Artificial intelligence ECG localization facilitates focused activation mapping and improves outcomes in hemodynamically unstable ventricular tachycardia

Le résumé fourni par la source

Catheter ablation is an established therapy for ventricular tachycardia (VT) but remains limited in patients with hemodynamically unstable arrhythmias, who constitute up to 70% of patients.1 In such cases, activation mapping is often not feasible. Freedom from recurrence may be as low as 50%.2,3 Furthermore, acute hemodynamic decompensation occurs up to 11% of VT ablations, resulting in cardiogenic shock or death.4 Artificial intelligence (AI)-based electrocardiographic (ECG) mapping enables single-beat, non-invasive localization of VT exit sites.5 We evaluated whether integrating AI-ECG mapping into a workflow with brief VT inductions could enable targeted activation mapping and ablation, thereby improving outcomes in unstable VT without increasing procedural risk. Patients undergoing ablation of hemodynamically unstable VT at two centres (2020–24) guided by AI-ECG mapping were enrolled in chronologic order. Hemodynamically unstable VT was defined by hypotension, need for multiple vasopressors, urgent cardioversion/defibrillation, or cycle length < 333 ms.6 Control patients undergoing ablation for unstable VT were enrolled in reverse consecutive order starting immediately prior to availability of this algorithm (2017–20), matching in 1:1 ratio to the same institutions and operators. The forward solution ECG mapping algorithm is based on a library of arrhythmia simulations to accommodate patient-specific variations. Its accuracy was externally validated in a multicentre clinical trial.5 In the study group, the prospective workflow was induction of VT, AI-ECG mapping, overdrive pacing or cardioversion, strategic catheter placement, reinduction to allow focused activation mapping, and termination with burst pacing, cardioversion, or ablation. Then, a contiguous focused ablation lesion set was performed. Controls underwent mapping using substrate, pace-mapping, and activation/entrainment at operator discretion. The primary endpoint was freedom from VT recurrence requiring anti-tachycardia pacing (ATP) or defibrillation, or all-cause mortality at 1 year.2 Secondary endpoints included acute termination, feasibility of activation mapping, arrhythmia inducibility post-ablation, and procedural complications. Outcomes were analysed using Cox regression adjusted for PAINESD score, ejection fraction, age, and cardiomyopathy type. Sixty patients were included: 30 study patients and 30 controls (age: 66 ± 11 vs. 62 ± 13 years, P = 0.2; ischaemic cardiomyopathy: 57% vs. 30%, P = 0.067). The study group had higher baseline risk, with lower mean ejection fraction (36 ± 16% vs. 46 ± 15%, P = 0.01), higher PAINESD score (16.5 ± 8.0 vs. 10.5 ± 6.5, P = 0.002), and more frequent VT storm (70% vs. 40%, P = 0.02). Activation mapping was feasible in 93% of study patients vs. 27% of controls (P < 0.001). Entrainment mapping was performed in 23% vs. 10%, respectively. In study patients, the AI-predicted site was a median of 4.0 mm (IQR 1.0–9.8) from the critical VT site confirmed by invasive mapping or termination with ablation. Low voltage/fractionated electrograms were located near ECG-predicted sites in 38 of 49 VT morphologies (78%) in study patients and in 33 of 34 VT morphologies (97%) in study patients with ischaemic cardiomyopathy. A representative patient is shown in Figure 1. Panel (A) AI 12-lead ECG mapping localized VT to the basal inferoseptal LV in a 60-year-old male with h/o CABG. Panel (B) and (C): integration of this mapping result with the electroanatomic map (left). These images guided invasive activation mapping (right) during hemodynamically unstable VT, which delineated the critical isthmus located in a large scar (Panel D) with LAT −111 ms. Concealed entrainment (Panel E) correlated the ECG-predicted site with the VT exit (PPI-TCL = 0, stim-QRS = EGM-QRS = -111 ms, 30% of TCL). VT termination with ablation at this site with mid-diastolic signals (not shown). Panel (F) The study arm (ECG-mapping group) experienced a significant improvement in freedom from the first occurrence of ATP, defibrillation, or death at 1 year vs. control patients both at baseline and after adjustment using Cox regression survival analysis, adjusting for age, EF, and PAINESD risk score [83% vs. 50%, hazard ratio 0.25 (CI 0.09–0.69), P = 0.007]. VT, ventricular tachycardia; LV, left ventricular; CABG, coronary artery bypass grafting; CTO, chronic total occlusion; LAD, left anterior descending artery; RCA, right coronary artery; LAT, local activation time; PPI, post-pacing interval; TCL, tachycardia cycle length; ATP, anti-tachycardia pacing; EF, ejection fraction; CI, confidence interval. Acute termination of induced VT occurred in 37% of study patients compared with none of the controls (P < 0.001). Substrate-only ablation was used in 17% of controls. Non-inducibility of sustained VT was achieved in 80% of study patients and 67% of controls. At 1 year, there was a significantly higher freedom from the composite endpoint of VT recurrence or death in the study group (80% vs. 43%, HR 0.25, 95% CI 0.10–0.64, P = 0.004) despite adjustment for confounders, (Figure 1, panel F). Individual secondary outcomes showed consistent trends. Freedom from ATP therapy was 97% in the study group vs. 60% in controls; freedom from defibrillation shocks was 87% vs. 80%; survival was 97% vs. 93%. Procedural time (326 ± 64 vs. 311 ± 80 min, P = 0.4), ablation duration (29 ± 12 vs. 29 ± 21 min, P = 0.5), and fluoroscopy use (23 ± 13 vs. 28 ± 17 min, P = 0.3) were similar between groups. No patient in either cohort had acute hemodynamic collapse or required rescue mechanical circulatory support. One study patient and two controls required prolonged ICU care. This study demonstrates that AI-ECG mapping enables focused invasive mapping in patients with hemodynamically unstable VT. Despite a sicker baseline profile, study patients achieved significantly higher 1 year arrhythmia-free survival compared to controls, without increased procedural risk. Traditional strategies for unstable VT rely on substrate modification or pace-mapping, which can miss critical isthmuses and are associated with relatively high recurrence rates.7,8 Data from multicentre registries suggest that more than half of patients with unstable VT experience recurrence within 6 months. Our findings suggest that supplementing invasive mapping with AI-ECG guidance can enable mechanistic mapping in this challenging cohort. The use of AI mapping during brief VT inductions enabled rapid identification of the critical isthmus without hemodynamic collapse. Acute termination was achieved more frequently in the study group, reinforcing that ablation targeted physiologically relevant sites. Importantly, this strategy was safe: outcomes were achieved without increased complications nor prolonging procedure time. This contrasts with experiences using prophylactic hemodynamic support, where complications and mortality have been higher. The clinical implications are significant. AI-ECG mapping allowed activation mapping to be feasible in over 90% of unstable VT, with prior literature achieving only 60% including stable VT.9 This demonstrates that the technology can meaningfully expand the scope of activation mapping in patients who otherwise would have been managed with substrate-only ablation. This workflow is widely applicable, requiring only standard 12-lead ECG data without prolonging procedures (Figure 1).10 These results build on prior work demonstrating the value of VT induction and mapping whenever feasible.8 AI-ECG guidance may enable prioritization of critical substrate and can combine mechanistic mapping with substrate mapping in a ‘single shot’ without adding significantly more time/risk. Because this was not randomized, there is a chance of confounders/bias. We have limited this by matching controls by institution/operator in reverse consecutive order and used a Cox regression model to adjust for clinical confounders. Even though this was

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.