Machine learning-based information flow analysis of ECG signals for long QT syndrome

Area of Science:

• Cardiology
• Biophysics
• Computational Biology

Background:

• Cardiovascular diseases are a leading global cause of death, necessitating advanced non-invasive risk assessment.
• The heart's complex nonlinear dynamics involve subsystem interactions, quantifiable by information-theoretic measures.
• Entropy-based methods offer a framework to analyze information flow and detect pathological cardiac alterations.

Purpose of the Study:

• To investigate information flow between heart rhythm and ventricular repolarization using ECG.
• To identify potential markers for pathological alterations in cardiac electrical activity.
• To discriminate patients with congenital long QT syndrome (LQTS) from healthy controls using machine learning.

Main Methods:

• Derived entropy-based measures of information transfer from beat-to-beat ECG time series.
• Employed supervised machine learning models (Random Forest, Support Vector Machine) for classification.
• Utilized SHapley Additive exPlanations (SHAP) for model explainability and feature importance assessment.

Main Results:

• Achieved high and stable classification performance in distinguishing LQTS patients from controls.
• Random Forest model showed mean accuracy of 95.9%, sensitivity of 95.9%, and specificity of 92.9%.
• Support Vector Machine model achieved mean accuracy of 93.1%, sensitivity of 93.1%, and specificity of 92.0%.

Conclusions:

• Multivariate and conditional information flow features were more critical than single-source entropy measures.
• Joint and conditional interactions in cardiac electrical activity are relevant for classifying LQTS.
• The approach demonstrates potential for non-invasive risk assessment of cardiovascular conditions.