Ventricular Fibrillation and Ventricular Tachycardia Classification Using Support Vector Machine

Abstract:
Correct detection and classification of ventricular fibrillation and ventricular tachycardia are crucial for the monitoring of patients with cardiovascular diseases. This study focuses on designing and optimizing a classification algorithm based on support vector machine (SVM) specifically for detecting ventricular tachycardia and ventricular fibrillation, using the Creighton University Ventricular Tachycardia Database (CUDB). Optimization was targeted at enhancing classifier performance through sophisticated data preprocessing, meticulous feature extraction, and refined classifier design. During the preprocessing stage, a comparative analysis was conducted between digital filtering and wavelet threshold denoising methods, evaluated based on signal-to-noise ratio, root mean square error, and time complexity metrics, to ascertain the most appropriate preprocessing technique. Subsequently, a feature extraction framework that integrates clinical expertise and functional analysis tools was developed. This framework employed the discrete Fourier transform to augment the feature information. The random forest algorithm was employed to prioritize and discern the optimal subset of features. In the final phase, the support vector machine parameters were optimized through the application of the particle swarm optimization algorithm, and a rigorous 10-fold cross-validation was performed over a 5-second time window to ensure robustness. The experimental outcomes reveal that the methodology proposed herein exhibits a sensitivity exceeding 98% on the CUDB dataset, complemented by an accuracy of over 97% and a specificity of 96%, thereby demonstrating a formidable classification capability for differentiating between ventricular tachycardia and ventricular fibrillation.