Video is an effective medium for learning procedural knowledge, such as surgical techniques. However, learning procedural knowledge through videos remains difficult due to limited access to procedural structures of knowledge (e.g., compositions and ordering of steps) in a large-scale video dataset. We present Surch, a system that enables structural search and comparison of surgical procedures. Surch supports video search based on procedural graphs generated by our clustering workflow capturing latent patterns within surgical procedures. We used vectorization and weighting schemes that characterize the features of procedures, such as recursive structures and unique paths. Surch enhances cross-video comparison by providing video navigation synchronized by surgical steps. Evaluation of the workflow demonstrates the effectiveness and interpretability (Silhouette score = 0.82) of our clustering for surgical learning. A user study with 11 residents shows that our system significantly improves the learning experience and task efficiency of video search and comparison, especially benefiting junior residents.

Zhiyu Chen, Wenhu Chen, Charese Smiley, Sameena Shah, Iana Borova, Dylan Langdon, Reema Moussa, Matt Beane, Ting-Hao Huang, Bryan Routledge, William Yang Wang. Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing. 2021.