Assessing, Optimizing, and Delivering Surgical Feedback with AI for Benign Urological Robotic Surgeries

ABSTRACT The feedback exchange between trainer and surgical trainee is critical for determining the rate of skill acquisition and subsequently improving patient outcomes. However, such feedback is inconsistent during formal surgical training and almost non-existent once training is complete, when surgeons may yet become fully proficient. For complex robotic procedures, surgeons may expect learning curves of up to 100-400 cases before achieving stable patient outcomes. Optimized surgical training, through real-time feedback measured objectively for effectiveness, can shorten the path toward surgical proficiency and ultimately decrease patient morbidity. To address this need for consistent and effective feedback, we will assess the present exchange of surgical dialogue throughout 500 benign urological training cases, recorded in the live operating room among five nationally recognized medical centers (Aim 1). Benign urological surgeries serve as an ideal test case for systematically categorizing and assessing live surgical feedback from a trainer to surgical trainee. Through the development of robust statistical tools and performance of randomized control trials in our training lab, we will investigate how to optimize human-delivered feedback (Aims 1-2). We will simultaneously harness artificial intelligence (AI) to automatically identify the need for and deliver video-based surgical feedback to novice surgeons in the lab (Aim 3), filling the observable gap during training and beyond. PI Hung and his team have two active R01 grants based on developing AI tools to assess surgeon performance, providing basic skills evaluation automatically and in near real-time. This proposal seeks to discover how feedback can be delivered back to the surgical trainee. PI Hung has piloted a system to trigger feedback in response to the assessment of low skill, in collaboration with AI scientist Co-I Huang and psychologist Shtulman, demonstrating as well in their preliminary work that such feedback can meaningfully improve surgical action. We hypothesize that 1) human-delivered live surgical feedback can be optimized, and 2) AI-delivered feedback is effective in improving surgical performance. Despite its critical importance for surgical training, understanding and improving surgical feedback has received little investment. Our work, if successful, will create a framework by which surgical educators’ live instruction can be objectively assessed, allowing for downstream, data-driven optimization of surgical training. Our AI-based automated feedback system will provide maturing surgeons access to consistent and effective surgical feedback in the training lab, promising to reduce procedure learning curves and ultimately improve patient outcomes.