Multi-modal machine learning to guide adjuvant therapy in surgically resectable colorectal cancer

项目摘要/摘要 结直肠癌(CRC)是第三大常见的恶性肿瘤和第二大致病原因。 世界范围内的癌症死亡。对准确、经济高效和可广泛获取的风险的需求尚未得到满足- 分层工具，以识别复发风险增加的患者，这些患者最有可能受益于 辅助治疗。目前的护理标准风险分层方法是不够的。每一次结直肠癌手术 候选人接受肿瘤的病理和放射学评估；这两种方式代表了丰富的， 为开发新的风险分层工具提供容易获得且迄今未得到充分利用的资源。深沉 学习(DL)已经显示出极大的潜力，可以提高医生在越来越广泛的诊断方面的能力 以及病理学、放射学和临床医学的预后任务。我们假设应用集成 对多模式(病理、放射学和电子病历(EMR))数据的基于DL的分析将产生 大大改善了结直肠癌患者的分层辅助治疗计划。我们计划建造第一座 全面的、公开可用的、专家注释的多模式CRC数据集，用于深度学习，包括 结直肠癌病理全片图像(WSI)、术前CT和MRI图像及结构化临床 电子病历数据。使用此数据集，我们将开发用于风险分层的单通道和多通道DL模型 手术可切除(I-III期)的结直肠癌患者。为了检验我们的假设，我们将比较 多通道模型与单通道模型和现有的分层方法。这个项目 受益于跨领域的独特跨学科团队的互补专业知识和资源 机器学习、病理学、放射学和肿瘤学。

Project Summary / Abstract Colorectal cancer (CRC) is the third most commonly diagnosed malignancy and the second leading cause of cancer death worldwide. There is an unmet need for accurate, cost-efficient, and broadly accessible risk- stratification tools to identify patients at increased risk of recurrence , who are most likely to benefit from adjuvant therapy. Current standard-of-care risk stratification approaches are inadequate. Every CRC surgical candidate undergoes pathologic and radiologic evaluation of their tumor; these two modalities represent a rich, readily accessible and, thus far, underutilized resource for developing new risk-stratification tools. Deep learning (DL) has demonstrated great potential for augmenting physicians on an increasing range of diagnostic and prognostic tasks in pathology, radiology, and clinical medicine. We hypothesize that applying integrated DL-based analysis to multimodal (pathologic, radiologic, and electronic medical record (EMR)) data will yield greatly improved stratification of CRC patients for adjuvant treatment planning. We propose to build the first comprehensive, publicly-available, expert-annotated multimodal CRC dataset for deep learning, containing annotated CRC pathology whole-slide images (WSI), preoperative CT and MRI images, and structured clinical EMR data. Using this dataset, we will develop both single and multi-modality DL models for risk stratification of surgically-resectable (Stage I-III) CRC patients.To test our hypothesis, we will compare the performance of multi-modality models with that of single-modality models and existing methods of stratification. This project benefits from the complementary expertise and resources of a unique interdisciplinary team spanning the fields of machine learning, pathology, radiology, and oncology.