Multiparametric Magnetic Resonance Imaging Artificial Intelligence Pipeline for Oropharyngeal Cancer Radiotherapy Treatment Guidance

口咽癌（OPC）是少数几种发病率上升的国内癌症之一，主要是由于 人乳头瘤病毒（HPV）感染率增加。放射学图像对于评估 OPC和援助的疾病检测和放射治疗（RT）。然而，传统的RT计划 成像需要依赖于操作者的肿瘤分割，这是治疗误差的主要来源 导致对正常组织的非预期剂量和随后的使人衰弱的口腔-牙齿后遗症。HPV+ OPC表达不同的肿瘤/淋巴结中RT反应（快速反应）率，导致显著的 计划和输送的RT剂量之间的差异。此外，对于治疗期间HPV+ OPC患者， 在抵抗子体积中，正常组织保留的程度取决于残余活性物质的位置。 疾病多参数MRI（mpMRI）技术结合了同时进行的高维解剖成像， 功能信息与人工智能（AI）方法相结合可以改善临床决策 通过为适应性RT计划提供可立即采取行动的临床依据来支持OPC。的 该F31项目的假设是，mpMRI技术和AI算法将有助于分割、快速 反应预测和OPC的治疗内耐药分类。为了验证这一假设，我将首先 使用mpMRI开发AI模型，以准确分割原发性肿瘤和转移性颈部淋巴结 并将模型与人类专家进行基准测试（具体目标1）。接下来，我将研究 快速治疗应答者和非应答者的原发性肿瘤/淋巴结之间的mpMRI以及随后的使用 AI构建响应预测模型（具体目标2）。最后，我将描述原发性肿瘤的区域 mpMRI上区域和体素水平的治疗阻力，随后使用AI建立阻力 分类模型（具体目标3）。通过F31奖项中提出的专门培训，我将获得 临床决策支持工具实施和设计的专业知识（培训目标1），制定方法 医学成像深度学习的创新（培训目标2），获得统计建模方面的专业知识， 临床信息学方法（培训目标3），从研究生研究过渡到指导研究生研究，最终成为独立的主要研究者（培训目标4）。成功 为了完成我提出的具体目标，实现我的培训目标，我组建了一个专门的小组， 导师和合作者将在整个项目期间为我提供出色的指导。 此外，该项目将在MD安德森癌症中心进行，这是一个国际知名的癌症 该机构拥有世界上一些最大的头颈癌患者成像数据集。 因此，我具有独特的地位，通过这个F31奖在这个研究领域进行开创性的工作。

Oropharyngeal cancer (OPC) is one of the few domestic cancers that is rising in incidence, primarily due to increased human papillomavirus (HPV) infection rates. Radiographic images are crucial for assessment of OPC and aid in disease detection and radiotherapy (RT) treatment. However, RT planning with conventional imaging requires operator-dependent tumor segmentation, which is the primary source of treatment error leading to unintended dose to normal tissues and subsequent debilitating oro-dental sequelae. Further, HPV+ OPC expresses differential tumor/node mid-RT response (rapid response) rates, resulting in significant differences between planned and delivered RT dose. Moreover, for HPV+ OPC patients with intra-treatment resistant sub-volumes, the degree of normal tissue sparing is dependent on the location of residual active disease. Multiparametric MRI (mpMRI) techniques that incorporate simultaneous high-dimensional anatomical and functional information coupled to artificial intelligence (AI) approaches could improve clinical decision support for OPC by providing immediately actionable clinical rationale for adaptive RT planning. The hypothesis of this F31 project is that mpMRI techniques and AI algorithms will facilitate segmentation, rapid response prediction, and intra-treatment resistance classification of OPC. To test this hypothesis, I will first develop an AI model using mpMRI to accurately segment primary tumors and metastatic cervical lymph nodes and benchmark the model against human experts (Specific Aim 1). Next, I will investigate the differences in mpMRI between primary tumors/nodes of rapid therapy responders and non-responders and subsequently use AI to build a response prediction model (Specific Aim 2). Finally, I will characterize areas of primary tumor treatment resistance at the regional and voxel level on mpMRI and subsequently use AI to build a resistance classification model (Specific Aim 3). Through dedicated training proposed in this F31 award, I will gain expertise in clinical decision support tool implementation and design (Training Goal 1), develop methodological innovations for deep learning in medical imaging (Training Goal 2), gain expertise in statistical modeling and clinical informatics approaches (Training Goal 3), and transition from graduate research to mentored post-graduate research and eventual independent principal investigator status (Training Goal 4). To successfully complete my proposed specific aims and achieve my training goals, I have assembled a dedicated group of mentors and collaborators that will provide me with excellent guidance throughout this project period. Moreover, this project will take place at MD Anderson Cancer Center, an internationally renowned cancer institution that is home to some of the largest imaging datasets of head and neck cancer patients in the world. Therefore, I am uniquely positioned to conduct pioneering work in this research space through this F31 award.