P.140 Automated video tracking and behavior recognition in rodents: deep learning improves tracking robustness and classification accuracy

Abstract

The ability to hold an endoscope is an important skill for surgeons and assistants performing endoscopic neurosurgery. Motion tracking can provide an objective assessment for surgical skills and may aid in endoscopic neurosurgery. We developed a cost-effective system to study the feasibility of objectively distinguishing performance levels for operating an endoscope.The study was divided into 2 parts. First, a video motion tracking analysis tool was built based on a printed mark and free software (Kinovea 0.8.15). Second, the tool was used to distinguish the holding endoscope performance of the robotic arm by experts (surgeon, n = 3) and novice users (residents, n = 6) under both 0′ and 30′ endoscopes.Through the printed mark and free software, we successfully built a system for video motion tracking. The data analysis showed that for both 0′ and 30′ endoscopes, the experts had a shorter total distance and depth, smaller average speed and maximum acceleration, and a fewer number of extreme accelerations than their novice counterparts. Compared with experts and residents, the fixed arm had better results.The simple low-cost video motion tracking system can provide an objective assessment of an endoscope holding skill, which allows for discrimination between an expert and a novice, and can be used by clinical neurosurgeons to select a qualified assistant.