Disocclusion-based 2D-3D registration for aortic interventions.

Area of Science:

• Medical imaging
• Computer-assisted interventions
• Image processing

Background:

• Intensity-based medical image registration algorithms struggle with accuracy and robustness due to occlusions from medical instruments.
• Occlusion and dissimilarity present significant challenges in 2D-3D registration tasks.

Purpose of the Study:

• To present a novel disocclusion-based 2D-3D registration method that effectively handles occlusions and dissimilarities.
• To introduce and evaluate new disocclusion techniques and robust similarity measures for improved registration performance.

Main Methods:

• Developed two disocclusion techniques: Spline Interpolation and Stent-editing.
• Introduced two robust similarity measures: Huber and Tukey Gradient Correlation.
• Validated the proposed methods on both synthetic and real interventional medical image data.

Main Results:

• Integration of disocclusion techniques significantly enhances accuracy and robustness in 2D-3D registration.
• The chosen robust similarity measures (Huber and Tukey Gradient Correlation) demonstrate varying impacts based on the nature of the occluding structure.
• Comparative analysis against established methods confirms the superiority of the proposed approach.

Conclusions:

• Disocclusion-based strategies are crucial for overcoming occlusion challenges in medical image registration.
• The developed disocclusion techniques and robust similarity measures offer a more accurate and reliable solution for 2D-3D registration.
• Further investigation into robust measures tailored to specific occluding structures can optimize registration outcomes.