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

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Traditional electrocardiogram (ECG) criteria for diagnosing left ventricular hypertrophy (LVH) demonstrate limited diagnostic accuracy.
• There is a need for improved diagnostic tools for LVH, a condition associated with significant cardiovascular risk.

Purpose of the Study:

• To develop and validate machine learning (ML) models for the diagnosis of LVH using ECG data.
• To compare the diagnostic performance of ML models against established ECG criteria and deep learning approaches.

Main Methods:

• Extracted diverse ECG features, including calculations, QRS amplitudes, voltage-time integrals, and deep learning-derived embeddings, from 12-lead, vectorcardiographic, and 3D ECGs.
• Trained various ML models (Logistic Regression, Random Forest, LGBM, ResNet, MLP) and a Convolutional Neural Network (CNN) using these features and patient sex.
• Validated models on a large dataset of 482,734 ECG-echocardiogram pairs, reporting Area Under the Receiver Operating Characteristic Curves (AUC).

Main Results:

• ML models utilizing ECG features achieved superior AUC values (e.g., LGBM 0.794, ResNet 0.795) compared to the best traditional criterion (Cornell voltage-duration product 0.716) and the CNN (0.788).
• In patients without baseline LVH, ML model false positives showed a significantly higher likelihood (3.07-fold odds) of developing future LVH over one year.
• The study demonstrated the effectiveness of deep learning embeddings within ML models for enhanced LVH classification.

Conclusions:

• Machine learning models significantly outperform traditional ECG criteria in classifying left ventricular hypertrophy.
• Models leveraging extracted ECG features, including deep learning representations, demonstrate superior performance over CNNs trained solely on ECG signals.
• ML-based ECG analysis holds promise for more accurate and predictive diagnosis of LVH.