Malignant arrhythmia risk assessment based on lead-I mobile ECG measurements using machine learning

Area of Science:

• Cardiology
• Biomedical Engineering
• Machine Learning

Background:

• Malignant arrhythmias pose a significant health risk.
• Accurate and timely risk assessment is crucial for patient outcomes.
• Current methods often require complex equipment or extensive data.

Purpose of the Study:

• To develop and validate an algorithm for estimating malignant arrhythmia risk using only a short lead-I ECG recording.
• To evaluate the performance of different machine learning classifiers for this task.
• To explore the potential for mobile, out-of-hospital risk assessment.

Main Methods:

• Feature extraction using mean and relative standard deviation of beat-to-beat QRST integrals.
• Training supervised machine learning models including 3 nearest neighbors (3-NN) and 1-D/2-D Bayesian classifiers.
• Validation using a test set of healthy subjects and patients with confirmed malignant arrhythmia records.

Main Results:

• The 2-D Bayesian classifier achieved the highest decision efficiency (87.30% for normal, 94.23% for malignant arrhythmia).
• The 3-NN classifier also showed strong performance (80.95%, 94.23%).
• Using QTc parameter yielded lower efficiency compared to QRST integrals.

Conclusions:

• The proposed algorithm, particularly with the 2-D Bayesian classifier, effectively estimates malignant arrhythmia risk from single-channel ECG.
• This method offers a valuable tool for portable ECG systems, facilitating out-of-hospital risk assessment.
• It represents a novel approach requiring only lead-I ECG for efficient risk stratification.