3D pose estimation of epidural catheter markers based on a single-view 2D X-ray fluoroscopy

초록

Minimal invasive surgery requires precise control of flexible instruments to ensure safe and accurate procedures within complex anatomical structures. Accurate estimation of the three-dimensional (3D) pose of flexible surgical tools such as catheters is therefore crucial. While various sensing-based approaches, including fiber Bragg grating sensors, electromagnetic tracking, and ultrasound localization, have been explored, these methods depend on additional sensing hardware and are sensitive to calibration and environmental conditions. X-ray imaging offers distinct advantages by allowing visualization without specialized catheters or external devices, enabling universal applicability; however, existing image-based tracking methods either rely on large training datasets are limited to thick or structured catheters, or lack real-time performance. In this study, we present a slender-marker-based method to accurately estimate the 3D pose of a catheter from a single X-ray image using thin radiopaque marker sets compatible with commercially available epidural catheters. The proposed algorithm resolves the inherent 3D ambiguity in single-plane X-ray projections by combining a mixture-of-experts model with geometry-constrained refinement, and achieves real-time and precise tracking without any pre-obtained experimental data through simulation-based learning, enabling strong generalization across catheter types. Validation with X-ray and C-arm experiments confirmed clinically acceptable accuracy, below 2 mm, showing that the proposed approach preserves the advantages of X-ray-based interventions while reducing radiation exposure for both surgeons and patients.

키워드