Comparison of model-based iterative vs non-iterative 3D device reconstruction for continuous-sweep limited-angle fluoroscopy

Abstract

Continuous-sweep limited angle (CLA) fluoroscopy is a recently proposed method for live 3D catheter reconstruction using a single-plane C-arm that continuously rotates back-and-forth within a narrow angle during fluoroscopic image acquisition. This study compares the accuracy and computation time of two real-time 3D catheter reconstruction algorithms. The first approach, iterative model-based device reconstruction (MDR), optimizes the 3D location of 8 control points representing the catheter as a cubic spline. Optimization was performed by minimizing a cost function that describes the data consistency between reconstructed shape and 2D catheter segmentations in the current and previous fluoroscopic images. Alternatively, the second approach is a topology observing reconstruction (TOR) algorithm, which reduces the solution space by performing a graph-based analysis of the 3D vessel tree followed by an analytical minimization of a simplified cost function. Both approaches were evaluated in an experimental study where a microcatheter was navigated through a 3D printed vessel phantom. Accuracy was determined by comparing the reconstructed final catheter pose to a reference 3D cone beam CT scan. The root mean squared distance (RMSD) between reference and CLA reconstruction was considerably smaller for the TOR approach (1.88 ± 0.34mm) than for MDR (6.78 ± 4.4mm). Computation time was also shorter for TOR (0.34ms) compared to MDR (518.6ms) making it suitable for real-time reconstruction with clinically relevant frame rates. The clinical translation of CLA could improve procedure efficiency by providing an intuitive 3D visualization of the device within a vessel. Further work is needed to evaluate the approach in vivo in the presence of respiratory motion.