Abstract 4146244: Late Gadolinium Enhancement-Derived Heart Digital Twins Non-Invasively Predict Ventricular Tachycardia Risk in Duchenne Muscular Dystrophy

Introduction: Patients with Duchenne muscular dystrophy (DMD) universally develop dilated cardiomyopathy. Characteristic left ventricular (LV) fibrosis and decreased LV systolic function can increase the likelihood of fatal ventricular arrhythmias (VA). However, the benefits of ICD therapy in DMD remain uncertain, and there are no established VA risk stratification methods to guide clinical decision making. Here, we developed a digital twin approach that uses standard-of-care late-gadolinium enhanced cardiac magnetic resonance (LGE CMR) imaging alone to non-invasively screen for VA risk in DMD. Methods: Thirty-nine DMD patients from three centers with LGE CMR and Ziopatch monitoring were included, with fast nonsustained VT (NSVT; >170 BPM) as the primary clinical outcome (positive control). From short-axis 2D LGE CMR, we constructed 3D computational models of the LV with patient-specific fibrosis distribution (Fig A). These models were separated into positive and negative groups by performing unsupervised clustering on regional substrate percentages (Fig B). DMD-specific EP properties were then assigned to the 3D models, and a rapid pacing protocol was used to assess in silico VT inducibility (Fig C). Patients were predicted to have VA if they tested positive with both clustering and simulation methods. Tests were performed blinded to clinical outcome, cohort positivity rate, and covariates such as ejection fraction. Results: Five patients were observed to have fast NSVT. Our approach achieved 80% (4/5) sensitivity, 79.4% (27/34) specificity, and 79.5% (31/39) accuracy in predicting these events. The sole false negative patient had lower LGE CMR resolution relative to the rest of the cohort (p < 0.01). While extensive LGE was present in all patients, the positive cluster showed higher inferolateral, apical, and total fibrosis burden (p < 0.01) compared to the negative cluster. Without using simulation, specificity and accuracy decreased to 58.8% and 64.1%. Conclusion: LGE-derived digital twin models show strong potential for screening for VA risk in DMD patients. Future application of these models may help define optimal indications for ICD implantation in DMD.