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

Figures

• 

  FIGURE 1. Conventional method for predicting the cryoballoon ablation difficulty. (A) a longer left superior pulmonary vein (LSPV) ostium‐bifurcation distance (>26.1 mm) (B) a shorter right inferior pulmonary vein (RIPV) ostium‐bifurcation distance (<10.4 mm) (C) a thinner left lateral ridge (<4.7 mm) (D) a higher ovality index of the LSPV (>50.5%).

• 

  FIGURE 2. The overview of the deep learning architecture. Computed tomography (CT) images were extracted into a simplified data format that represents only the three‐dimensional (3D) geometry, and the data size was reduced through edge collapse. These data were input into the 3D deep learning model, which output binary data indicating either cryoballoon ablation (CBA) difficulty or CBA nondifficulty. CBA difficulty was defined as those that required touch‐up ablation and/or more than 3 times per pulmonary vein.

• 

  FIGURE 3. (A) The number of cryoballoon ablation (CBA) difficulty and nondifficulties in each pulmonary vein (B) The number needed to disconnect (NND) and the number of touch‐up ablations in each pulmonary vein. LIPV, left inferior pulmonary vein; LSPV, left superior pulmonary vein; RIPV, right inferior pulmonary vein; RSPV, right superior pulmonary vein.

• 

  FIGURE 4. The area under the receiver operating characteristic curve of the deep learning model. The area under the receiver operating characteristic curve (AUROC) value was 0.845.

• 

  FIGURE 5. Prediction accuracy of cryoballoon ablation difficulty using a three‐dimensional deep learning model. LIPV, left inferior pulmonary vein; LSPV, left superior pulmonary vein; RIPV, right inferior pulmonary vein; RSPV, right superior pulmonary vein.