Sparse-EWI: Towards an automated and real time focal arrhythmia mapping technique - in silico to clinical validation

Cardiac arrhythmias are a leading cause of morbidity and mortality worldwide, and identifying the origin of focal ventricular arrhythmias such as Premature Ventricular Contractions (PVC) is essential for precision radiofrequency ablation treatment. Electromechanical Wave Imaging (EWI) is a noninvasive high-frame-rate echocardiography-based technique that maps the cardiac electromechanical activation throughout the entire heart. However, estimation of activation times across the myocardium requires manual annotation of approximately 150 points per apical view, which is labor-intensive and limits clinical adoption. This study introduces an efficient, clinically applicable sparse-EWI method, in which only 15 seed points per view are manually annotated and activation times at the remaining points are estimated automatically through recursive activation-time selection based on the strain correlation between adjacent points. Sparse-EWI was compared to conventional EWI using a simulated electromechanical heart model coupled with a 3-D Field II ultrasound simulation, an open-chest canine paced at five different left-ventricular locations, and three PVC patients, with high-frame-rate acquisitions performed at 2000 fps using a 2.5 MHz phased-array transducer and a single diverging-wave transmit. Sparse-EWI reduced the number of initialization points by 10× without compromising accuracy. Relative to conventional EWI, the sparse-EWI activation-time maps achieved correlation coefficients of 0.74, 0.82±0.13, and 0.75±0.12; accuracies of 66.64%, 85.9±7.4%, and 84.2±8.2%; and source-localization errors of 7.10 mm, 2.40±2.80 mm, and 1.91±2.36 mm across the simulated, preclinical, and clinical datasets, respectively. The spatial reduction also resulted in approximately five-fold higher computational speed compared to conventional EWI, decreasing processing time from ∼900 s to ∼182 s per view. In summary, sparse-EWI achieved a 90% reduction in required initialization points and approximately five-fold faster processing across simulated, preclinical, and clinical datasets, increasing EWI's translation and providing rapid feedback during ablative treatment.