A Dynamic Spectrum extraction method for extracting blood scattering information - Dual-position extraction method

摘要

Dynamic Spectrum (DS) can effectively suppress the influences of individual differences (such as skin, muscle, fat) and the variations of measurement conditions, which shows its potential in the clinical applications. To suppress the non-linear effect of blood scattering on DS and improve the accuracy of noninvasive blood component analysis, a dual-position extraction method was proposed. By extracting DS from the upper and lower halves in a spectrum photoplethysmography (SPPG) (photoplethysmography signal at multiple wavelengths) signal separately, the DS of pulsatile partial of arterial blood contracting in two different positions can be extracted from one sample. Then, the DS containing gross errors is screened by statistical method, and the final DS output is obtained by superimposing the remainder DS averagely. To verify the effectiveness of the new method, single-trial extraction, optimized difference extraction and dual-position extraction method were used to extract DS from 231 volunteers' clinical experimental data respectively. Partial Least Squares (PLS) and Radial Basis Function (RBF) neural network were used to establish hemoglobin concentration prediction model. The results show that the modeling indicators of the dual-position extraction are lower than the other two extraction methods. However, combining the spectral data extracted by the dual-position extraction with the spectral data extracted by the single-trial extraction, the obtained modeling indicators are superior to other methods. To make full use of the advantages of the dual-position extraction method, the RBF neural network, a non-linear modeling method is used to model. The result means that, increasing spectral information at different positions can obtain more scattering information of blood. In the modeling, especially for the non-linear modeling method, dual-position extraction can better suppress the nonlinear effects of blood scattering and improve the prediction accuracy of non-inv