Surgical Tool Detection and Pose Estimation using YOLOv8-pose Model: A Study on Clipper Tool

Analyzing or understanding medical images obtained from an endoscopic camera is important in computerassisted interventions (CAI). Compared to prior research, this paper introduces a method for detecting surgical tools and simultaneously estimating their pose. We collected and labeled data for the clipper tool to train based on the YOLOv8-pose model. The model was assessed on a test dataset using 4 different versions of the YOLOv8-pose model. Among these versions, YOLOv8n, with just 3M parameters (the lightest variant), demonstrates notably high accuracy in pose estimation. It achieves Precision, Recall, respectively. Furthermore, this version shows impressive accuracy in clipper detection, with Precision, Recall, mAP50, and mAP5095 scores of $\mathbf{9 7. 9 \%}, \mathbf{9 6. 0 \%}, \mathbf{9 9. 2 \%}$, and $\mathbf{6 4. 6 \%}$, respectively. These impressive results can pave the way for further development of this approach for various types of surgical tools in the future.