STTR Phase I: A Fully-Automated Endoscopic Scoring System for Ulcerative Colitis

该STTR一期项目的广泛/商业影响将开发一种自动化系统，显著提高溃疡性结肠炎（UC）患者疾病严重程度评估的速度、可靠性和准确性。这种使人终生衰弱的疾病影响了近100万美国患者，开发有效的治疗方法需要准确、可靠和及时的评分。目前，fda批准的主要诊断是梅奥评分的内窥镜部分，临床试验需要由专业胃肠病学家进行耗时且资源有限的专家阅读。拟议的技术将有多种用途：加快评分过程以确定患者是否有资格进行药物试验，测量基线和疾病变化以获得UC药物试验的FDA终点，并为GI医生提供特定治疗对特定患者有效性的见解。这种系统评分——在几分钟而不是几天内完成——将提高效率，加快招募和保留试验参与者（药物试验中的最大挑战）。据估计，每年花费在专家阅读结肠镜检查视频以进行药物试验上的费用为5600万美元，这项技术不仅可以节省时间，还可以显著降低成本。此外，该方法将转化为临床护理，为社区医生提供自动化胃肠道专业知识和对疾病进展和患者对治疗反应的宝贵见解。最后，该技术可作为医学生和GI住院医师的教学工具。这个STTR一期项目旨在创建一个用于UC监测的结肠镜视频评估的自动化系统。这项研究的独特创新因素是对结肠镜检查视频中所有可用数据的自动处理，以创建可靠、可重复、有效和定量的UC疾病负担评估。该方法结合了有效的信息帧分类器、位置估计系统和疾病严重程度分类器来生成整个视频的评分。对临床捕获的视频进行自动、全面、基于机器学习的评估的算法是该系统的基础。该项目将提高(1)信息视频帧与非信息视频帧之间的分类准确性，(2)摄像机位置的估计，以及(3)针对来自多个临床提供者和不同制造商的结肠镜的异质结肠镜视频数据集验证系统。该算法将针对内窥镜评估和评分系统进行优化，并通过与学术合作伙伴的持续数据收集进行扩展。该项目将产生一种新的医学视频分析方法，使用有效的机器学习方法来创建一个实用的、数据驱动的解决方案，用于评估和改进UC护理。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The broader/commercial impact of this STTR Phase 1 project will develop an automated system that significantly increases the speed, reliability, and accuracy of disease severity assessment in ulcerative colitis (UC) patients. This lifelong debilitating disease impacts almost 1 M US patients, and developing effective treatments requires accurate, reliable, and timely scoring. Currently, the FDA-approved primary diagnostic is the endoscopic component of the Mayo score, and clinical trials require a time-consuming, resource-constrained process of expert reading by specialized gastroenterologists. The proposed technology will serve multiple purposes: expediting the scoring process to determine patient eligibility for drug trials, measuring baseline and disease change to obtain FDA endpoints for UC drug trials, and providing insights that inform GI physicians on the effectiveness of a particular therapy for a specific patient. This system scoring – performed in minutes, rather than days – will improve efficiency and expedite recruitment and retention of trial participants (the greatest challenge in drug trials). With an estimated $56 M spent annually on expert reading of colonoscopy videos for drug trials, this technology would not only save time but would also significantly reduce costs. In addition, this methodology will translate into clinical care, providing community physicians with automated GI expertise and valuable insight into disease progression and patient response to therapy. Finally, this technology could be utilized as a teaching tool for medical students and GI residents. This STTR Phase I project is designed to create an automated system for video assessment of colonoscopies taken for UC monitoring. The unique innovative factor in this research is the automated processing of all data available from colonoscopy videos to create a reliable, repeatable, efficient, and quantitative assessment of the burden of UC disease. The approach uses a combination of an effective informative frame classifier, location estimation system, and disease severity classifier to generate scoring of the entire video. Algorithms for automated, comprehensive, machine-learning-based assessment of clinically-captured videos are the foundation of the system. The project will improve (1) classification accuracy between informative vs. non-informative video frames, (2) estimation of the camera location, and (3) validate the system against a heterogenous colonoscopy video dataset from multiple clinical providers and from colonoscopes from various manufacturers. The algorithms will be optimized for an endoscopic assessment and scoring system and extended through ongoing data collection with academic partners. This project will result in a novel approach to medical video analysis using effective machine learning methods to create a practical, data-driven solution for assessment and improvement of UC care.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.