Computer Vision Techniques for Detection and Tracking of Surgical Instruments in the Operating Room

Abstract

In response to the ever-growing demand for real-time intraoperative intelligence in modern operating rooms, surgical instrument detection and tracking as a part of computer vision techniques for surgical instruments has made significant progress. These methods are applied as tool usage analysis and skill evaluation, autonomous assistance, and surgical workflow optimization. Deep learning architectures like CNNs, transformers, as well as HA, although they have recently emerged for SPTIC, still face challenges associated with occlusion, visual ambiguity, domain variability, or real-time performance. The state-of-the-art approaches, their experimental performance are reviewed, and gaps preventing clinical deployment are identified. The emphasis is made in terms of the evaluation metrics, the algorithmic advancements, and the system integration. The last section concludes with discussions of unresolved issues and possible directions to further develop the clinical utility of these technologies.