Abstract
Medical errors are critical issues in medical practice. Serious accidents are deeply investigated, but most slight cases are ignored as careless mistakes by clinical staff. Identification of potential risks in surgery and treatment of risks are important to decrease errors. Video recording and analyzing system was developed to record intraoperative information and to find risk factors, but visual mining by medical doctors requires much time and effort and the results will be subjective. Thus we adopted quantity of motion in the recorded video as a quantitative index to indicate “candidates” of incidents. This system was evaluated in a clinical case (brain tumor removal) to compare detecting ability of incidents between human observation and computer processing. While human observation took 8.7 hours (equivalent to operative time) and found 4 incidents, computer processing took only 2.7 hours and extracted 81 candidates under tentative extraction threshold. Two events were common to both methods, but results of computation contained many false positive cases and does not detect rest two cases which human observation succeeded. Computer detection reduced the time to find risk factors, but it contained false detection and could not detect motionless incidents. We will integrate other featuring quantity and machine learning methods.
