Predicting left ventricular hypertrophy from the 12-lead electrocardiogram in the UK Biobank imaging study using...

Area of Science:

• Cardiology
• Medical Imaging
• Biomarkers

Background:

• Left ventricular hypertrophy (LVH) is a significant predictor of cardiovascular disease.
• Electrocardiogram (ECG) indices offer limited sensitivity for LVH detection.
• Cardiovascular magnetic resonance (CMR) imaging is a definitive method for LVH diagnosis.

Purpose of the Study:

• To classify LVH using 12-lead ECG data.
• To develop cost-effective patient stratification tools for LVH.
• To integrate ECG biomarkers and clinical variables for improved LVH prediction.

Main Methods:

• Extracted ECG biomarkers from 37,534 UK Biobank participants.
• Developed classification models using logistic regression, SVM, and RF.
• Utilized 80% training and 20% test sets with 10-fold cross-validation.

Main Results:

• QRS amplitude and blood pressure were key predictors of LVH.
• Logistic regression achieved 81% accuracy (AUC 0.86), SVM 81% (AUC 0.85), and RF 72% (AUC 0.83).
• ECG biomarkers significantly enhanced classifier performance over clinical variables alone.

Conclusions:

• Combined ECG biomarkers and clinical variables accurately predict CMR-defined LVH.
• The ECG serves as an economical screening tool for LVH risk stratification.
• Validated models demonstrate robustness for clinical application.