Contrast-Based 3D/2D Registration of the Left Atrium: Fast versus Consistent

Area of Science:

• Medical Imaging
• Computational Anatomy
• Interventional Cardiology

Background:

• Augmented fluoroscopy in cardiac ablation relies on registering pre-acquired 3D left atrium (LA) models to intra-procedural X-ray images.
• Accurate registration is crucial for navigation but challenging with limited contrast agent (CA) visualization.

Purpose of the Study:

• To develop and evaluate an automatic registration method for 3D LA models to fluoroscopic images, effective even with minimal CA.
• To improve the accuracy and robustness of image registration for augmented fluoroscopy during cardiac ablation procedures.

Main Methods:

• Proposed two novel similarity measures: one based on anatomical edges and another using a contrast agent distribution estimate (CADE) within the 3D model.
• Applied temporal filtering using a Markov chain framework to registration results from image sequences.
• Evaluated the method on clinical angiographic sequences with varying levels of CA.

Main Results:

• Achieved a mean error of 7.9 ± 6.3 mm for all frames in well-contrasted sequences, reduced to 4.6 ± 4.0 mm when registering to an optimal frame.
• Temporal filtering decreased the overall error from 7.9 ± 6.3 mm to 5.7 ± 4.6 mm.
• The combined similarity measures outperformed previous methods, with accuracy comparable to manual registration for well-contrasted sequences.

Conclusions:

• The proposed automatic registration methods, particularly the combination of edge-based and CADE similarity measures, offer improved accuracy for augmented fluoroscopy in cardiac ablation.
• The techniques show promise for enhancing navigation and safety in procedures, especially when limited contrast agent is available.