MICCAI 2021 Daily - Tuesday

In this paper, Kun proposes a model for anticipating surgical instrument and phase occurrence before they present in surgery. This would support instrument preparation and operating room automation, as well as surgical coaching. Originally, the team planned to investigate the workflow analysis problem to recognize the current phase of the surgery taking place. However, thanks to deep learning , other works are already showing promising results here, so they decided to turn to something more challenging: the anticipation problem. The anticipation problem has been rarely explored by current researchers. There are some methods in the conventional computer vision field, but they are not yet fully applicable to a surgical scenario. Having carefully considered different options, Kun tells us they opted for remaining time prediction as their final solution. “ The whole network is composed of a spatial feature extractor and a temporal model, ” he explains. “ The spatial feature extractor benefits from our newly proposed instrument interaction module. This module has been designed to capture the surgical instrument and its surrounding interaction. The temporal model is a multi- stage temporal convolutional neural network. It is lightweight and has a 10 DAILY MICCAI Tuesday Oral Presentation Surgical Workflow Anticipation using Instrument Interaction Kun Yuan is a master’s student at the University of Ottawa in Canada under the supervision of Professor Won-Sook Lee and Professor Matthew Holden. His work explores building a more generalized computer- assisted surgery system for anticipation in the surgical workflow. He speaks to us ahead of his oral presentation and poster session today.