Evaluating similarity measure for multimodal 3D to 2D registration

Area of Science:

• Medical Imaging
• Spine Surgery
• Computer-Aided Surgery

Background:

• 3D to 2D registration is essential in spine surgery to determine vertebral pose and prevent wrong-site surgery.
• Accurate spatial correspondence between pre-operative MRI and intra-operative X-ray images is critical.
• Various similarity measures are employed to evaluate this correspondence.

Purpose of the Study:

• To evaluate and compare the performance of three distinct similarity measures for 3D to 2D image registration in spine surgery.
• To assess registration accuracy using metrics like Target Registration Error, iteration times, and success rate.
• To identify the most effective similarity measure for feature-based registration.

Main Methods:

• Implementation of a framework using Binary Image Matching, Dice Coefficients, and Normalized Cross-correlation for 3D to 2D image registration.
• Comparison of similarity measures based on pixel position analysis.
• Evaluation of registration accuracy on simulated AP and Lateral test images.

Main Results:

• All three evaluated similarity measures performed effectively in feature-based 3D to 2D registration.
• Binary Image Matching (BIM) yielded superior results compared to Dice Coefficients and Normalized Cross-correlation.
• High registration accuracy was maintained with initial displacements up to ±20 mm and ±10°.

Conclusions:

• Binary Image Matching is a highly effective similarity measure for 3D to 2D registration in spine surgery.
• The proposed framework demonstrates robust performance across various similarity measures.
• Accurate vertebral pose estimation is achievable even with significant initial positional variations.