1. Prediction of difficulty in cryoballoon ablation with a 3D deep learning model using polygonal mesh representation
- Author
-
K Nakasone, M Nishimori, K Kiuchi, M Shinohara, K Fukuzawa, T Takami, T Nakamura, Y Sonoda, H Takahara, K Yamamoto, Y Suzuki, K Tani, H Iwai, Y Nakanishi, and K Hirata
- Subjects
Cardiology and Cardiovascular Medicine - Abstract
Background Cryoballoon ablation (CBA) is a useful treatment for pulmonary vein isolation (PVI). Some cases, however, are difficult to treat and may require multiple freezing procedures and/or touch-up ablation. Although several predictors of CBA difficulty have been reported, no report has been able to assess the spatial location and morphology of the left atrium (LA) and pulmonary veins (PVs). A polygonal mesh is a collection of vertices, edges, and faces that defines the shape of a polyhedral object, and is able to represent a spatial location with a small amount of information. We hypothesized that a deep learning model that learns mesh representation datasets could more accurately detect the CBA difficulty and that we could establish a novel evaluation method in CBA. Purpose The aim of this study was to create a model to predict CBA difficulty with a 3D deep learning model using polygonal mesh representation. Methods and results All the 140 patients who underwent CBA for drug-resistant atrial fibrillation between January 2015 and January 2022 were included. A 28-mm cryoballoon (Arctic Front Advance, Medtronic) was used in all cases. We defined CBA difficulty as requiring a touch-up ablation procedure to create complete PVI. We converted the volume data in DICOM format of the computed tomography images of PVs and LA to obj file format (shown in Figure 1), which supports the definition of the geometry for object surfaces using polygonal meshes. Next, we developed a deep learning model that could learn polygonal meshes and classify whether the CBA required touch-up ablation or not. Only a training dataset is used to train the deep learning model, and finally, a test dataset is used to evaluate the model metrics. The accuracy, area under the ROC curve, recall, precision, and f1-score of the deep learning model using the test dataset was 86.5%, 87.7%, 66.7%, 75.0%, 70.6%, respectively. Conclusions We developed a 3D deep learning model that can detect a difficulty in CBA using polygonal mesh representation. By predicting difficult cases in advance, we will be able to develop strategies to increase the success rate. Funding Acknowledgement Type of funding sources: None.
- Published
- 2022
- Full Text
- View/download PDF