Robust Catheter and Guidewire Tracking Using B-Spline Tube Model and Pixel-Wise Posteriors

Abstract

In endovascular surgery and cardiology, robotic catheters are emerging as a promising technology for enhanced catheter manipulation and navigation while reducing radiation exposure. For robotic catheter systems especially with tendon actuation, a key challenge is the localisation of the catheter shape and position within the anatomy. An effective approach is through image-based catheter/guidewire detection and tracking. However, these are difficult problems due to the thin appearance of the instruments in the image and the low signal-to-noise ratio of fluoroscopy. In this letter, we propose a deformable B-spline tube model, which can effectively represent the shape of a catheter and guidewire. The model allows fitting using a region-based probabilistic algorithm, which does not rely on intensity gradients but exploits a signed distance function and the nonparametric distributions of measurements. Unlike previous B-spline fitting approaches, which optimise the spline with respect to control points, we propose a knot-driven scheme with an equidistance prior to better fit complex curves. Our probabilistic framework shows promising results for catheter and guidewire tracking in different procedures even with handling overlapping instrument segments. We present empirical studies using phantom model data and in vivo fluoroscopic sequences with annotated ground truth. Our results indicate that the proposed approach can precisely model the catheter and guidewire contours in near real time, and this information can be embedded in a robotic catheter control loop or utilised for image-guidance.