Lymph Node Quantification System for Multisite Clinical Trials

项目摘要/摘要 在淋巴瘤和其他癌症患者中，对肿瘤负担程度的定量评估是必要的。 对分期、再分期和治疗反应或复发的评估很重要；但总体测量 用目前的工具，尤其是当淋巴结汇合或难以完全覆盖时，肿瘤负担是具有挑战性的。 与周围的建筑区分开来。精确医学和新疗法正在强调这一需要 引入风险适应的方法，为癌症患者量身定做适当的治疗策略。有能力 用功能和解剖成像定量评估癌症表型，可以有效地和主动地评估肿瘤表型。 精确地将患者映射到基因表达谱、临床信息、匹配队列和新疗法 这些方案可能会导致对癌症患者进行更优化的管理。 这一学术-产业合作伙伴关系旨在将最近开发的技术转化为半 将淋巴结的自动图像分割和量化放入强大的工具中，并将它们集成到 现有的基于云的多中心肿瘤学临床试验管理系统。有能力半- 利用计算机断层扫描(CT)自动分割淋巴结病理，并对结节组织进行量化。 利用正电子发射断层扫描(PET)的代谢将能够全面追踪形态和 与疾病进展和治疗反应相关的功能变化。 自2004年以来，Dana-Farber/哈佛癌症中心(DF/HCC)的肿瘤成像度量核心(TIMC) 已经开发了Precision Image Metrics，LLC(PIM)平台来管理临床试验图像评估 工作流程。目前，TIMC中有近50,000个持续测量的淋巴结数 数据库。PIM系统每年用于在8个NCI进行超过20,000个时间点的成像评估- 指定的癌症中心，目标是通过过渡到完全云托管的系统来快速增长。 如果有足够的训练数据，最先进的机器学习和人工智能(AI)技术- 在特定的成像分析任务上，GES可以达到甚至超过人类的表现。最近的研究表明 表明基于人工智能的CT扫描淋巴结分割正在接近人类的表现水平，并且 我们将扩展这项工作，并将其转化为商业工具。具体地说，我们的目标是翻译最近的广告- 将基于AI的细分功能增强为可部署的服务，并将这些服务集成到临床 审判工作流程。拟议系统的设计将纳入图像分析提供的专家反馈-- LYST和放射科医生返回地面真实数据集，允许不断提高准确性和 临床验收。我们将扩展我们的半自动CT分割技术来量化淋巴结 PET/CT中的代谢，以淋巴瘤为模型疾病。这些技术与PIM的集成将 提供整个癌症中心网络中持续测量的量化数据来源。

Project Summary / Abstract In patients with lymphomas and other cancers, quantitative evaluation of the extent of tumor burden is im- portant for staging, restaging, and assessment of therapeutic response or relapse; yet measurement of overall tumor burden is challenging with current tools, particularly when lymph nodes are confluent or difficult to fully differentiate from surrounding structures. Precision medicine and novel therapeutics are emphasizing the need to introduce a risk-adapted approach to tailor appropriate treatment strategies for cancer patients. The ability to quantitatively assess cancer phenotypes with functional and anatomical imaging that could efficiently and ac- curately map patients to gene expression profiling, clinical information, matching cohorts, and novel treatment regimens could potentially result in more optimal management of patients with cancer. This Academic-Industry Partnership aims to translate recently developed technologies for semi- automated image segmentation and quantification of lymph nodes into robust tools and integrate them into an existing cloud-based system for management of multicenter oncology clinical trials. The ability to semi- automatically segment lymph node pathology with computed tomography (CT), as well as quantify nodal me- tabolism with positron emission tomography (PET) will enable comprehensive tracking of morphological and functional changes related to disease progression and treatment response. Since 2004, the Dana-Farber/Harvard Cancer Center's (DF/HCC) Tumor Imaging Metrics Core (TIMC) has developed the Precision Imaging Metrics, LLC (PIM) platform to manage clinical trial image assessment workflows. Currently, there are nearly 50,000 consistently measured lymph node measurements in the TIMC database. The PIM system is used to make over 20,000 time point imaging assessments per year at eight NCI- designated Cancer Centers and aims to grow quickly by transitioning to a fully cloud-hosted system. Given sufficient training data, state-of-the-art machine learning and artificial intelligence (AI) technolo- gies can meet or even exceed human performance on specific imaging analysis tasks. Recent studies have indicated that AI-based lymph node segmentation from CT scans is nearing human performance levels, and we will extend and translate this work into a commercial tool. Specifically, our aim is to translate recent ad- vancements in AI-based segmentation into deployable services, and integrate these services into the clinical trial workflow. The proposed system will be designed to incorporate expert feedback provided by image ana- lysts and radiologists back into the ground truth dataset, allowing for continuous improvement in accuracy and clinical acceptance. We will extend our semi-automatic CT segmentation technologies to quantify lymph node metabolism in PET/CT, using lymphoma as the model disease. Integration of these technologies with PIM will provide an ongoing source of consistently measured quantitative data across a network of cancer centers.