A machine learning algorithm identified spatiotemporal dispersion in 93% of atrial fibrillation patients, with higher dispersion sites in persistent AF (21±3 sites) vs. paroxysmal AF (6±2 sites).
AI-guided ablation of spatiotemporal dispersion in atrial fibrillation is feasible and correlates with left atrial functional impairment, though it prolongs procedure time.
Absolute Event Rate: 0% vs 0%
Abstract Introduction and Purpose: Artificial intelligence (AI) excels at detecting subtle patterns within large and complex datasets such as electroanatomical 3D maps (EAM). Recently, its application for identifying spatiotemporal dispersion (stD) has gained further support through the results of a first randomized controlled trial. However, its performance and clinical benefits remain to be fully established. Methods: This retrospective, single-center analysis included all-comer patients with atrial fibrillation (AF). Patient selection was based on the availability of stD mapping rather than patient characteristics. EAM for voltage, activation, and stD was performed, followed by pulmonary vein isolation and targeted ablation of identified dispersion regions. Demographic data and markers of atrial remodeling were systematically collected. Results: Our analysis included 27 patients (9 paroxysmal, 18 persistent), with a mean age of 65.3±1.9 years, a mean BMI of 28±1 kg/m², and an average AF duration of 7±1 years (c.f. Figure 1). stD was confined to low-voltage areas (LVA) in 56% of patients, while extending into normal-voltage regions in 44%. Baseline parameters were as follows: left ventricular ejection fraction (LVEF) 53±2%, global longitudinal LV strain −16±1%, left atrial (LA) reservoir strain 20±2%, left atrial volume index (LAVI) 41±3 ml/m², right atrial area 22±1 cm², mean left atrial pressure 8.6 ± 1 mmHg and mean procedure time 111±5 minutes. LVA represented 27±6% of the LA surface, while stD was annotated in 1.3±1.3%. The machine learning algorithm identified stD in 93% of patients. Patients with persistent AF exhibited a higher number of dispersion sites (21±3) compared to those with paroxysmal AF (6±2). Although LVA was more frequently observed in the anterior left atrial wall than in the posterior wall (81.5% vs. 66.6%), the total number of stD annotations was similar between the two regions (167 anterior, 158 posterior). StD burden correlated significantly with age (τ=0.388), LVA burden (τ=0.412), NT-proBNP levels (τ=0.494), and LAVI (τ=0.377) but not right atrial size (τ=-0.003), left atrial pressure (τ=0.113), or left ventricular strain (τ=0.099). When compared to LVA mapping, stD annotation demonstrated a stronger correlation with LA strain in all phases (mean τ=0.374 vs. τ=0.237) (c.f. Figure 2). At follow-up, recurrence was reported in 33% of paroxysmal and 60% of persistent AF patients. Despite these recurrences, most patients reported a reduction in symptom burden. Conclusion: The use of a machine learning algorithm to guide ablation in an all-comer patient population is safe and feasible, though it is associated with prolonged procedure times. Dispersion appears to be closely associated with low-voltage areas but is not confined to them. Compared to low-voltage annotation, dispersion burden may more accurately reflect left atrial functional impairment.
Heil et al. (Thu,) reported a other. A machine learning algorithm identified spatiotemporal dispersion in 93% of atrial fibrillation patients, with higher dispersion sites in persistent AF (21±3 sites) vs. paroxysmal AF (6±2 sites).