Performance Evaluation of Various Pre-Processing Techniques for R-Peak Detection in ECG Signal

Abstract

ABSTRACT In recorded Electrocardiogram (ECG) signal, clinical information is masked by several noises and distortion resulting in low signal-to-noise-ratio (SNR). In this situation, an efficient pre-processing technique is required to improve SNR for efficient analysis of ECG signals. In this research article, performance of five pre-processing techniques viz. digital bandpass filter (DBPF), wavelet transform (WT), independent principal component analysis (IPCA), savitzky golay digital FIR filter (SGDFF) and fractional wavelet transform (FrWT) have been evaluated and compared for their effects on the efficiency of R-peak detection. FrWT has been utilized for pre-processing of ECG signal for the first time in this paper. A FrWT-based technique is also proposed using Yule Walker autoregressive modeling (YWARM) and Principal Component Analysis (PCA) for feature extraction and R-peak detection, respectively. YWARM is selected due to its more stable output for long time recorded ECG signal than existing techniques, whereas PCA is selected to get optimal dimensional feature vectors out of higher dimensional feature vectors. The proposed technique has been evaluated and compared with others on the basis of various performance parameters; SNR, mean squared error (MSE), sensitivity (SE), accuracy (Acc), and positive predictive value (PPV). The proposed technique yielded interesting results among all the methods; 34.37 dB of output SNR, 0.026% of MSE, 99.98% of SE, 99.97% of Acc and 99.99% of PPV on real time ECG database (RT DB) and 24.81 dB of output SNR, 0.099% of MSE, 99.96% of SE, 99.93% of Acc, and 99.97% of PPV on MIT-BIH Arrhythmia database (MIT-BIH Arr DB).