Statistical atlases of brain tumor MRI:do imaging phenotypes predict progression?

描述（由申请人提供）：统计地图集和相关的图像分析方法在几个神经成像领域得到了广泛的应用，提供了一种强大的方法来整合不同的成像信息，将其与遗传和临床测量相关联，了解疾病对大脑结构和功能的影响，并构建诊断工具。该建议将统计图像分析、可变形配准和生物物理建模方法结合起来，构建一个集成框架，用于构建和临床使用脑肿瘤患者的统计地图集。重点是胶质瘤，由于肿瘤浸润超出可见肿瘤边界，预后很差。因此，本研究的最终临床目的是识别可能被肿瘤浸润的脑组织的细微影像学特征，以及在相对较短的时间内可能出现复发的组织。这将通过研究健康和病理组织的多模态成像表型与空间信息（包括肿瘤的空间模式和恶性组织与白质纤维通路的接近程度）以及将这些表型与临床信息（包括肿瘤复发）相关联来实现。他们的假设是，信号和空间信息结合在一起，将能够识别出以后可能出现复发的脑组织。需要克服的主要技术挑战是：1)开发计算效率高的肿瘤生长、扩散和质量效应生物物理模型；2)可变形配准方法的发展，使我们能够共同配准带有肿瘤的脑图像并建立基于人群的图谱——这里的主要挑战是估计适当的肿瘤参数以及通常被水肿、浸润和极端变形混淆的肿瘤周围解剖结构的位置；3)发展机器学习方法，以表征脑组织的细微异常，并识别切除和治疗后可能出现复发的组织。该方法在更大规模临床研究中可行性的试点研究将在神经胶质瘤患者的脑磁共振图像数据库上进行，该数据库通过丰富而广泛的采集方案获得，包括灌注、扩散张量成像、光谱和常规成像。

DESCRIPTION (provided by applicant): Statistical atlases, and associated image analysis methods, have found widespread use in several neuroimaging fields, presenting a powerful way to integrate diverse imaging information, correlate it with genetic and clinical measurements, understand effects of disease on brain structure and function, and construct diagnostic tools. This proposal will combine statistical image analysis, deformable registration, and biophysical modeling approaches to an integrated framework for constructing and clinically using statistical atlases from brain tumor patients. Emphasis is placed on gliomas, which have very poor prognosis due to cancer infiltration beyond the visible tumor boundary. Accordingly, the ultimate clinical goal of this study is to identify subtle imaging characteristics of brain tissue that is likely to be infiltrated by tumor, as well as of tissue that is likely to present recurrence in relatively shorter time period. This will be achieved by studying the multi-modal imaging phenotypes of healthy and pathologic tissues in conjunction with spatial information, including the spatial pattern of the tumor and the proximity of malignant tissue to white matter fiber pathways, and by correlating these phenotypes with clinical information, including tumor recurrence. The hypothesis is that signal and spatial information together will be able to identify brain tissues that are likely to later present recurrence. The main technical challenges that will be overcome are 1) development of computationally efficient biophysical models of tumor growth, diffusion, and mass effect; 2) development of deformable registration methods that will allow us to co-register tumor-bearing brain images and build a population-based atlas-the main challenges here are to estimate the appropriate tumor parameters as well as the location of peri-tumor anatomy that is typically confounded by edema, infiltration and extreme deformations; and 3) development of machine learning methods for characterizing subtle abnormalities of brain tissue, and for identifying tissue that is likely to present recurrence after resection and treatment. Pilot studies on the feasibility of this approach to larger clinical studies will be performed on a database of brain MR images obtained from glioma patients via a rich and extensive acquisition protocol, including perfusion, diffusion tensor imaging, spectroscopy, and conventional imaging.