Dynamic 2-D/3-D Rigid Registration Framework Using Point-To-Plane Correspondence Model.

Area of Science:

• Medical Imaging
• Image-Guided Interventions
• Computer-Aided Surgery

Background:

• Accurate alignment of 3-D (CT/MRI) and 2-D (X-ray) images is essential for image-guided interventions.
• Current methods face challenges in achieving precise registration, especially with dynamic motion.

Purpose of the Study:

• To present a dynamic rigid 2-D/3-D registration framework for enhanced spatial perception in image-guided procedures.
• To improve the accuracy and robustness of 3-D to 2-D image registration.

Main Methods:

• Developed a dynamic rigid 2-D/3-D registration framework utilizing a novel point-to-plane correspondence model.
• Constrained updates for planar and non-planar 3-D rigid transformations.
• Evaluated using phantom studies, dynamic motion compensation, and a clinical angiogram dataset.

Main Results:

• Achieved high accuracy in simulations (0.07 mm head, 0.05 mm thorax) using single-view X-ray images.
• Demonstrated significant accuracy improvement in dynamic motion compensation compared to baseline methods.
• On clinical data, achieved mean 3-D accuracy < 0.8 mm and 2-D accuracy < 0.3 mm, outperforming state-of-the-art single-view registration.

Conclusions:

• The proposed framework offers an intuitive and generic solution for both initial and dynamic 2-D/3-D registration.
• It enhances accuracy and robustness, making it suitable for real-time image-guided interventions.
• The method provides superior performance in single-view registration accuracy and robustness.