Novel RANSAC-based method for detecting and estimating externally attached marker spheres in craniomaxillofacial CT...

Area of Science:

• Medical Imaging
• Computer-Aided Surgery
• Surgical Navigation

Background:

• Accurate estimation of spherical parameters from CT images is crucial for image-physical space registration in craniomaxillofacial surgical navigation.
• Existing methods are susceptible to errors from artifacts and interference, impacting precision.
• Robust methods are needed to meet high-precision surgical application requirements.

Purpose of the Study:

• To develop a robust method for accurate spherical parameter estimation from CT images.
• To improve the reproducibility and accuracy of surgical navigation systems, especially with noisy data.
• To enhance the reliability of craniomaxillofacial surgical navigation.

Main Methods:

• Proposed a Local Evaluation and Optimization RANdom SAmple Consensus (LEO-RANSAC) algorithm for refining spherical parameter detection.
• Introduced a novel metric combining multi-level adaptive curvature and local solutions to filter models.
• Utilized custom equipment for fiducial localization error (FLE) measurement and skull phantom studies for fiducial registration error (FRE) and target registration error (TRE) evaluation.

Main Results:

• Evaluated on 72-point clouds with inlier ratios from 30% to 90%.
• 87.50% of maximum FLEs were less than 0.9 mm, and 95.83% of FLE variances were less than 0.01.
• Skull phantom studies yielded FREs of 0.4222, 0.5223, 0.372 mm and TREs of 0.8546, 0.9471, 0.8537 mm.

Conclusions:

• The proposed LEO-RANSAC method demonstrates superior accuracy and reliability compared to existing approaches.
• The method effectively handles low inlier ratio data, improving robustness in surgical navigation.
• Results highlight the method's potential for high-precision craniomaxillofacial surgical navigation applications.