Identifying Mitral Valve Prolapse at Risk for Arrhythmias and Fibrosis From Electrocardiograms Using Deep Learning

Abstract

Background: Mitral valve prolapse (MVP) is a common valvulopathy, with a subset developing sudden cardiac death or cardiac arrest. Complex ventricular ectopy (ComVE) is a marker of arrhythmic risk associated with myocardial fibrosis and increased mortality in MVP.

Objectives: The authors sought to evaluate whether electrocardiogram (ECG)-based machine learning can identify MVP at risk for ComVE, death and/or myocardial fibrosis on cardiac magnetic resonance (CMR) imaging.

Methods: A deep convolutional neural network (CNN) was trained to detect ComVE using 6,916 12-lead ECGs from 569 MVP patients from the University of California-San Francisco between 2012 and 2020. A separate CNN was trained to detect late gadolinium enhancement (LGE) using 1,369 ECGs from 87 MVP patients with contrast CMR.

Results: The prevalence of ComVE was 28% (160/569). The area under the receiver operating characteristic curve (AUC) of the CNN to detect ComVE was 0.80 (95% CI: 0.77-0.83) and remained high after excluding patients with moderate-severe mitral regurgitation [0.80 (95% CI: 0.77-0.83)] or bileaflet MVP [0.81 (95% CI: 0.76-0.85)]. AUC to detect all-cause mortality was 0.82 (95% CI: 0.77-0.87). ECG segments relevant to ComVE prediction were related to ventricular depolarization/repolarization (early-mid ST-segment and QRS from V₁, V₃, and III). LGE in the papillary muscles or basal inferolateral wall was present in 24% patients with available CMR; AUC for detection of LGE was 0.75 (95% CI: 0.68-0.82).

Conclusions: CNN-analyzed 12-lead ECGs can detect MVP at risk for ventricular arrhythmias, death and/or fibrosis and can identify novel ECG correlates of arrhythmic risk. ECG-based CNNs may help select those MVP patients requiring closer follow-up and/or a CMR.

Keywords: artificial intelligence; computers; echocardiography; valvular heart disease.

Graphical abstract

Figure 1

Echocardiographic and ECG Examples of MVP Bileaflet MVP (A) and inverted T-wave inversions in the inferior ECG leads (B). Posterior MVP (D) without repolarization abnormalities on ECG (E). Both MVP cases had complex ventricular ectopy (Cand F). ECG = electrocardiogram; MVP = mitral valve prolapse.

Figure 2

Diagram of Study Cohorts and Data Sets MVP patients with and without complex ventricular ectopy (ComVE) were randomly split into training, development, and test data sets. AF = atrial fibrillation; ECG = electrocardiogram; L/RBBB = left/right bundle branch block; MVP = mitral valve prolapse; PVCs = premature ventricular contractions.

Central Illustration

Identifying Mitral Valve Prolapse at Risk for Ventricular Arrhythmias and Myocardial Fibrosis From 12-Lead Electrocardiograms Using Deep Learning AUC = area under the receiver operating characteristic curve; CNN = convolutional neural network; ComVE = complex ventricular ectopy; dx = diagnosis; ECG = electrocardiogram; LGE = late gadolinium enhancement.

Figure 3

Performance of the Convolutional Neural Network (A) Receiver operating characteristic curve for the CNN to predict ComVE (green), all-cause mortality (magenta), and composite cardiac death (orange). (B) Confusion matrix demonstrating CNN performance to predict ComVE by ECG in the holdout test data set at the chosen score threshold of 0.39. AUC = area under the receiver operating characteristic curve; CNN = convolutional neural network; ComVE = complex ventricular ectopy.

Figure 4

Distribution of ComVE CNN Scores by Strata in the Test Data Set Box and whisker plots showing (A) distributions of ComVE CNN scores for mitral valve prolapse (MVP) patients with and without ComVE; (B) Distributions of ComVE CNN scores for ComVE patients with and without progression. (C) Averaged CNN scores (y-axis) by patient strata (ComVE, ComVE with progression, and no ComVE) plotted over time, as a percent of total follow-up time since ComVE diagnosis (x-axis). The red line indicates CNN score threshold used for binary classification of ComVE. CNN = convolutional neural network; ComVE = complex ventricular ectopy.

Figure 5

Importance of ECG Segments to the Prediction of ComVE Highlighted ECG segments indicate the top 10 most important ECG segments for prediction of ComVE. ComVE = complex ventricular ectopy; ECG = electrocardiogram.

Figure 6

Performance of the ECG-Based CNN to Detect LGE by CMR in the Test Data Set (A) CMR showing LGE in the papillary muscles (arrows). (B) Confusion matrix demonstrating CNN performance to detect LGE with performance metrics shown in the chart (lower). AUC = area under the receiver operating characteristic curve; CMR = cardiac magnetic resonance; CNN = convolutional neural network; ECG = electrocardiogram; LGE = late gadolinium enhancement; NPV = negative predictive value; PPV = positive predictive value.