Comparison of the characteristics between machine learning and deep learning algorithms for ablation site...

Area of Science:

• Cardiology
• Medical Informatics
• Artificial Intelligence in Medicine

Background:

• CARTONET is a cloud-based system for analyzing CARTO system ablation procedures.
• The current CARTONET R14 model utilizes deep learning, but its performance metrics require thorough evaluation.

Purpose of the Study:

• To compare the performance characteristics of the CARTONET system between the R12.1 and R14 models.
• To assess the accuracy and predictive value of the CARTONET R14 deep learning model.

Main Methods:

• Analysis of data from 396 atrial fibrillation ablation cases.
• Investigation of sensitivity and positive predictive value (PPV) of the CARTONET R14 automated anatomic location model.
• Comparison of data between CARTONET R12.1 and CARTONET R14 models.

Main Results:

• The CARTONET R14 model demonstrated significantly improved sensitivity (77.5%) and PPV (86.2%) compared to the R12.1 model (71.2% and 85.6%, respectively).
• Analysis of 39,169 points and 625 segments revealed high incidence of reconnections in the posterior areas of pulmonary veins.
• The highest possibility of reconnection was observed in the roof area of both right and left pulmonary veins.

Conclusions:

• The CARTONET R14 model offers significant improvements in sensitivity and PPV over the R12.1 model.
• The R14 model maintains a similar tendency for predicting potential reconnection sites as the R12 model.