Automatic Centroid Angle Measurement From CT Image for Preoperative Rod Design of Robotic-Assisted Screw-Rod System Implantation

Abstract

Robotic-assisted implantation of screw-rod systems serves as an advanced therapy for spinal diseases. A precise curvature fit between rods and spines is critical to postoperative spinal stability. Currently, rod curvature is determined intraoperatively to accommodate screw positions, which is hardly conducive to optimal rod bending and is vulnerable to surgeons’ expertise. To address this challenge, we propose an automated and efficient method for measuring the centroid angle to guide preoperative rod design from CT images. The centroid angle is defined by lines connecting centroids of the upper and lower vertebrae pairs, providing a reliable measurement for spinal deformities. The proposed pipeline includes (1) 3D spine segmentation with multiscale multitask deep learning; (2) vertebrae recognition using graphical morphology; (3) automatic and reproducible centroid angle measurement. Our method is evaluated on both healthy and pathological spines. Compared to manual measurements by professional surgeons, our method achieves an accuracy of 94.50% and 91.93% on adjacent and non-adjacent vertebrae, respectively. A Slicer-based plugin for robotic-assisted screw-rod systems implantation is built, providing a new clinical tool to personalize screw-rod systems consistent with the natural spinal curvature, thereby enhancing biomechanical properties.