3D/2D Vessel Registration Based on Monte Carlo Tree Search and Manifold Regularization

Area of Science:

• Medical Imaging
• Computer-Aided Surgery
• Biomedical Engineering

Background:

• Digital subtraction angiography (DSA) lacks depth information and requires excessive contrast agents.
• Augmented intra-operative imaging aims to improve vascular interventional surgery navigation.
• Accurate 3D/2D vessel registration is crucial for effective image augmentation.

Purpose of the Study:

• To develop and evaluate a robust 3D/2D vessel registration method for augmented intra-operative imaging.
• To improve the accuracy and efficiency of image guidance in vascular interventions.
• To address limitations of current DSA-based navigation techniques.

Main Methods:

• A 3D/2D registration method based on vessel graph matching is proposed.
• Rigid registration utilizes Monte Carlo tree search for optimal vessel graph matching.
• Nonrigid registration employs a novel manifold regularization model with simplified gradient formulas for fast registration.

Main Results:

• The proposed method demonstrated superior performance against seven rigid and three nonrigid techniques.
• Achieved average registration errors of 2.14 mm (rigid) and 0.34 mm (nonrigid).
• Exhibited high computational efficiency with an average computation time of 0.51 seconds.

Conclusions:

• The novel 3D/2D vessel registration method enhances intra-operative imaging accuracy and efficiency in vascular surgery.
• The technique shows robustness against pose variations, noise, and deformations.
• This approach offers a significant advancement for DSA-based navigation in interventional procedures.