Artificial Intelligence Enhanced Electrocardiogram Analysis for Age and Sex Classification in Youth

Area of Science:

• Cardiology
• Biomedical Engineering
• Machine Learning

Background:

• Electrocardiogram (ECG) values exhibit significant age and sex variations, especially in pediatric populations.
• Existing age- and sex-specific ECG standards may not fully capture complex relationships and are underutilized in machine learning (ML) applications.
• Automated ECG analysis using ML has improved clinical accuracy, but pediatric studies are scarce.

Purpose of the Study:

• To develop age- and sex-specific standards for pediatric electrocardiograms (ECGs) using machine learning (ML) modeling.
• To enhance the accuracy of automated ECG analysis in children and adolescents.
• To investigate the utility of ML in establishing new pediatric ECG reference ranges.

Main Methods:

• Analysis of 29,408 curated resting 12-lead ECGs from healthy individuals aged 0-21 years.
• Utilized 177 digitized ECG variables with various ML models, including regression, classification, and semi-supervised neural networks.
• Evaluated model performance using F1-score, AUROC, and confusion matrices across repeated train-test splits.

Main Results:

• Support Vector Machine (SVM) demonstrated the highest accuracy in modeling both age and sex.
• Key predictive ECG features included heart rate, PR interval, QRS duration, and T-wave amplitude.
• SVM achieved 94% accuracy for age-group classification (allowing one-group misclassification) and high F1-scores (0.91) and AUROC (0.95) for sex classification in adolescents.

Conclusions:

• Supervised ML models effectively capture age- and sex-related physiological ECG changes, outperforming semi-supervised approaches, especially in smaller subgroups.
• Findings support the creation of age- and sex-specific ML-enhanced ECG standards for pediatric cardiology research and clinical practice.
• This study highlights the potential of ML to refine diagnostic criteria for pediatric ECG interpretation.