Ventricular Fibrillation Detection by an Improved Time Domain Algorithm Combined with SVM

Abstract

Correct detection of ventricular fibrillation (VF) is of great importance to real-time electrocardiogram (ECG) monitoring systems and automatic external defibrillator (AED). First, the paper gives a brief review of threshold crossing sample count algorithm (TCSC), and analyzes this algorithm's drawbacks. Then the authors present an improved algorithm combined TCSC with support vector machine (SVM), which is more accuracy than the TCSC algorithm. For assessment of the performance of the algorithm, the complete CU database and MIT-BIH database are used. The authors compare the new algorithm with other VF detection methods under the same conditions. The ROC curve is created and the AUC is also calculated. The results show that the proposed algorithm has a high Accuracy of 91.2%, Specificity of 96.8%, and the AUC is 92.5. The new algorithm is fast, accurate and reliable, showing strong potential to be applied in real-time ECG monitor system.