3D Multi-Contrast MRI with Automatic Brain Tissue and Lesion Segmentation

项目总结 这项工作的目标是开发和评估一种用于多对比成像和自动损伤的新策略 胶质瘤的分割将显著影响临床成像工作流程，并使稳健的措施 监测治疗反应。生成的图像、量化地图和分割区域将具有 1 mm各向同性分辨率，全脑覆盖，6分钟扫描时间，方便日常使用 评估反应的客观标准，并揭示病变随时间增长的细微变化 目测评估或肿瘤横断面直径测量漏诊。除了生成 定量的T1、T2和大分子质子分数图以及常规的T2-、FLAIR-和T1- 加权图像，我们自动分割对应于对比度增强和T2的区域- 高密度病变将在不注射基于Gd的造影剂的情况下进行。 我们的策略包括使用高速加速的3D MRI序列和多个反转脉冲来实现 全脑，通过在不完全反转恢复过程中连续采集数据进行多对比成像 平衡稳态自由进动。这种独特的方法克服了传统方法固有的局限性 解剖学成像在有限的窗口内获取数据，并需要通过合并 在MR指纹方法中使用的词典搜索。我们最近的研究证明了 通过利用增加的对比在脑肿瘤患者中对这一序列进行成像 自动分割对比度增强的病变、浸润性肿瘤和水肿，并突出显示需要 用于进一步改进参数和评估患者。具体来说，在目标1中，我们将制定和 评价：1)基于水的多室模型的定量多参数映射 和大分子质子池，并包括磁化传递效应；2)患者特定的自动化 利用定量组织T1、T2和大分子质子分数映射的组织分割，3) 使用运动补偿来改进组织和病变分割。由此产生的量化地图， 合成图像和组织分割将通过与它们各自的参考文献进行比较来评估 在正常脑组织和病变中。然后，目标2将利用最佳参数集来评估结果 强化高级别胶质瘤患者的节段性病变。派生自自动 我们的多对比扫描在注射Gd前后的分割将与手动进行比较 定义感兴趣区域，以确定预对比剂注射多对比扫描是否可以 准确：1)勾画增强灶；2)区分浸润性肿块和水肿性肿块。

PROJECT SUMMARY The goal of this work is to develop and evaluate a novel strategy for multi-contrast imaging and automatic lesion segmentation of gliomas that will significantly impact clinical imaging workflow and enable robust measures for monitoring therapeutic response. The resulting images, quantitative maps, and segmented regions will have 1mm isotropic resolution, full brain coverage, in 6 minutes scan time, facilitating their routine use in providing objective criteria for response assessment and revealing subtle changes in lesion growth over time that can be missed by visual assessment or measurements of cross-sectional tumor diameter. Besides generating quantitative T1, T2, and macro-molecular proton fraction maps along with conventional T2-, FLAIR-, and T1- weighted images, our automatic segmentation of regions that correspond to contrast-enhancing and T2- hyperinense lesions will be performed without the injection of a gadolinium-based contrast agent. Our strategy involves using a highly accelerated 3D MRI sequence with multiple inversion pulses to achieve whole brain, multi-contrast imaging by continuously acquiring data during incomplete inversion recovery with balanced steady state free precession. This unique approach overcomes the limitations inherent in conventional anatomical imaging that acquire data during a limited window and require full inversion recovery by incorporating dictionary searching that is used in the MR fingerprinting approach. Our recent studies demonstrate the potential of this sequence in imaging patients with brain tumors by taking advantage of the added contrasts to automatically segment the contrast-enhancing lesion, infiltrative tumor, and edema, as well as highlight the need for further refinement of parameters and evaluation in patients. Specifically, in Aim 1 we will develop and evaluate: 1) quantitative multi-parametric mapping based on a multiple-compartment model comprised of water and macromolecular proton pools and includes magnetization transfer effects; 2) a patient-specific automated tissue segmentation that utilizes quantitative tissue T1, T2, and macromolecular proton fraction mapping, 3) using motion compensation to improve tissue and lesion segmentation. The resulting quantitative maps, synthetic images, and tissue segmentations will be evaluated through comparison with their individual references in normal brain tissue and lesions. Aim 2 will then utilize the best set of parameters to evaluate the resulting segmented lesions in patients with enhancing high-grade gliomas. Volumes derived from the automatic segmentation of our multi-contrast scan pre- and post-injection of gadolinium will be compared to manually defined regions of interest in order to determine whether the pre-contrast injection multi-contrast scan can accurately: 1) delineate the contrast-enhancing lesion and 2) separate infiltrative tumor from edema.