Improving ultrasound-based bone registration using the iterative closest point algorithm paired with a complex...

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Computational Anatomy

Background:

• Iterative Closest Point (ICP) is a standard algorithm for bone registration using ultrasound.
• ICP is susceptible to local minima, potentially compromising registration accuracy.
• Gradient-based optimization methods commonly used in ICP can be less robust than alternative approaches.

Purpose of the Study:

• To investigate the impact of initial estimate and registration point quantity on bone registration accuracy.
• To compare the performance of ICP with a Complex optimization routine against Quadratic Sequential Programming (SQP).
• To evaluate the robustness of Complex optimization for ultrasound-based bone registration.

Main Methods:

• Ultrasound (A-mode) measurements were taken on bovine and ovine bone specimens in ballistic gel.
• Bone and ultrasound probe positions were tracked in 3D space using retroreflective markers.
• Bone registrations were performed using two ICP solvers and a Complex optimization routine, varying initial estimates and the number of registration points.

Main Results:

• Complex optimization achieved <1 mm translation and <1° rotational error in 68% of registrations with perturbed initial estimates, compared to 35% for SQP.
• Both Complex optimization and SQP showed similar median registration errors when varying the number of registration points.
• Complex optimization consistently provided accurate bone registrations across different conditions.

Conclusions:

• Complex optimization demonstrates superior robustness and accuracy for ultrasound-based bone registration compared to SQP.
• While accurate, the objective function in Complex optimization does not always identify the most precise registrations.
• Further research is needed to refine Complex optimization objective functions for improved registration error determination.