Introduction
Video-based coaching and debriefing of laparoscopic surgical procedures has been demonstrated to contribute to enhanced surgical performance; however, modern pressures on training and productivity preclude spending hours viewing and editing surgical video for routine video-based coaching. As an alternative, artificial intelligence (AI) can be used to automatically segment, summarize, and classify a surgical video into its constituent steps.
Method
Laparoscopic sleeve gastrectomy (LSG) videos were collected from a single institution from 2014-2016. Three surgeons provided manual segmentation and annotation of the videos into seven representative steps such as port placement, gastrocolic ligament dissection, and stapling. Machine learning algorithms (e.g. support vector machine classifiers, hidden Markov models, neural networks) and coresets were utilized to analyze video in real-time.
Results
Thirty LSG videos were collected with some videos used as training data and the others as novel test data. Compared to a “ground-truth” surgeon segmentation/identification of operative steps, AI had 92% accuracy in segmentation and identification of steps. Data compression up to 90% was achieved using coresets to enable efficient segmentation of large amounts of video data.
Conclusion
AI can be effectively trained to deliver a reliable and efficient segmentation/labeling of operative video into its constituent steps. Automated segmentation can allow important operative steps to be efficiently indexed for training or expert consultations for feedback and coaching. Identification of deviations from an expected operative course lays the foundation for automated detection of intraoperative adverse events. This work demonstrates that AI can enable real-time, evidence-based clinical decision support in an operative setting.