Pathomics biomarkers for stratification of clear cell kidney cancers

项目摘要 没有对癌症的病理属性，如组织学和肿瘤生长模式进行定量评估 到目前为止。在每种癌症类型中，这些参数都会影响患者的预后，并包含在 肿瘤复发和总存活率。使用机器学习和卷积神经网络的算法 使我们能够量化病理和开发病理组学生物标记物。在这里，我们建议获得病理组学 计数透明细胞组织学生长模式(HGP)的癌症复发/进展生物标志物 肾细胞癌(CcRCC)。CcRCC是肾癌最常见的亚型。在其本地化阶段，它是 接受肾切除的治疗。然而，约30%的患者在手术后经历疾病进展和 可能受益于辅助治疗。决定是否有必要进行治疗需要确定 有高复发风险的患者。在这里，我们假设定量生物标记物将会改善 对慢性肾细胞癌患者的风险评估并建议开发计算机生成的肿瘤特征 增长模式。我们之前定义了13个HGP，并展示了它们预测总存活率的能力 接受慢性肾细胞癌治疗的患者。将使用以下工具生成和验证每个HGP的独特功能 卷积神经网络的框架，产生跨癌症区域表达的概率。 此外，概率分布将被用于获得每个HGP表达的生物标记物。vbl.使用 参数模型和非参数模型、HGP-生物标记物与肿瘤的相关性将被检验 肾细胞癌进展的分期和局部机制，如癌栓形成、区域淋巴转移 肾周脂肪组织的转移或侵袭。每种算法在项目中的性能将 被评估。总之，在这个项目中开发的生物标志物将为选择患者提供一个起点 术后辅以ccRCC治疗。

Project Summary Pathologic attributes of cancers, such as histology and tumor growth patterns are not quantitatively assessed to date. In every cancer type these parameters effect patient outcomes and are included in risk models of tumor recurrence and overall survival. Algorithms using machine learning and convolutional neural networks allow us to quantify pathology and develop Pathomics biomarkers. Here, we propose to obtain pathomics biomarkers of cancer recurrence/progression that enumerate histology growth patterns (HGPs) in clear cell renal cell cancer (ccRCC). ccRCC is the most common subtype of kidney cancer. In its localized stage, it is treated by nephrectomy. However, about 30% of patients experience disease progression after surgery and may benefit from adjuvant treatment. Deciding whether or not treatment is warranted requires identifying patients who are at a high risk of recurrence. Here, we hypothesize that quantitative biomarkers will improve the risk assessment of patients with ccRCC and propose to develop computer-generated features of tumor growth patterns. We previously defined 13 HGPs and demonstrated their ability to predict overall survival in patients treated for ccRCC. Distinctive features for each HGP will be generated and validated using frameworks of convolutional neural networks that produce probabilities of expression across cancer regions. Further, the distribution of probabilities will be used to obtain biomarkers of expression of each HGP. Using parametric and non-parametric models, HGP-biomarkers will be examined for their association with tumor stage and local mechanisms of ccRCC progression, such as formation of tumor thrombi, regional lymph node metastases or invasion into perinephric adipose tissues. The performance of each algorithm in the project will be evaluated. Altogether, biomarkers developed in this project will provide a starting point to select patients with ccRCC for adjuvant treatment after surgery.