Robust Self-Supervised Learning of Deterministic Errors in Single-Plane (Monoplanar) and Dual-Plane (Biplanar) X-Ray...

Area of Science:

• Medical Imaging
• Surgical Technology
• Computer Vision

Background:

• Fluoroscopic imaging is crucial for guiding minimally invasive procedures like catheter insertions.
• Enhanced accuracy in fluoroscopy data is needed to improve surgical precision in endovascular procedures.
• Current methods may lack the precision required for complex interventions.

Purpose of the Study:

• To develop and validate a robust self-calibration algorithm for single-plane and dual-plane fluoroscopy systems.
• To improve the accuracy of three-dimensional (3D) target localization and two-dimensional (2D) image reprojection in fluoroscopic imaging.
• To assess the effectiveness of k-nearest-neighbour (kNN) regression in modeling and correcting systematic errors.

Main Methods:

• A robust self-calibration algorithm was developed using a 3D target field imaged by fluoroscopy.
• Simultaneous estimation of 3D target positions and fluoroscope pose via Maximum Likelihood Estimation (MLE) with Student-t distribution.
• Incorporation of smoothed k-nearest-neighbour (kNN) regression to model image reprojection errors within a bundle adjustment framework.
• Iterative refinement of MLE and kNN steps until convergence, with comparison of four error modeling schemes.

Main Results:

• The self-calibration algorithm significantly reduced 3D mapping error from 0.61-0.83 mm to 0.04 mm (up to 95.7% improvement).
• 2D reprojection error decreased from 1.17-1.31 pixels to 0.20-0.21 pixels (up to 83.8% improvement).
• Biplanar fluoroscopy showed a 47.2% improvement in 3D measurement accuracy, reducing error from 0.60 mm to 0.32 mm.
• Smoothed kNN regression effectively modeled systematic fluoroscopy errors, comparable to expert performance with small datasets.

Conclusions:

• The proposed self-calibration algorithm robustly enhances fluoroscopic imaging accuracy for surgical guidance.
• The method demonstrates significant improvements in both 3D localization and 2D reprojection accuracy.
• kNN regression offers an effective automated approach to systematic error correction in fluoroscopy, reducing reliance on manual expertise.