A robust automated surface-matching registration method for neuronavigation.

Area of Science:

• Neurosurgery
• Medical Imaging
• Computer-Aided Surgery

Background:

• Current neuronavigation systems rely on manual landmark selection for coarse registration, leading to increased procedure times and potential inaccuracies.
• Automatic and efficient patient-to-image registration is crucial for practical and precise surgical navigation.

Purpose of the Study:

• To develop and validate a fast, accurate, and automatic coarse-to-fine spatial registration method for patient-to-image alignment in neuronavigation.
• To eliminate the need for manual landmark identification and reduce registration time without compromising accuracy.

Main Methods:

• Proposed a coarse-to-fine registration approach utilizing 3D Harris corner detection for keypoint extraction from point clouds.
• Employed the 4-points congruent sets (4PCS) algorithm for automatic coarse registration of keypoints, followed by an iterative closest point (ICP) variant for fine registration.
• Validated the method using phantom and patient head data, comparing its efficiency and effectiveness against existing techniques.

Main Results:

• Keypoint extraction was achieved within 7.0 seconds.
• Phantom experiments yielded a mean target registration error (TRE) of 1.17 ± 0.04 mm with an average registration time of 17.4 seconds.
• Clinical experiments on five patients showed mean TREs ranging from 1.64 ± 0.3 mm to 1.83 ± 0.38 mm, with an average registration time of 21.4 seconds.
• The proposed method demonstrated a significant speed advantage over traditional and 4PCS/ICP-only methods while maintaining comparable or superior accuracy.

Conclusions:

• The developed method substantially enhances registration speed in neuronavigation.
• It guarantees equivalent or superior registration accuracy compared to existing methods.
• The automatic nature of the registration process eliminates the need for tedious manual interventions.