Considering blood scattering effect in noninvasive optical detection of blood components using dynamic spectrum along with time varying filter based empirical mode decomposition

Absorbance in photoplethysmography (PPG) depend linearly on concentration however blood scattering can induce the absorbance and transmittance resulting in nonlinear errors. Based on an adaptive spectroscopy technique called dynamic spectrum (DS), a multi-position multi-spectral PPG processing method potentially for detecting pulsating arterial blood from upper and lower parts of the PPG signal was proposed to probe blood scattering effects. The proposed method employs the time varying filter-based empirical mode decomposition (TVF-EMD) algorithm for DS extraction from accumulated narrowband signals for linear DS components extraction. The extraction in this paper uses the transmittance and reflectance modalities of the PPG signal from 600 nm and 1100 nm wavelengths. The hemoglobin concentration samples from 220 volunteer subjects were collected for modeling. A reduced range of absorbance values was obtained from the combined-DS extraction which is significant for high transmittance and high precision against nonlinear errors due to blood scattering. Quantitative modeling with partial least squares (PLS) and radial basis function neural network (RBF-NN) was used for further evaluation. Low and under-fitting training results were obtained from the PLS modeling of the original dual-position DS, confirming the nonlinear effect of scattering. Conversely, modeling results from the combined-DS (TVF-EMD dual-position DS) with the nonlinear RBF-NN model improved significantly indicating the potential reduction of blood scattering effects. Furthermore, significant noise reduction was obtained with the TVF-EMD extraction method. The results suggest that DS from dual-position with adaptive filtering and nonlinear modeling can be a promising approach to eliminate blood scattering errors in noninvasive optical detection of blood components.