Stationary wavelet transform based technique for automated external defibrillator using optimally selected classifiers

Early and accurate detection of ventricular fibrillation (VF) and rapid ventricular tachycardia (VT) is vital for defibrillation therapy. Various techniques have been proposed based on various parameters extracted from the electrocardiogram (ECG), which are mostly slow and requires comparatively wider ECG segment. The proposed technique requires a 4.1 s segment of ECG which results in an early detection and thus proposed to help in timely diagnosis and treatment of these life-threatening arrhythmias. Stationary Wavelet Transform (SWT) has been used for decomposition of the signal followed by calculation of sample entropy of the wavelet bands selected using filter-type feature selection procedure. Sample entropy of these bands, working as attributes for the classifier was fed to three different classifiers, Support Vector Machine (SVM), k-Nearest Neighbors (k-NN), and Random Forest Algorithm (RFA) and their performance was analyzed with variation in their key model parameters. The proposed technique has been analyzed in two scenarios; VFVT vs Non-VFVT and VF vs Non-VF. The Sensitivity (Se%), Specificity (Sp%), Positive Predictivity (+P%), Accuracy (Ac%), and area under Receiver Operating Characteristics Curve (Roc%) analyzed over Creighton University Ventricular Tachyarrhythmia (CUVT) database were used for performance analysis and comparison. As per observations of the results, under VFVT vs Non-VFVT scenario SVM has highest +P%=96.53 and Sp%=97.08, RFA has the highest Roc%=98.10, and k-NN has the highest Se%=95.64 and Ac%=96.01. For VF vs non-VF classification, SVM gives best Sp%=96.86, RFA has the highest Roc%=98.00,Se%=95.74 and Ac%=95.80, and k-NN gives best +P%=96.52.