The Challenge in Surgical Video Annotation

To teach an AI model surgical intraoperative phases, a surgeon will watch a video and label each phase of the surgery.

Then an AI algorithm will be given both the labels and the video to learn what constitutes each phase.

The challenge is that surgical data is too complex and needs trained annotators.

Those expert annotators are rare.

One of the solutions for this problem is to pre-train models with unlabeled data.

Another solution is to let machine learning (ML) algorithms do “auto-annotation” from a few videos.