Since from last two decades the electrocardiogram (ECG) signals can used to predict the heart conditions. While designing the ECG signals based heart prediction framework, one must consider the key steps preprocessing, features extraction and selection, and categorization. The conventional methods mainly rely on methods those are tedious, time consumption, and bound to specific pattern of signals. The current pre-processing methods are not working dynamically as the signals gets pre-processed even if there is no artefacts /or noise present. The ineffective pre-processing leads the inaccurate prediction results, thus designing robust pre-processing technique that removes the artefacts and noise from the raw ECG signals without distorting the original heart beats. In this paper, we proposed such pre-processing algorithm using the twin median filtering of different frequency effects followed by the notch filtering to eliminate the power line interference. The pre-processed signal further used for the features extraction. To estimate the important beats like Q, R, and S of ECG signal, we proposed the Normalized Stationary Wavelet Transform (NSWT) in which the pre-processed signal is normalized prior applying the SWT. Rather than using the just approximation coefficients of NSWT for the QRS complex extraction, we used the both approximation and detail coefficients of third level to extract the features. The individual features further combined and selected using the Discrete Coefficient Transform (DCT). The proposed framework is evaluated using different classifiers and with state-of-art methods to claim the robustness and efficiency.
Volume 12 | Issue 7
Pages: 380-394
DOI: 10.5373/JARDCS/V12I7/20202019