Prediction of surgery start for automated anesthesia using draping detection from surveillance videos.

One of the primary goals of automated anesthesia is to reduce human intervention and reduce the workload of anesthesiologists. However, switching modes before the start of surgery still requires manual operation. The present study aims to develop a system that predicts the start of surgery by analyzing the actions of medical staff in the operating room using surveillance camera footage, thereby enabling automated mode transitions in anesthesia systems. We analyzed 110 surveillance videos of elective laparoscopic surgeries at Kyoto University Hospital. Key medical staff actions to predict the start of surgery were identified, and the time intervals between each action and skin incision were recorded. We then developed a detection system to identify draping, the best key action, and evaluated it by comparing system-detected draping times with manually annotated times in 96 videos. Five key actions were identified: hand washing, sterilization, light activation, bed cradle set-up, and draping. The start of draping had the shortest median time interval to the skin incision (7.71 min, interquartile range: 5.89-9.72), which was significantly shorter than that of the other actions (p < 0.05), and also had the shortest interquartile range. In the system evaluation, the median time error for detecting draping was 19.0 s (interquartile range: 16.0-50.0). The start of draping is a reliable predictor of the start of surgery, and the draping detection system demonstrated high accuracy. These results support advances in anticipatory automated anesthesia systems, enhancing workflow efficiency and patient safety in the operating room.