Surgical Suture Thread Detection and 3-D Reconstruction Using a Model-Free Approach in a Calibrated Stereo Visual System

In robot-assisted surgery (RAS), three-dimensional (3-D) position estimation of a surgical suture thread is essential in automating the wound suturing procedure. Nevertheless, accurate suture thread's segmentation and its 3-D coordinates reconstruction remain challenging because of its flexible property and lacking indistinctive feature points. In this paper, a novel model-free method is proposed to enable the 3-D computation of a suture thread through a calibrated stereo camera system. In this method, only the suture thread's tip and grasping points need to be indicated on one image with mouse clicks, and their accurate image positions can be detected using computed contour information. Afterward, key points that represent the suture thread are generated on one image, while their matched points on the second camera frame are evaluated through an iterative algorithm for 3-D model reconstruction. Experiments using different backgrounds and environmental conditions were conducted to examine the accuracy and robustness of the 2-D segmentation. The suture threads were then formed in various 3-D shapes, and their respective 3-D models were reconstructed for comparison. The experimental results confirm the feasibility of our method in reducing human intervention in the detection of matched key points on an image pair, thereby guaranteeing an accurate 3-D coordinates computation of a suture thread.