PPG Signal Denoising using a New Method for the Selection of Optimal Wavelet Transform Parameters

Photoplethysmogram (PPG) is a vital biological signal which provides valuable information related to our cardiovascular system. However, noise artifacts like high frequency noise and movement noise can corrupt the signal during recording. The success of the widely used Wavelet Shrinkage denoising method largely depends on the optimal selection of its control parameters. In this paper, we propose a new method for the selection of the Decomposition Level and Noise Threshold in PPG signal wavelet shrinkage denoising. The Crest Factor (CF) of the detail coefficient is used as a metric for the selection of decomposition level and to determine that the detail coefficient comprises noise component, noisy component or signal component. The noise components are removed and the noisy components are delimited using a level-dependent threshold, which do not demand noise estimation. Finally, the denoised signal is reconstructed from the delimited noisy components, signal components and approximate component by Inverse Wavelet Transform (IWT). Experiments are carried out using simulated PPG signals with MIT-BIH noise stress test database and real PPG signals from PPG-BP Database and Wrist PPG During Exercise Database. The denoising performance is compared with the state-of-art-methods in terms of improved Signal to Noise Ratio (SNR), Root Mean Square Error (RMSE) and Percent Root Mean Square Difference (PRD). Results shows that the proposed method yields an improved SNR of 15 dB and 37 dB at input SNR of 2 and 20 dB respectively, whereas the best of the standard methods increases the SNR by 12 dB and 28 dB for the same input SNR. Thus, the proposed method has promising research value and can invariably be applied to other biomedical nonstationary signals like ECE and EEG.