Quantitative PET/CT Analysis to Improve Evaluation of Tumor Response

定量 PET/CT 分析可改善肿瘤反应评估

• 批准号: 8578743
• 负责人: Wei Lu
• 金额: $ 31.85万
• 依托单位: UNIVERSITY OF MARYLAND BALTIMORE
• 依托单位国家: 美国
• 财政年份: 2013
• 资助国家: 美国
• 起止时间: 2013-07-01 至 2017-04-30
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/8578743
• 关键词: Algorithms; Anatomy; Caliber; Cancer Patient; Characteristics; Clinical; Clinical Trials; Complement; Diagnostic; Diagnostic tests; Esophageal Neoplasms; Evaluation; Failure; Future; Goals; Health Sciences; Image; Image Analysis; Institution; Logistic Regressions; Machine Learning; Malignant Neoplasms; Malignant neoplasm of esophagus; Manuals; Maryland; Measures; Metabolic; Methods; Modality; Modeling; Morbidity - disease rate; Operative Surgical Procedures; Oregon; Outcome; Pathologic; Patients; Physicians; Positron-Emission Tomography; Property; Quality of life; Receiver Operating Characteristics; Relative (related person); Reporting; Sensitivity and Specificity; Site; Spatial Distribution; Specificity; Staging; Survival Rate; Techniques; Testing; Texture; Universities; Validation; Variant; Work; X-Ray Computed Tomography; base; chemoradiation; clinical practice; cost; demographics; forest; image registration; improved; mortality; novel; predictive modeling; prospective; public health relevance; response; tool; treatment strategy; tumor; uptake

DESCRIPTION (provided by applicant): We propose to develop and validate a quantitative Positron Emission Tomography / Computed Tomography (PET/CT) image analysis framework to improve the evaluation of esophageal tumor response to chemoradiotherapy (CRT) in patients with locally advanced esophageal cancer. In Aim 1, we will extract comprehensive spatial and temporal features of a tumor from PET/CT images and evaluate their ability in predicting tumor response to CRT. These features will quantify the spatial characteristics of a tumor along with their changes due to CRT, adding a great amount of information to the current non-volumetric PET/CT response measures. Also, we will use image registration techniques to align pre-CRT images with post-CRT images, making it possible to quantify the spatial changes at the original tumor site. In Aim 2, we will construct and test reliable predictive models of tumo response to CRT using machine learning techniques with spatial- temporal PET/CT features, clinical parameters and demographics as input. The models will identify an optimal subset of predictive features and utilize PET and CT information in chorus. In Aim 3, we will develop a novel multi-modality adaptive region-growing algorithm for tumor delineation in PET/CT. We will compare the accuracy and precision of the resulting predictive models against those in which tumor is delineated using conventional methods (manual contouring or thresholding). This comparison will help us understand to what degree the prediction of tumor response depends on the tumor delineation methods. Finally, we will use pathologic response and survival as the end points and ground truth to cross-validate each predictive model. If all aims are achieved, the proposed PET/CT image analysis framework may provide a highly accurate diagnostic tool. This, will complement other diagnostic tests in assisting physicians in making a treatment decision to more appropriately select patients for surgery, thus avoiding the mortality and morbidity of surgery in responders for whom surgery can be safely deferred; while improving local control and survival in non- responders for whom surgery should be considered. Therefore, this work has the potential to improve outcomes by safely deferring surgery which will improve our locally advanced esophageal cancer patient's quality of life while simultaneously reducing costs.

描述（由申请人提供）：我们建议开发并验证定量正电子发射断层扫描/计算机断层扫描（PET/CT）图像分析框架，以改善局部晚期食管癌患者食管肿瘤对放化疗（CRT）反应的评估。在目标 1 中，我们将从 PET/CT 图像中提取肿瘤的全面空间和时间特征，并评估它们预测肿瘤对 CRT 反应的能力。这些特征将量化肿瘤的空间特征以及 CRT 引起的变化，为当前的非体积 PET/CT 反应测量添加大量信息。此外，我们将使用图像配准技术将 CRT 前图像与 CRT 后图像对齐，从而可以量化原始肿瘤部位的空间变化。在目标 2 中，我们将使用机器学习技术，以时空 PET/CT 特征、临床参数和人口统计数据作为输入，构建并测试 CRT 肿瘤反应的可靠预测模型。这些模型将识别预测特征的最佳子集，并同时利用 PET 和 CT 信息。在目标 3 中，我们将开发一种新颖的多模态自适应区域生长算法，用于 PET/CT 中的肿瘤描绘。我们将比较所得预测模型与使用传统方法（手动轮廓或阈值处理）描绘肿瘤的预测模型的准确性和精确度。这种比较将帮助我们了解肿瘤反应的预测在多大程度上取决于肿瘤描绘方法。最后，我们将使用病理反应和生存作为终点和基本事实来交叉验证每个预测模型。如果所有目标都实现了，所提出的 PET/CT 图像分析框架可能会提供高度准确的诊断工具。这将补充其他诊断测试，协助医生做出治疗决定，以更适当地选择接受手术的患者，从而避免可以安全推迟手术的反应者的手术死亡率和发病率；同时改善应考虑手术的无反应者的局部控制和生存。因此，这项工作有可能通过安全推迟手术来改善结果，这将改善我们局部晚期食管癌患者的生活质量，同时降低成本。