Accelerated Video-Based Surgical Skills Assessment through Saliency-Guided Surgical Highlights Reel

Marzie Lafouti, Liane S. Feldman, Amir Hooshiar

Keywords: Surgical skill assessment, Video-based artificial intelligence, Saliency-guided highlight reels, Laparoscopic surgery

Abstract: Objective evaluation of surgical skill is critical for training, credentialing, and quality assurance. However, current approaches rely heavily on expert raters manually reviewing surgical videos, which is time-consuming, labour-intensive, and provides delayed feedback. These limitations restrict the scalability of performance assessment and slow the feedback cycle for trainees [1,2]. Video-based artificial intelligence (AI) offers a promising pathway to automate parts of this process while maintaining clinician oversight. To address these gaps, we developed an interpretable spatiotemporal AI pipeline that classifies surgical expertise directly from operative video while generating saliency-based “highlight reels”. These highlights reveal the critical frames and time points most influential to the model’s inference.

Submission Number: 50