| Due to physical variability of ECG waves, detection of the QRS complex becomes a difficult task in a real time situation. Jiau Pan and Willis J. Tompkins of the University of Wisconsin developed a real time QRS detection algorithm for a Z-80 microprocessor. They demonstrated a overall performance of 99.325% when tested against the MIT-BIH arrhythmia database.;Pan and Tompkins performed denoising of the ECG signal using a Band Pass filter which was built using cascaded High Pass and Low Pass filters. Pan and Tompkins used to detect fiducial point by finding the highest squared slope during the high wave energy field, which resulted in much more number of fiducial points than the actual QRS complexes. In this situation we start looking for the fiducial points on negative slope where the square of it was more than the positive slope. The adaptive thresholds, then applied, were the highest among the two threshold extracted out of ECG signal and the integration of the ECG signal. These situations make the complexity more than what is needed actually.;Looking at the non-stationary behavior of the ECG signal where T-wave and R-wave exhibit the same frequency characteristic over different times, we propose to use Wavelet denoising method, which works better than the Band Pass filtering suggested by Pan-Tompkins.;We found that the two threshold needed for the detection is not necessary because the one threshold calculated using integration of the signal is always low. Wherein, the maximum of the two threshold suggested by Pan and Tompkins ignores the threshold from integration wave all the time. Therefore, using only one threshold extracted from ECG signal is sufficient enough for the detection.;We recommend to use the multiscale wavelet peak detection for the fiducial point extraction from the ECG wave which eliminates the negative slope points from being observed as fiducial point. This reduces the complexity of algorithm by half compared to Pan-Tompkins.;Finally, we implemented the modified algorithm in the LabVIEW graphical programming language, and tested it against first 60 seconds of the entire 48 MIT-BIH Arrhythmia database files. Overall performance of modified algorithm were found to be 99.48% which is better than the overall performance of the Pan-Tompkins algorithm. Overall performance of modified algorithm was found to be 99.93% applied on the first 60 seconds of first 23 data files in MIT-BIH arrhythmia database.;We also compared our results with one of the commercially available software from Monebo Technologies, Inc.;Keywords: ECG Morphology, QRS complex, R-wave, T-wave, Fiducial mark, Wavelet De-noising, Wavelet de-trending, Estimation, Detection, arrhythmia, cardiac causes, ECG heterogeneity, LabVIEW, Monebo, MIT-BIH arrhythmia database, R-wave onset, R-wave offset. |
