Video-based Methods for Automated Surgical Training and Skill Assessment

Abstract

Training physicians in technical skills is a critical aspect of medical education, essential for ensuring patient safety and the readiness of new doctors for independent practice. However, traditional methods of skill assessment, which rely heavily on expert observation, are time-consuming and prone to subjectivity, even when structured rubrics are used. This issue is exacerbated by the growing demand for competency-based medical education (CBME), which requires frequent and detailed assessments, further burdening medical educators. To address these challenges, I developed a novel, video-based, deep learning approach for assessing technical skills in surgical procedures. Specifically, this work in this thesis focused on percutaneous procedures, areas that have been largely neglected in current research. The core of this approach lies in leveraging object detection and motion-based metrics to evaluate the usage and movement of surgical instruments, offering a more accessible and scalable alternative to traditional tracking systems. The primary aim of this research was to develop and validate computer-based training and skill assessment systems that leverage video-based methods to provide objective performance evaluation. Key contributions include the design of training platforms that simultaneously support skill development and serve as data collection tools to advance research in computer-assisted skill assessment. Additionally, this work demonstrated the feasibility of using video-based motion analysis for skill evaluation by developing and deploying a system in a low-resource clinical setting. Finally, the effectiveness of these methods was validated by comparing video-based assessments to traditional motion-tracking techniques and expert evaluations, ensuring their reliability as scalable alternatives to conventional assessment methods. By focusing on underrepresented procedures and combining state-of-the-art deep learning methods with practical, real-world applications, this research bridges existing gaps in the field of computer-assisted skill assessment, ultimately contributing to more objective, efficient, and scalable training tools for medical professionals.