Validation of Automated Low Voltage Area Measurement Using CARTONET in Atrial Fibrillation Patients.

BACKGROUND

Low-voltage areas (LVAs) detected via electroanatomic mapping (EAM) are key indicators of atrial remodeling and predictors of atrial fibrillation (AF) recurrence following pulmonary vein isolation (PVI). However, manual LVA quantification is time-intensive, operator-dependent, and prone to variability.

OBJECTIVE

To validate the accuracy, consistency, and efficiency of automatic LVA quantification function of CARTONET-a cloud-based automated platform for LVA quantification-by comparing it with conventional manual analysis in AF patients.

METHODS

This retrospective study included 100 patients who underwent PVI for AF at a single center between January 2022 and December 2023. Manual LVA measurements were performed by experts using CARTO3. LVA quantification and anatomical segmentation were performed using CARTONET. LVA was stratified into four categories: low (<5%), mild (5%-20%), moderate (20%-35%), and severe (>35%). Statistical analyses included Spearman's rank correlation (ρ), Bland-Altman agreement, and confusion matrices for categorical classification, with subgroup analysis based on segmentation errors and map point density. The number and the locations of segmentation errors were quantified.

RESULTS

CARTONET demonstrated strong correlation with manual measurements across all LA regions (ρ = 0.928-0.983, p < 0.01). Significant differences were observed in anterior and posterior walls (p < 0.01), especially in low-density maps. In high-density maps, no significant differences were observed (p ≥ 0.07). Bland-Altman analysis showed minimal bias (mean difference: -0.8%), and 96% of values fell within limits of agreement. The confusion matrix showed that categorical agreement for LVA stratification exceeded 90%. Segmentation errors were most common at the LSPV but did not significantly impact LVA quantification.

CONCLUSION

CARTONET offers reliable, efficient LVA quantification with high concordance to manual methods. Its performance in high-quality maps supports its use in clinical and research settings, reducing operator workload and promoting standardized substrate evaluation.