Administrative Supplement to Support Collaborations to Improve AIML-Readiness of NIH-Supported Data for Parent Award SCH: Personalized Rescheduling of Adaptive Radiation Therapy for Head & Neck Cancer

项目摘要 在我们的母公司奖项（1 R 01 CA 257814 -01）下，我们收集了一个系列多参数数据库 磁共振（MR）图像以及患者报告的和客观的毒性指标， 400例头颈部（HNC）患者，治疗前、治疗中和治疗后。我们计划利用这些数据， 与父母奖励的第一个具体目标保持一致，以有效量化治疗相关反应 在肿瘤/淋巴结和正常组织上，以开发个性化的治疗计划适应性， 个别HNC患者。然而，据我们所知，作为数据最丰富的图像毒性队列， 该数据库需要严格的管理才能用于人工智能/机器学习（AI/ML） 预测例如肿瘤并发症概率（TCP）和正常组织并发症的方法 概率（NTCP）。具体地，对感兴趣的肿瘤和正常组织区域的多观察者分割是 必需的.此外，传播工作需要吸引AI/ML社区的专家， 为自动分割模型和TCP/NTCP预测开发AI/ML方法。为此，我们计划 实现三个具体目标。通过我们的第一个具体目标，我们计划策划我们的系列多参数， 治疗反应和TCP的多时间点MRI数据集（伴随提取的放射组学） 通过组建一个由三名医生组成的团队来进行预测，以获得地面实况分割图像。我们 进一步计划存款的策划分割图像作为数据集的癌症成像档案（TCIA）。作为 我们的第二个具体目标，我们计划策展和公共沉积匹配的图像剂量多时间点 急性和晚期毒性指标将分发给AI/ML专家进行NTCP建模。我们将 尤其包括患者报告的MD安德森症状量表-头颈部（MDASI-HN）毒性 结局、常见毒性标准-不良事件（CTC-AE）、医生分级毒性和目的 吞咽功能障碍的测量，如改良钡吞咽和管饲评估。在 第三个具体目标，我们计划设计和执行一个公共众包的系列图像剂量的挑战- TCP和NTCP预测建模任务的响应预测。基于我们计划的测试数据集， 为了在挑战执行后发布，我们将对提交的 模型（例如，假阳性和假阴性病例），并将最佳结果作为手稿传播， 提交出版物和演示文稿。如果成功，拟议的努力将直接响应 需要AI/ML就绪数据集用于癌症治疗。

Project Summary We have collected, under our parent award (1R01CA257814-01), a database of serial multi-parametric magnetic resonance (MR) images as well as patient-reported and objective toxicity measures for more than 400 head and neck (HNC) patients, at pre-, on-, and post-therapy. We plan to utilize this data, in complete alignment with the first specific aim of the parent award, to effectively quantify treatment-related response on tumor/node and normal tissue in order to develop personalized treatment planning adaptations for individual HNC patients. As the most data-rich image toxicity cohort to the best of our knowledge, however, this database necessitates rigorous curation to be utilized for artificial intelligence/machine learning (AI/ML) approaches to predict, for example, tumor complication probability (TCP) and normal tissue complication probability (NTCP). Specifically, multi-observer segmentation of tumor and normal tissue regions of interest is required. Additionally, dissemination efforts are necessary to engage experts from AI/ML communities to develop AI/ML-approaches for auto-segmentation models, and TCP/NTCP predictions. To this end, we plan to undertake three specific aims. Through our first specific aim, we plan to curate our serial multi-parametric, multi time-point MRI dataset (accompanied with extracted radiomics) for therapeutic response and TCP prediction through assembling a team of three physicians to obtain the ground-truth segmented images. We further plan to deposit the curated segmented images as a dataset to The Cancer Imaging Archive (TCIA). As our second specific aim, we plan for curation and public deposition of matched image-dose multi-time-point acute and late toxicity metrics to be disseminated to both AI/ML experts for NTCP modeling. We will particularly include patient-reported MD Anderson Symptom Inventory-Head and Neck (MDASI-HN) toxicity outcomes, Common Toxicity Criteria- Adverse Events (CTC-AE) physician-ranked toxicity, and objective measures of swallowing dysfunction such as modified barium swallowing and tube-feeding assessments. In the third specific aim, we plan to design and execute a public crowdsourced challenge for serial image dose- response prediction for both TCP and NTCP prediction modeling tasks. Based on the test dataset that we plan to release after the execution of the challenge, we will conduct a post-challenge analysis on the submitted models (e.g., false-positive, and false-negative cases), and disseminate the best results as manuscripts to be submitted for publications and presentations. If successful, the proposed efforts are directly responsive to the need for AI/ML-ready datasets to be utilized for cancer treatment.