Development of Machine Learning Algorithms to Assess and Train Vesico-Urethral Anastomosis during Robot Assisted Radical Prostatectomy

项目概要/摘要 候选人（安德鲁J洪，医学博士）：我的长期目标是建立一个创新的培训方法的职业生涯 机器人手术，这将导致缩短外科医生的学习曲线，并最大限度地提高患者的安全性。我的第一 实现这一目标步骤侧重于理解衡量外科医生表现的客观指标， 机器学习算法如何处理这些数据来指导训练。我已经发展了一项事业 开发计划，建立在我在机器人泌尿外科手术的临床培训和以前的研究， 外科训练通过指导，奖学金和正式的课程，这个K23奖将为我提供 在我没有接受过正式培训但对我的职业生涯至关重要的三个领域， 成功：（1）机器学习;（2）外科教育;（3）先进的统计技能和研究设计。 导师团队：我的职业发展和研究计划利用现有的机构资源， 包括南加州大学机器学习中心，由共同主要导师Yan Liu博士领导;以及南加州大学凯克医院， 美国第二繁忙的机器人中心和南加州大学泌尿学研究所（由联合国领导）， 主要导师和主席Inderbir Gill博士），是几种泌尿外科手术技术的先驱者， 一个强大的研究机构，支持几位NIH资助的临床科学家。我的指导团队是 由共同导师罗伯特·斯威特博士补充，他是国防部资助的外科教育专家;职业导师 博士Larissa Rodriguez，联邦政府资助的临床医生/科学家，在指导K获奖者方面经验丰富; 教育心理学合作者Kenneth Yates博士，认知任务分析的权威; Intuitive Surgical的顾问Anthony Jarc博士支持了许多关于客观的试点数据， 性能指标。拟议的K23工作确实需要机器人专家的大力合作。 手术、教育和机器学习。研究：外科医生执行机器人的学习曲线 辅助根治性直肠癌切除术（RARP）是陡峭的：超过100例。目前的“黄金标准”手术方法 评估依赖于主观的专家审查，但这种评价既费时又不一致。 尽管如此，授予外科医生进行机器人手术的资格具有巨大的意义--病人的结果 有危险，外科医生的职业生涯也岌岌可危根据我在机器人泌尿外科手术方面的临床经验 和初步数据，我将开发一种利用机器学习（ML）算法的新方法， 客观地评估机器人外科医生的表现，并指导膀胱尿道手术的培训。 吻合术（VUA）是机器人辅助根治性直肠癌切除术（RARP）中最关键的重建部分。我 将开发和验证在VUA（目标1）期间直接从da芬奇机器人捕获的客观指标， 训练机器学习算法以评估外科医生的VUA表现（目标2），并利用ML 指导外科医生学习VUA的算法（目标3）。有了这些数据和技能从这个奖项，我将 独特地适合于利用机器学习来概括客观的外科医生评估， 泌尿外科内外的外科手术最后，本研究的结果将提供初步数据 通过NIH R 01赠款等机制获得独立资金。

PROJECT SUMMARY/ABSTRACT CANDIDATE (Andrew J. Hung, MD): My long-term goal is to establish a career in innovating training methods for robotic surgery which will lead to curtailing surgeon learning curve, and maximize patient safety. My first step towards that goal focuses on understanding objective metrics that measure surgeon performance, and how machine learning algorithms can process that data to guide training. I have developed a career development program that builds on my clinical training in robotic urologic surgery and prior research in surgical training. Through mentorship, a fellowship, and formal coursework, this K23 award will provide me the necessary support to develop expertise in 3 areas where I do not have formal training, yet are critical to my success: (1) Machine learning; (2) Surgical education; (3) Advanced statistical skills and study design. MENTORING TEAM: My career development and research plans leverage existing institutional resources, including the USC Machine Learning Center, led by co-primary mentor Dr. Yan Liu; and Keck Hospital of USC, the second busiest robotic center by volume in the United States and the USC Institute of Urology (led by co- primary mentor and chairman Dr. Inderbir Gill), home to pioneers of several urologic surgical techniques with a robust research apparatus supporting several NIH-funded clinical scientists. My mentoring team is complemented by co-mentor Dr. Robert Sweet, a DOD-funded expert on surgical education; career mentor Dr. Larissa Rodriguez, a federally funded clinician/scientist experienced in mentoring K awardees; educational psychology collaborator Dr. Kenneth Yates, an authority on cognitive task analysis; and consultant Dr. Anthony Jarc, at Intuitive Surgical who has supported much of the pilot data on objective performance metrics. The proposed K23 work truly requires the robust collaboration of experts in robotic surgery, education, and machine learning. RESEARCH: The learning curve for surgeons performing robot assisted radical prostatectomy (RARP) is steep: over 100 cases. Current ‘gold standard’ methods of surgical assessment rely on subjective expert review, but such evaluations are time consuming and inconsistent. Nonetheless, credentialing a surgeon to perform robotic surgery has enormous implications - patient outcomes are at risk, and a surgeon’s career is on the line. Informed by my clinical expertise in robotic urological surgery and preliminary data, I will develop a novel method of utilizing machine learning (ML) algorithms to objectively assess robotic surgeon performance and to guide training for the vesico-urethral anastomosis (VUA), the most critical reconstructive part of the robot-assisted radical prostatectomy (RARP). I will develop and validate objective metrics directly captured from the da Vinci robot during the VUA (Aim 1), train machine learning algorithms to assess a surgeon’s performance of VUA (Aim 2), and utilize ML algorithms to guide surgeons learning the VUA (Aim 3). Armed with these data and skills from this award, I will be uniquely suited to utilize machine learning to generalize objective surgeon assessment for robot-assisted surgical procedures within and beyond urology. Finally, the results from this study will provide preliminary data for independent funding through mechanisms such as an NIH R01 grant.