Fig. 4

De-trending experiments on one QT database patient (No.sel15814) using sym3 wavelet based on different trends. (1) a trend with a sinusoidal wave at 0.3 Hz added to a normalized raw ECG data (only wavelet coefficients at levels 1–7 were preserved), (a) the normalized raw ECG signal, (b) semi-synthetic ECG signal formed by superimposing the normalized raw ECG signal and a simulated trend, (c) the normalized processed semi-synthetic data with removed trend after applying a WT, (d) the extracted trend from (b); the MSE between (a) and (c) was 0.0018 (2) a trend with a step function added to a normalized raw ECG data (only wavelet coefficients at levels 1–7 were preserved), (a) to (d) following the same process in (1), (e) the reconstructed signal by preserving wavelet coefficients only at level 1 and removing others at other levels; MSE between (a) and (c) was 0.0337 (3) a trend with a spike added to a normalized raw ECG data (only wavelet coefficients at levels 3–7 were preserved), (a) to (e) following the same process in (2); MSE between (a) and (c) is 0.0009, (4) a trend with a spike added to a normalized raw ECG data (only wavelet coefficients at levels 1–7 were preserved), (a) the normalized raw ECG signal, (b) semi-synthetic ECG signal formed by superimposing the normalized row ECG signal and a simulated trend, (c) the non-normalized processed semi-synthetic data with removed trend after applying a WT, (d) normalized processed semi-synthetic data with removed trend after applying a WT