What does this research mean for the field?

Machine learning models utilizing variational autoencoder latent embeddings and engineered electrocardiographic features classify left ventricular hypertrophy with higher diagnostic accuracy than traditional ECG criteria or end-to-end convolutional neural networks. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

March 24, 2026Heart Rhythm O2Open Access

Machine learning to classify left ventricular hypertrophy using electrocardiographic feature extraction by variational autoencoder

View Full Paper

Ask AI

Bookmark

Why the study?

Traditional ECG criteria for left ventricular hypertrophy (LVH) have modest diagnostic yield.

Do machine learning models using extracted ECG features and deep-learning latent embeddings improve the diagnostic classification of left ventricular hypertrophy compared to traditional ECG criteria?

Structured PICO

Do machine learning models using extracted ECG features and deep-learning latent embeddings improve the diagnostic classification of left ventricular hypertrophy compared to traditional ECG criteria?

Population

482,734 ECG-echocardiogram pairs (± 45 days) used for model training and validation, with a hold-out testing set of n=54,984.

Intervention

Machine learning models (Logistic regression, random forest, LGBM, ResNet, MLP) trained on ECG features (calculations, amplitudes, QRS voltage-time integrals) and deep-learning latent embeddings extracted via a variational autoencoder, along with sex.

Comparator

Best individual ECG variable (VTI QRS-Z), best traditional ECG criterion (Cornell voltage-duration product), and a convolutional neural network (CNN) trained on raw ECG signals alone.

Outcome

Area under the receiver operating characteristic curve (ROC-AUC) for classifying left ventricular hypertrophy (LV mass indexed >95 g/m2 for women, >115 g/m2 for men).surrogate

Machine learning models utilizing extracted ECG features and deep-learning latent embeddings significantly outperform traditional ECG criteria and raw-signal CNNs for detecting left ventricular hypertrophy.

Implication

ML ECG models may aid LVH detection; leaves open prospective validation before clinical adoption.

Abstract

Background Traditional ECG criteria for left ventricular hypertrophy (LVH) have modest diagnostic yield. Objective Develop and validate machine learning models for LVH diagnosis from ECG. Methods ECG calculations (rate, intervals, axis), R-wave, S-wave and overall-QRS amplitudes, and QRS voltage-time integrals (VTI QRS ) were obtained from 12-leads, vectorcardiographic X-Y-Z-leads, and 3D (L2-norm)-lead ECGs. Deep-learning enabled latent embeddings (30 per ECG) were extracted using a variational autoencoder (pretrained on unselected 1.18 million ECGs) from representative-beat signals. Logistic regression, random forest, light gradient boosted machine (LGBM), residual network (ResNet) and multilayer perceptron network (MLP) models using ECG features (calculations and embeddings) and sex, and a convolutional neural network (CNN) using ECG signals alone, were trained to predict LVH (LV mass indexed, women >95 g/m 2 , men >115 g/m 2 ) on 482,734 ECG-echocardiogram pairs (± 45 days). ROC-AUCs were reported from a hold-out testing set. Results In the testing set (n = 54,984), AUC for LVH classification was higher for ML models using ECG features (LGBM 0.794, MLP 0.793, ResNet 0.795) compared with the best individual ECG variable (VTI QRS-Z 0.707), the best traditional criterion (Cornell voltage-duration product 0.716), and the CNN using ECG signals (0.788). Among patients without LVH who had a follow-up echocardiogram >1 year later, LGBM false-positives, compared to true-negatives, had a 3.07 (95% CI 2.44, 3.86)-fold higher odds of developing future LVH (p<0.0001). Conclusions ML models are superior to traditional ECG criteria for classifying LVH. Models trained on extracted ECG features, including deep-learning latent space representations, can outperform CNN models trained on ECG signals.