Abstract

Info

Title:: Guidewire registration method based on PointNet and curvature constraints

Author(s):: DENG Zihan; HU Zhi; XIN Shaozong; LI Shufan; ZHANG Beilang; School of Electronic and Electrical Engineering, Shanghai University of Engineering Science, Shanghai 201620, China

Keywords:: Keywords: vascular intervention PointNet regression prediction iterative closest point curvature

Abstract:: Abstract: To address the inherent limitations of traditional registration algorithms in vascular interventional procedures, including high dependency on the initial position and orientation of flexible guidewires, insufficient registration accuracy, high susceptibility to local optima, and low computational efficiency, a novel guidewire registration method integrating PointNet and curvature constraints is proposed. Specifically, a physical model of the guidewire and blood vessels is established to capture motion points. Then, key geometric features are extracted using PointNet, and a pose regression network is used to predict the transformation matrix between the guidewire point cloud and the vascular centerline point cloud. The registration is further optimized by accelerating point cloud search via KD-Tree and refining results with a curvature feature-constrained iterative closest point algorithm. Experimental results reveal that compared with traditional methods and learning-based approaches, the proposed method achieves the smallest mean square error and mean absolute error, demonstrating its effectiveness in vascular interventional guidewire registration.