Improved recognition of sustained ventricular tachycardia from SAECG by support vector machine

Area of Science:

• Cardiology
• Biomedical Engineering
• Machine Learning

Background:

• Sustained ventricular tachycardia (SVT) is a critical condition following myocardial infarction (MI).
• Accurate risk stratification is essential for managing patients at risk of SVT.
• Traditional signal-averaged electrocardiography (SAECG) methods have limitations in SVT detection.

Purpose of the Study:

• To develop and validate an improved method for recognizing sustained ventricular tachycardia (SVT).
• To enhance the accuracy of SAECG analysis by incorporating new parameters and filtering techniques.
• To apply a support vector machine (SVM) classifier for robust SVT classification.

Main Methods:

• A dataset of 376 patients (SVT+, SVT-, healthy controls) was analyzed.
• SAECG analysis utilized two filters: IIR Butterworth and Finite Impulse Response (FIR) with Kaiser window.
• Nine SAECG parameters were calculated, including six novel ones, alongside standard parameters.
• A support vector machine (SVM) classifier with a Gaussian kernel was employed for SVT+ classification.

Main Results:

• The FIR filter combined with 9 SAECG parameters yielded the highest recognition score of 95.21%.
• Models demonstrated good generalization capabilities on validation data.
• The improved method significantly outperformed standard SAECG analysis in SVT recognition.

Conclusions:

• The developed method, utilizing FIR filtering, extended SAECG parameters, and SVM, significantly improves SVT risk stratification.
• This approach enhances diagnostic accuracy for sustained ventricular tachycardia, achieving up to 95% recognition.
• The study highlights the potential of advanced signal processing and machine learning in cardiac arrhythmia detection.