Using Connectomics and Machine Learning to Predict Survival in Diffuse Glioma

摘要 弥漫性胶质瘤是最常见的恶性原发性脑肿瘤。临床结局，包括总体 存活率，在个体患者中差异显著，且不能通过已知的预后因素充分解释。 因素如年龄、组织学和分子病理学以及给予的治疗。更深入地了解 弥漫性胶质瘤的预后影响有助于治疗决策和患者咨询， 让我们进一步了解这种疾病的生物学基础。磁共振成像（MRI）扫描 大脑是这些患者标准术前评估的一部分。大脑的结构和功能 受生物和环境因素的调节，MRI衍生的指标通常提供敏感的预后 生物标志物。我们已经初步表明，连接组学，一种测量大脑连接的方法， MRI可以准确预测弥漫性胶质瘤患者的生存率。我们将回顾性地获得至少 1，150个数据集，并尝试验证我们的初步模型（目标1）。我们还将检查连接体 与肿瘤基因型相关的表型，以帮助完善和改进我们的模型（目的2）。准确的预- 对结果的手术预测最终可以允许更好地为个体患者定制干预 并协助临床医生在治疗决策中优化肿瘤控制和神经功能。

ABSTRACT Diffuse gliomas are the most common malignant primary brain tumors. Clinical outcomes, including overall survival, vary significantly across individual patients and are not adequately explained by known prognostic factors such as age, histologic and molecular pathology, and imparted treatments. A deeper understanding of prognostic influences in diffuse glioma can facilitate therapeutic decision making and patient counselling and lend further insight into the biologic underpinnings of the disease. Magnetic resonance imaging (MRI) scans of the brain are part of standard pre-operative evaluation in these patients. As brain structure and function are modulated by biologic and environment factors, MRI-derived metrics often provide sensitive prognostic biomarkers. We have preliminarily shown that connectomics, a method of measuring brain connectivity from MRI, can accurately predict survival in patients with diffuse glioma. We will retrospectively obtain at least 1,150 datasets and attempt to validate our preliminary models (Aim 1). We will also examine connectome phenotypes associated with tumor genotypes to help refine and improve our models (Aim 2). Accurate pre- operative prediction of outcome may ultimately allow for better tailored interventions for the individual patient and assist clinicians in optimizing both tumor control and neurologic function in treatment decision making.