Automated and Objective Assessment of Surgical Training: Detection of Procedural Steps on Videotaped Performances

With the rapid growth in competency-based performance requirements for medical resident education, there is a critical need for validated assessments of technical skills. Skill evaluation today is predominantly based on manual evaluations by expert surgeons-which is time-consuming, non-uniform, and cumbersome to work into the natural flow of training or testing events. In this paper, we propose an algorithm for automated recognition of surgical procedural steps using video analysis for the objective assessment of technical skills during surgical training. We employ a bouquet of computer vision techniques, such as template matching, for the automated detection of correct and incorrect surgical procedural steps during tracheoesophageal fistula repair. We consider specifically a simulated model of human infant anatomy used to train surgeons to perform the tracheoesophageal fistula repair procedure. Using a simulated model is key for gaining expertise in repairing this relatively rare but deadly abnormality. Our automated detection approach provides an appropriate, clinically-relevant scenario using a well-designed and validated simulator and produces a uniform, manageable and verifiable solution. The algorithm was validated on nine performances of thoracoscopic tracheoesophageal fistula ligation from surgeons with a broad range of surgical skills. The algorithm result imitates the groundtruth for the evaluations, and thus demonstrates the feasibility of the proposed work for efficient, practical and objective assessment of surgical skill during training.