Methods To Improve Signal-to-noise Ratio (SNR) Of Dynamic Spectrum And Its Application

Noninvasive measurement of blood components has great significance fordiagnosis of some chronic diseases (e.g. diabetes and anemia) and continuousmonitoring of patients in the emergency department or during the perioperative period.It is also an effective method to screen for diseases at early stages, and can savemedical and environmental resources. Research on noninvasive measurement ofhuman blood components can promote development in noninvasive biomedical signaldetection, weak signal detection, basic medical research, and clinical medicine.Spectroscopy, because of its characteristics of convenience, noninvasiveness,multi-dimensional information, high speed and high accuracy, has become the mostpromising method. In vivo spectral measurements based on safety and comfort areaffected by signal weakness, spectral overlapping, baseline drift, and some uncertaintyfactors such as individual differences, measurement conditions and psychologicalchange of test subjects. Therefore, the primary target of in vivo spectral measurementsis to improving signal-to-noise ratio (SNR). Dynamic spectrum (DS) based ontransmission photoplethysmography (PPG) waves extracts optical densities of arterialpulsatile blood by changes in absorption spectrum at several given wavelengths. TheDS method can reduce the influence of individual differences and measurementconditions in theory. Compared to other spectral methods, there are some remarkableadvantages.Firstly, the paper presents the basic principle of DS. Considering thecharacteristics of human tissue, the DS values with different hemoglobinconcentrations were calculated by Monte Carlo (MC) simulations, and the correlationcoefficient between DS values and hemoglobin concentrations was used to verify thefeasibility of the DS method. Moreover, the measurement system based on a gratingspectrometer and a halogen lamp was constructed to collect the clinical data anddevelop the calibration model by partial least squares (PLS) regression. Theexperimental results further validate the DS method. However, the measurementaccuracy needs to be improved, and the SNR in signal detection and processingshould be enhanced. The paper studies the topic mainly from the following fouraspects. The first aspect is to improve the SNR in the signal sensing system. For the dataacquisition system, the DS values under the conditions of three kinds of lightingmethods (narrow flat beam, broad flat beam and broad fiber optic beam) by MCsimulations were compared to analyze the influence of different optical paths due tolighting methods on DS. And the DSs of two light sources (a tungsten-halogen lampand supercontinuum sources) were extracted to verify the simulation result that thenarrow beam can enhance the sensitivity of PPG waves. For the data acquisitionmethod, an optimum method “double-sampling” was proposed. Compared with thetraditional method, the double-sampling method can enhance the dynamic range ofdetectors, significantly improve the SNR of DS, and widen the effective spectralrange.The second aspect is to improve the SNR of the extracted DS. Two existingmethods: frequency domain analysis and single edge analysis, were compared witheach other by theoretical analysis and experiments. The results demonstrate thefeasibility and consistency of two extraction methods, and the superiority of singleedge analysis on removing the abnormal segment of PPG waves. Then, combining theadvantages of both existing methods, a new method based on a fast digital lock-inamplifier (FDLIA) was developed, which cut PPG waves into several segments intime domain, and extracted the DS of each segment by FDLIA in real time. The newmethod greatly reduces the complex computation and increases the accuracy of DSs.Simulation and experimental results show the feasibility of the new method andevaluate its performance in the speed and accuracy.The third aspect is to improve the SNR in the signal processing and modeling.For scattering influence, simulations on three kinds of tissue models by the MCmethod were used to analyze its way and degree before correction. For spectralanalysis, a coding method was proposed to study of internal mechanism of spectralanalysis methods. To show its implementation process, spectra of the mixture withthree components were developed to construct calibration models. The resultsdemonstrate that the coding method can be used to qualitatively and quantitativelyanalyze the influence from absorbance of three components, is useful in wavelengthselection, and helps us grasp mechanism of modeling methods. Moreover, the paperanalyzed the influence of water absorption on hemoglobin measurement, and theexperimental results show that the spectra covering absorption peaks of non-detected components can improve the SNR of the calibration model, which opens up an avenuefor reducing the infuence of non-detected components by selecting wavelength range.The fourth aspect is to design a portable measurement system based on LED andDSP. The system adopts optical modulation technology, and synchronously collectseight-channel PPG waves with high accuracy by FDLIA together with oversampling.The design can simplifier the external hardware circuit and make the system flexible,portable and low-cost. The algorithm combining FDLIA and oversampling wasdiscussed in detail. Simulations and experiments were performed to verify thefeasibility of the algorithm and evaluate the performance of weak signal detection.Meanwhile, the algorithm was extended to the multi-channel signal detection, wherethe carrier frequency setting and adjacent channel crosstalk were taken intoconsideration. The parameters settings and implementation method were proposed forgeneral applications, which make the method more universal.Based on the feasibility of DS by noninvasive hemoglobin measurement, thepaper presents a series of measures and methods from four aspects including signaldetection, DS extraction, spectral processing and modeling, and the portable systemdesign, to improve the accuracy and sensitivity of spectra. All of these can help form acomplete system of methods for noninvasive blood component measurement of DS,deal with common problems of in vivo spectral measurement and weak signaldetection, and provide references for other similar applications.