Detection of Left Ventricular Systolic Dysfunction from Electrocardiographic Images

Abstract

Background: Left ventricular (LV) systolic dysfunction is associated with over 8-fold increased risk of heart failure and a 2-fold risk of premature death. The use of electrocardiogram (ECG) signals in screening for LV systolic dysfunction is limited by their availability to clinicians. We developed a novel deep learning-based approach that can use ECG images for the screening of LV systolic dysfunction. Methods: Using 12-lead ECGs plotted in multiple different formats, and corresponding echocardiographic data recorded within 15 days from the Yale-New Haven Hospital (YNHH) during 2015-2021, we developed a convolutional neural network algorithm to detect LV ejection fraction < 40%. The model was validated within clinical settings at YNHH as well as externally on ECG images from Cedars Sinai Medical Center in Los Angeles, CA, Lake Regional Hospital (LRH) in Osage Beach, MO, Memorial Hermann Southeast Hospital in Houston, TX, and Methodist Cardiology Clinic of San Antonia, TX. In addition, it was validated in the prospective Brazilian Longitudinal Study of Adult Health (ELSA-Brasil). Gradient-weighted class activation mapping was used to localize class-discriminating signals in ECG images. Results: Overall, 385,601 ECGs with paired echocardiograms were used for model development. The model demonstrated high discrimination power across various ECG image formats and calibrations in internal validation (area under receiving operation characteristics [AUROC] 0.91, area under precision-recall curve [AUPRC] 0.55), and external sets of ECG images from Cedars Sinai (AUROC 90, AUPRC 0.53), outpatient YNHH clinics (AUROC 0.94, AUPRC 0.77), LRH (AUROC 0.90, AUPRC 0.88), Memorial Hermann Southeast Hospital (AUROC 0.91, AUPRC 0.88), Methodist Cardiology Clinic (AUROC 0.90, AUPRC 0.74), and ELSA-Brasil cohort (AUROC 0.95, AUPRC 0.45). An ECG suggestive of LV systolic dysfunction portended over 27-fold higher odds of LV systolic dysfunction on TTE (OR 27.5, 95% CI, 22.3-33.9 in the held-out set). Class-discriminative patterns localized to the anterior and anteroseptal leads (V2-V3), corresponding to the left ventricle regardless of the ECG layout. A positive ECG screen in individuals with LV ejection fraction ≥ 40% at the time of initial assessment was associated with a 3.9-fold increased risk of developing incident LV systolic dysfunction in the future (HR 3.9, 95% CI 3.3-4.7, median follow-up 3.2 years). Conclusions: We developed and externally validated a deep learning model that identifies LV systolic dysfunction from ECG images. This approach represents an automated and accessible screening strategy for LV systolic dysfunction, particularly in low-resource settings. CLINICAL PERSPECTIVE: What is New?: A convolutional neural network model that accurately identifies LV systolic dysfunction from ECG images across subgroups of age, sex, and race. The model shows robust performance across multiple institutions and health settings, both applied to ECG image databases as well as directly uploaded single ECG images to a web-based application by clinicians. The approach provides information for both screening of LV systolic dysfunction and its risk based on ECG images alone. What are the clinical implications?: Our model represents an automated screening strategy for LV systolic dysfunction on a variety of ECG layouts. With availability of ECG images in practice, this approach overcomes implementation challenges of deploying an interoperable screening tool for LV systolic dysfunction in resource-limited settings. This model is available in an online format to facilitate real-time screening for LV systolic dysfunction by clinicians.