New treatment monitoring biomarkers for brain tumors using multiparametric MRI with machine learning

项目总结/摘要 该项目的目标是开发和评价使用多参数MRI的新型成像生物标志物 方法来确定胶质脑肿瘤的真实空间范围。标准RANO（缓解评估， 神经肿瘤学）标准将肿瘤范围定义为造影剂后T1w（T1 + C）上的亮信号区域 T2 WI FLAIR图像，称为对比增强病变（CEL），沿着瘤周亮信号 图像，称为非增强病变（NEL）。然而，CEL反映了血脑的渗透性 对造影剂的屏障，并且对于肿瘤和治疗效果可以表现为相同。同样，尽管 可能含有肿瘤，目前的成像无法区分肿瘤和水肿。这些困难导致 目前的解剖MRI方法无法确定神经胶质肿瘤的真实空间范围， 严重限制了脑肿瘤患者的治疗管理。 我们和其他人已经表明，先进的MRI方法，包括灌注和扩散MRI，是有用的， 评估肿瘤等级、预测结果或区分肿瘤与治疗效果。然而，几乎 排他地，该方法已经从以下各项中提取单个生理参数的平均值： 在一些实施例中，患者可以测量预定的感兴趣的肿瘤区域，然后测量它们与期望的临床指数的相关性。 虽然这种方法对于最初的生物标志物开发是有用的，但它没有充分利用丰富的生物标志物。 多参数和空间信息，从而激发了目前的研究。首先，两个多参数 将开发MRI生物标志物以识别增强和浸润肿瘤负荷。然后，他们将 单独和联合评价，以评估与标准相比的总肿瘤负荷 当前使用的体积度量。 这些生物标志物的开发和测试将在几个独立的步骤中完成 提出的目标。首先（目标1），我们建议开发一种MRI生物标志物， 增强CEL内肿瘤负荷的可能性，早期结果显示能够区分肿瘤 治疗效果。接下来，我们将开发一种能够识别浸润性肿瘤的多参数生物标志物 在NEL内（目标2）。这些努力利用了我们以前使用人工智能的结果， 机器学习，以及我们独特的脑肿瘤组织库，其中包含数百个空间匹配的活检样本 to imaging成像.最后（目标3），将测试CEL和NEL内肿瘤负荷的空间范围， 区分假进展/反应与真实进展/反应，这是一个主要问题， 目前的治疗管理混乱。 总之，增强型和浸润型脑肿瘤的多参数高级MRI生物标志物具有 有可能导致治疗管理方式的范式转变，最终改善结局。

Project Summary/Abstract The goal of this project is to develop and evaluate novel imaging biomarker(s) that use multiparameter MRI methods to identify the true spatial extent of glial brain tumors. The standard RANO (response assessment in neuro-oncology) criteria define tumor extent as the region of bright signal on post-contrast agent T1w (T1+C) images, termed the contrast enhancing lesion (CEL), along with the peritumoral bright signal on T2w FLAIR images, referred to as non-enhancing lesion (NEL). Yet, the CEL reflects the permeability of the blood-brain barrier to contrast agent and can appear the same for both tumor and treatment effect. Likewise, though NEL likely contains tumor, current imaging cannot distinguish tumor from edema. These difficulties result in the inability of current anatomical MRI methods to determine the true spatial extent of glial tumors, a serious limitation for treatment management of brain tumor patients. We and others have shown that advanced MRI methods, including perfusion and diffusion MRI, are useful for assessing tumor grade, predicting outcomes, or distinguishing tumor from treatment effect. Yet, almost exclusively, the approach has been to extract mean values of a single physiological parameter from predetermined tumor regions of interest and then measure their correlation with the desired clinical index. Although this approach has been useful for initial biomarker development, it underutilizes the rich multiparameter and spatial information available, thus motivating the current study. First, two multiparameter MRI biomarkers will be developed to identify enhancing and infiltrating tumor burden. Then, they will be evaluated individually and in combination to assess the total tumor burden in comparison with the standard volumetric metrics in current use. The development and testing of these biomarkers will be accomplished in several independent steps outlined by the proposed aims. First (Aim 1), we propose to develop an MRI biomarker that gives the voxelwise probability of enhancing tumor burden within CEL, with early results showing the ability to distinguish tumor from treatment effect. Next, we will develop a multiparameter biomarker capable of identifying infiltrating tumor within NEL (Aim 2). These efforts leverage our previous results using artificial intelligence, recent advances in machine learning, and our unique brain tumor tissue bank with hundreds of biopsy samples spatially matched to imaging. Finally (Aim 3), the spatial extent of tumor burden within CEL and NEL will be tested in their ability to distinguish pseudo-progression/response from true progression/response, which is a primary question that confounds treatment management today. In summary, multiparameter advanced MRI biomarkers of enhancing and infiltrative brain tumor have the potential to cause a paradigm shift in how treatment is managed, ultimately resulting in improved outcomes.