Clinical validation of a novel fully automated measurement method for echocardiographic mechanical dispersion using deep learning

Abstract

Listen

Translate article

Abstract Background Evaluation of left ventricular (LV) function by strain echocardiography and mechanical dispersion (MD) has demonstrated improved prediction of ventricular arrhythmia in patients with heart failure (HF) and reduced ejection fraction. However, analyzing MD is both time consuming and operator dependent, thus its clinical use remains limited. Fully automated measurements of MD using a novel deep learning (DL) method may be a solution to facilitate clinical implementation. Purpose We aimed to validate a novel DL based method for fully automated measurements of MD, by evaluating its correlation and agreement with manual reference measurements in a large cohort of patients with heart failure and left ventricular ejection fraction (LVEF) <40%. Methods In this prospective multicenter follow-up study, we consecutively included patients with recently diagnosed heart failure of all causes with LVEF ≤ 40%. All patients were treated according to contemporary guidelines with optimal medical therapy and revascularization if necessary. Two experienced cardiologists recorded all echocardiograms using Vivid E9 or E95 scanners (GE HealthCare, Horten, Norway). Measurements of MD were performed both by an experienced cardiologist using a commercial semi-automatic method, and an in-house developed fully automated deep learning-based method. Results MD was measured in 200 patients (73 ± 14 years old, 33% females). Mean LVEF was 32% ± 6. Mean MD by reference measurements was 61 ± 17ms and by the deep learning method 63 ± 19 ms . Deep learning-based measurements of MD had excellent agreement with reference, with bias of 2ms and Limits of Agreement only 28ms (Figure 1). The correlation between methods was good, with R2 46%, p<0.001. Calculation time for automated measurements of MD per patient was only 6.4 ± 2.5 seconds per echocardiogram. Conclusion The novel fully automated deep learning-based method provides rapid measurements of MD and excellent agreement with manual reference measurements, and may thus facilitate clinical implementation of MD for improved risk stratification in patients with HF.