Advanced prediction of GBM recurrence (TIME) for personalized radiotherapy

摘要： 多形性胶质母细胞瘤(GBM)是成人最常见的原发脑恶性肿瘤。GBM患者的 对包括手术切除、放射治疗、化疗和肿瘤治疗领域在内的治疗的反应 (TTF)不令人满意，导致高复发率，这被认为是致命的。抢救RT通常用于 延缓复发的基底膜肿瘤生长，延长患者生存期。然而，由于 GBM细胞，增加一个大的各向同性治疗边缘(~2 cm)，以覆盖超出 磁共振成像(MRI)上的放射学证实的肿瘤。因为有相当大的重叠 在经常性和主要规划目标容量(PTV)之间，因额外回收而产生的增长延迟 正常器官所允许的辐射剂量耐受性适中。更显著、更有效的剂量是有毒的。 至高危器官(桨)，包括脑干、视交叉、视神经及受累脑实质等。 安全地增加剂量，必须显著减少复发治疗量。与 放射学证实的肿瘤增加了非特异性边缘，亚临床复发的体积在 较早的时间点明显较小。我们的初步研究基于干细胞生态位(SCN)在 基底膜细胞迁移显示体素预测基底膜复发的可行性 在射线照相上变得明显。GBM递归(时间)算法的预测是 通过训练开发了一种基于纵向多参数跟踪MR的机器学习分类器 图像，量化复发和大脑中SCN之间的潜在联系。鉴于前景看好 结果，有必要进一步改进算法，以便更准确地预测并建立其对 介入临床试验前的放射治疗计划。提出了以下目标： 实现目标。目标1：开发一个受干细胞定位弱监督的神经网络来执行 体素级别的复发预测。目标2a：预期患者图像数据的获取、预处理和 体素递归预测模型验证。目标2b：证明显著的剂量递增可以 达到了早期预测的复发。该项目的成功将进一步阐明SCN参与GBM， 提供了一种早期预测复发的方法，并有助于提高靶向准确性和抢救效果 放射治疗。最后一点将为一项可付诸实践的前瞻性干预试验铺平道路。 改变为GBM管理。

Abstract: Glioblastoma multiforme (GBM) is the most common primary brain malignancy in adults. GBM patients' response to therapies including surgical resection, radiotherapy (RT), chemotherapy, and tumor treating fields (TTF) is unsatisfactory, leading to a high recurrence rate, which is considered fatal. Salvage RT is often used to delay recurrent GBM tumor growth and prolong patient survival. However, due to the diffusive nature of the GBM cells, a large isotropic treatment margin (~2cm) is added to cover microscopic disease beyond the radiographically confirmed tumor on magnetic resonance image (MRI). Because of the considerable overlap between the recurrent and primary planning target volumes (PTV), growth delay from the additional salvage radiation dose allowed by the normal organ tolerance is modest. A more significant, more effective dose is toxic to organs at risk (OARs), including the brain stem, chiasm, optic nerves, and involved brain parenchyma, etc. To safely escalate the dose, the recurrent treatment volume must be significantly reduced. Compared with the radiologically confirmed tumor with added non-specific margin, the volume of the subclinical recurrence at an earlier time point is markedly smaller. Our preliminary research based on the role of stem cell niches (SCN's) in GBM cell migration shows the feasibility of voxel-wise prediction of GBM recurrences 2-3 months before they become radiographically apparent. The prediction of the GBM recurrence (TIME) algorithm was developed through training a machine learning classifier on longitudinal multi-parametric follow-up MR images, quantifying the potential connection between the recurrence and SCN's in the brain. Given the promising results, it is necessary to further improve the algorithm for more accurate prediction and establish its impact on radiotherapy treatment planning before an interventional clinical trial. The following aims are proposed to achieve the goal. Aim 1: Develop a neural network weakly supervised by stem cell niche locations to perform voxel-level recurrence prediction. Aim 2a: Prospective patient image data acquisition, pre-processing, and voxel-wise recurrence prediction model validation. Aim 2b: Demonstrate that significant dose escalation can be achieved for early predicted recurrence. The project's success will further elucidate SCN's involvement in GBM, provide a way to early predict recurrence, and help improve the targeting accuracy and efficacy of salvage radiotherapy. The last point will pave a path towards a prospective interventional trial that can be practice- changing for GBM management.