How to Bring Surgery to the Next Level: Interpretable Skills Assessment in Robotic-Assisted Surgery Available to Purchase

Introduction: A surgeon’s technical skills are an important factor in delivering optimal patient care. Most existing methods to estimate technical skills remain subjective and resource intensive. Robotic-assisted surgery (RAS) provides a unique opportunity to develop objective metrics using key elements of intraoperative surgeon behavior which can be captured unobtrusively, such as instrument positions and button presses. Recent studies have shown that objective metrics based on these data (referred to as objective performance indicators [OPIs]) correlate to select clinical outcomes during robotic-assisted radical prostatectomy. However, the current OPIs remain difficult to interpret directly and, therefore, to use within structured feedback to improve surgical efficiencies. Methods: We analyzed kinematic and event data from da Vinci surgical systems (Intuitive Surgical, Inc., Sunnyvale, CA, USA) to calculate values that can summarize the use of robotic instruments, referred to as OPIs. These indicators were mapped to broader technical skill categories of established training protocols. A data-driven approach was then applied to further sub-select OPIs that distinguish skill for each technical skill category within each training task. This subset of OPIs was used to build a set of logistic regression classifiers that predict the probability of expertise in that skill to identify targeted improvement and practice. The final, proposed feedback using OPIs was based on the coefficients of the logistic regression model to highlight specific actions that can be taken to improve. Results: We determine that for the majority of skills, only a small subset of OPIs (2–10) are required to achieve the highest model accuracies (80–95%) for estimating technical skills within clinical-like tasks on a porcine model. The majority of the skill models have similar accuracy as models predicting overall expertise for a task (80–98%). Skill models can divide a prediction into interpretable categories for simpler, targeted feedback. Conclusion: We define and validate a methodology to create interpretable metrics for key technical skills during clinical-like tasks when performing RAS. Using this framework for evaluating technical skills, we believe that surgical trainees can better understand both what can be improved and how to improve.