Segment Specific Modeling of Electrocardiogram for Improved Reconstruction Error

TLDR: Segment specific modelling approach for different wave segments of ECG signal provides better reconstruction performance in comparison with the few published works using Gaussian and Fourier model.

Abstract: Electrocardiogram (ECG) modeling is useful for abnormality detection and data compression. The common research problem in modeling is retaining pathological information using minimum number of model coefficients. In this paper, a new modeling technique for different wave segments of ECG signal, viz., baseline to P-onset, P wave, P-offset to Q, QRS complex, S to T-onset, T wave and T-offset to next baseline is presented. The processing steps included preprocessing, R-peak detection, beat segmentation and waveform partitioning, followed by modeling of individual partitions. For P, QRS and T wave, Gaussian model was adopted and for other segments, Fourier model was adopted to minimize reconstruction error. For testing of the proposed model, normal sinus rhythm (NSR) and myocardial infarction (MI) data records of PTB Diagnostic ECG database (ptbdb) and atrial premature (APC), premature ventricular contraction (PVC), left bundle branch block (LBBB) and right bundle branch block (RBBB) data records of MIT-BIH arrhythmia database (mitdb) under PhysioNet were used. The average SNR, and MSE using proposed method for ptbdb NSR was 86.33, and 4.41×10-6, respectively; for AMI 96.18, and 3.70×10-6 respectively; for IMI 80.86, and 1.36×10-6 respectively; for mitdb NSR 90.94 and 3.50×10-6 respectively; for APC 89.42, and 2.34×10-6 respectively; for PVC 93.28 and 3.06×10-6, respectively; for LBBB 93.77 and 2.74×10-6, respectively; for RBBB 92.83 and 3.52×10-6 respectively. Segment specific modelling approach provides better reconstruction performance in comparison with the few published works using Gaussian and Fourier model.