Towards real-time critical view of safety detection

Yariv Colbeci, Maya Zohar, Daniel Neimark, Dotan Asselmann, Omri Bar

Keywords: Surgical Intelligence, Critical View of Safety, CVS, Multi Instance Learning

Abstract: Surgical procedures have a clear designated goal, which makes the art of performing surgery a task-oriented action. The performing surgeon follows specific workflow steps that describe the actions needed to reach the surgery goal. In ectomy procedures, such as Cholecystectomy and Appendectomy, the goal is to dissect and remove a specific organ. Safety measures are set to prevent injuries, and the surgeon needs to follow protective methods to avoid misidentification. In Laparoscopic Cholecystectomy (LC), this measure is known as Critical View of Safety (CVS). This work illustrates that machine learning can detect CVS accurately enough to be used routinely in the clinical setting, both for educational purposes and in real-time scenarios. We formulate CVS detection as a supervised Multi Instance Learning (MIL) problem designated to work in the wild in a production environment. Our proposed approach uses an attention-based MIL model, trained and evaluated on more than 2,000 different surgical videos. It achieves 82.6% mean unweighted accuracy in detecting LC CVS criteria and 84.2% accuracy in the final task of CVS detection.

Registration: I acknowledge that publication of this at MIDL and in the proceedings requires at least one of the authors to register and present the work during the conference.

Authorship: I confirm that I am the author of this work and that it has not been submitted to another publication before.

Paper Type: both

Primary Subject Area: Application: Endoscopy

Secondary Subject Area: Transfer Learning and Domain Adaptation

Confidentiality And Author Instructions: I read the call for papers and author instructions. I acknowledge that exceeding the page limit and/or altering the latex template can result in desk rejection.