation of white matter lesions within their neuroanatomical context; (2) develop and deploy spatially regularized models for predicting disability at the level of the individual patient; and (3) generalize, validate, and apply the proposed segmentation and prediction tools in longitudinal settings. The successful completion of this project will result in a set of computational imaging biomarkers in MS that correlate better with clinical observation than currently available methods; publicly available software tools for robustly segmenting longitudinal scans of MS patients across a wide range of imaging hardware and protocols; and a more detailed characterization of the morphological and temporal dynamics underlying disease progression and accumulation of disability in MS.