ECG-based prediction algorithm for imminent malignant ventricular arrhythmias using decision tree.

Area of Science:

• Biomedical Engineering
• Cardiology
• Machine Learning in Healthcare

Background:

• Spontaneous prediction of malignant ventricular arrhythmia (MVA) using electrocardiogram (ECG) is vital for prompt medical intervention.
• Existing MVA prediction algorithms face challenges including unclear feature impact, potential prediction delays, and performance uncertainty.

Purpose of the Study:

• To investigate the optimal number and types of ECG features for MVA prediction using a decision tree classifier.
• To minimize MVA warning delays by analyzing algorithm execution time prior to arrhythmia onset.
• To evaluate MVA prediction algorithm performance through sensitivity and specificity analysis.

Main Methods:

• Conducted a literature review on existing MVA prediction studies.
• Designed and developed four modules for MVA prediction, focusing on feature selection and classification.
• Compared decision tree classifiers with support vector machine and naive Bayes algorithms.

Main Results:

• Eight ECG features combined with a decision tree classifier demonstrated effective prediction performance regarding execution time and sensitivity.
• The highest sensitivity (95%) and specificity (90%) were achieved in the interval 15.1–20 minutes preceding MVA.
• Comparative analysis with other classifiers was performed to validate results.

Conclusions:

• The optimal prediction window for MVA is approximately 15-20 minutes before the event.
• A decision tree classifier utilizing eight ECG features provides a robust approach for early MVA detection.
• This approach enhances prediction accuracy and minimizes critical delays in rescue operations.