Prediction of difficulty in cryoballoon ablation with a three-dimensional deep learning model using polygonal mesh...

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Cryoballoon ablation (CBA) is a key technique for pulmonary vein (PV) isolation in treating atrial fibrillation.
• Challenging CBA cases often require multiple applications or touch-up ablations, impacting procedural efficiency.
• Existing predictors of CBA difficulty do not account for the spatial anatomy of the left atrium and PVs.

Purpose of the Study:

• To develop and evaluate a novel three-dimensional (3D) deep learning (DL) model for predicting CBA difficulty.
• To compare the predictive accuracy of the DL model against conventional manual measurement methods.

Main Methods:

• A 28-mm cryoballoon was used in 189 patients with drug-resistant atrial fibrillation.
• CBA difficulty was defined by the need for touch-up ablation or >3 applications per PV.
• A DL model analyzing polygonal meshes was developed and compared to traditional predictors (e.g., left lateral ridge thickness, PV ostium-bifurcation distances).

Main Results:

• The DL model demonstrated superior accuracy (0.793 vs. 0.630) and specificity (0.796 vs. 0.609) compared to conventional methods (p < 0.05).
• The area under the receiver operating characteristic curve (AUC-ROC) for the DL model was 0.821.
• Conventional predictors included left lateral ridge thickness and PV ostium-bifurcation distances.

Conclusions:

• A 3D DL model utilizing polygonal mesh representation effectively predicts CBA difficulty.
• Proactive identification of challenging cases allows for optimized procedural strategies.
• This AI-driven approach has the potential to enhance the success rates of pulmonary vein isolation via CBA.