A novel feature extraction-based ECG signal analysis

Abstract

An electrocardiogram (ECG) is an essential and fundamental diagnostic tool for assessing cardiac arrhythmias. It is mainly a combination of P, QRS, and T waves. But visual inspection of these waves may lead to wrong diagnosis due to physiological variability and noisy QRS complexes. Hence, computer-aided diagnosis (CAD) is required for accurate and efficient diagnosis of the clinical information. Therefore, in this paper, a novel technique, i.e., fractional wavelet transform (FrWT) is proposed to be used as a feature extraction technique. Afterward, Probabilistic Principal Component Analysis (PPCA) and K-Nearest Neighbor (KNN) are jointly used as classification (i.e., detection of R-peaks) tools for diagnosing heart abnormalities in various morphologies of the ECG signal robustly. The proposed technique has been evaluated on the basis of sensitivity (Se), detection error rate (Der), and positive predictive value (Ppv) for records in the MIT-BIH Arrhythmia database (M/B Ar DB). The proposed technique yields Se of 99.98%, Der of 0.036%, and Ppv of 99.98% for M/B Ar DB. These results establish robustness of the proposed technique, which will go a long way in assisting the cardiologists in improving overall health care system in hospitals.