Sparse Catheter Pathways for Neural Network-Based Left Atrial Reconstruction

Overview

This study introduces a neural network-based method to reconstruct the left atrial (LA) anatomy from sparse catheter pathway data, aiming to reduce mapping time while maintaining anatomical accuracy. The dense encoder–decoder (DED) network outperforms baseline methods in reconstructing LA shapes, particularly pulmonary vein ostia, using limited catheter traversal paths acquired in under 3 minutes.

Background

Atrial fibrillation (AF) is the most common cardiac arrhythmia and is primarily treated with catheter-based electro-anatomic mapping (EAM) and radiofrequency ablation targeting pulmonary vein isolation (PVI). Current EAM systems require extensive catheter contact along the LA blood pool and tissue boundaries, which is time-consuming and operator-dependent. Imaging modalities like MRI, CT, and ultrasound have limitations such as radiation exposure, noise, and restricted field of view. Efficient and accurate LA surface reconstruction from sparse catheter data could improve procedural efficiency and outcomes.

Data Highlights

Key Findings

• The proposed DED neural network reconstructs anatomically relevant LA surfaces from sparse catheter paths acquired in under 3 minutes.
• Reconstruction focuses on critical anatomical landmarks such as the four pulmonary vein ostia, improving identification and ablation targeting.
• The method accommodates common LA anatomical variations with four pulmonary veins and can be adapted for less common variations via model retraining.
• Compared to baseline methods, the DED network provides superior accuracy and time efficiency in real-world clinical data.
• Synthetic data generation using statistical shape models enables robust training despite limited patient data availability.
• Early LA surface visualization during mapping may reduce procedure time and improve safety by guiding catheter navigation in challenging regions.

Clinical Implications

This neural network-based approach can streamline the LA mapping process by reducing the need for extensive catheter maneuvering and manual editing, potentially shortening procedure times. Accurate early visualization of pulmonary vein anatomy may enhance ablation precision and safety, particularly in complex anatomical regions. Integration with existing EAM systems could improve clinical workflow and patient outcomes in AF treatment.

Conclusion

The study demonstrates that sparse catheter pathway data combined with a neural network reconstruction approach can efficiently and accurately model the left atrial anatomy. This advancement holds promise for optimizing AF ablation procedures by enabling faster and anatomically precise mapping.

References

1. Sciarra et al. 2019 -- Evaluation of CARTO® 3 Fast Anatomical Mapping
2. Biosense Webster 2021 -- Model FAM (mFam) for LA Reconstruction