Surgical Visual Understanding Banner
Join

**This Challenge has ended and is NO LONGER accepting applications. Please be on a lookout here for next year's challenge website.**

SurgVU - Surgical Visual Understanding

News:

  • 10th Oct, 2024: Final challenge results announced!
  • 17th Sept, 2024: Submission closed!
  • 1st Sept, 2024: Final testing phase submissions open
  • 5th Aug, 2024: Preliminary testing phase submissions open
  • 6th May, 2024: Challenge is live and accepting applications

Overview:

Machine learning models that can detect and track surgical context from endoscopic video will enable transformational interventions. For example, the ability to automatically categorize surgical progress (i.e. Phase, Step, Task, or Action) and the instruments used, will allow for improved assessments of surgical performance, efficiency, and tool choreography, as well as new analyses of OR resource planning. Indeed, the theoretical and practical applications are broad and far-reaching. Obtaining the data needed to train these models, however, is resource intensive and time consuming. Clinical videos need to be annotated, frame by frame, segmenting the surgical categories and identifying the bounding boxes and/or key points around surgical instruments, under a broad variety of conditions. Moreover, ongoing annotator training is needed to stay up to date with surgical methods and instrument innovation. Importantly, in robot-assisted surgery, instrument installation/uninstallation information can be programmatically harvested from the system, providing proxy annotations for tool presence in the video feed. Additionally, standard surgical ontologies are being developed with objective definitions, easing the burden on human annotators. In this challenge we invite the surgical data science community to take part in two categories. The first category requires participants to train a model to localize tools and their corresponding key-points in videos, using tool presence data as weak labels. The second category invites participants to segment videos into different surgical steps being performed. Winning solutions to either category would significantly reduce the annotation loads required for training models, and avail themselves to a wealth of clinical applications.

This challenge will take place as part of EndoVis challenge at MICCAI 2024 in Marrakesh, Morocco!
(Note: This year's challenge is an extension of the SurgToolLoc 2022/3 challenges (which was also held as part of EndoVis at MICCAI 2022/3). Please check out previous challenge's website for more details at https://surgtoolloc.grand-challenge.org/ and https://surgtoolloc23.grand-challenge.org/ for additional context)

Prizes:

Top performing teams will be awarded cash prizes that can total up to $8,000! See the prizes page for more details.

Interested in participating in the challenge? Check out the getting started page!

For any questions, please post it on the forum or use the contact us page to email direct queries.