Robustness and accuracy of feature-based single image 2-D-3-D registration without correspondences for image-guided...

Area of Science:

• Medical Imaging
• Computer-Aided Surgery
• Image Registration

Background:

• 2-D-to-3-D registration is crucial for image-guided interventions.
• Establishing accurate point correspondences between 2-D images and 3-D data is challenging due to intraoperative clutter and imaging physics.
• False detections in feature extraction can hinder precise registration.

Purpose of the Study:

• To develop an accurate, robust, and fast method for 2-D-to-3-D registration using a single image.
• To address the limitations of paired correspondences in the presence of false detections.
• To improve the reliability of image-guided interventions.

Main Methods:

• Formulated 2-D-to-3-D registration as a maximum likelihood estimation problem.
• Employed a coupled approach of expectation maximization and particle swarm optimization.
• Evaluated the method using phantom and cadaver studies.

Main Results:

• Achieved subdegree rotational and submillimeter in-plane translational errors in phantom studies.
• Outperformed state-of-the-art methods lacking paired correspondences.
• Demonstrated comparable accuracy to global optimal methods that rely on correct paired correspondences.

Conclusions:

• The proposed method enables accurate 2-D-to-3-D registration from a single image, even with false detections.
• It offers a robust alternative to methods requiring precise paired correspondences.
• This advancement has significant implications for improving image-guided interventions.