Characteristic Analysis And Modeling Study On Human Serum Spectrum

Along with the rising of the standard of living, people’s diet structure, livinghabit and environmental factors have very big changes, which makes the diseases,such as high blood sugar, high cholesterol and high blood fat metabolism syndromeand its complications, are gradually increasing. Therefore, it is very important to fastand accurately diagnose the physiological indexes of blood. This is one of the veryurgently topics needing to be solved in life science. Blood can characterize manyimportant life indexes. Therefore, the detection and research of the spectroscopy ofblood can test and analyze the blood physiological indexes, even might build newblood physiological indexes basing on the spectrum of blood. Obtaining informationfrom the blood spectrum of the complicated composition and putting it intoapplication in clinical practice are challenging research subjects. Compared with thetraditional method, spectral analysis technology is a method with speed and nocontact, no damage and other advantages. This will be an important medical diagnosisdevelopment direction.At present, though a lot of progress have been made to analyze the spectrum ofblood in the qualitative and quantitative aspects, there are still many questionsunresolved. For instance, in the blood qualitative analysis, the basic start point of mostresearches is to control the single factor, i.e., to research the reflection of spectruminformation of some a composition concentration, controlling other factors in a certainrange. The conclusions do not embody the generality, and it is also hard for somequalitative conclusion to be applied in clinical directly. The quantitative regressionresearch of the spectrum for the blood is still in the initial stage, and the researchmainly focus on the infrared spectrum of the blood. Although some fast detectioninstruments of the blood composition have been developed, there are low correlationbetween the test results and automatic biochemistry analyzer testing results, therefore,the instrument precision should to be further improved.With the background mentioned above, we study systematically the humanserum UV-vis absorption spectroscopy and fluorescence spectrum basing on the specific conditions of the laboratory. The main research content and innovative workare as follows:1. We have proposed the decomposition method using partial least squares basedon the ultraviolet-visible absorption spectrum, acquiring the absorption coefficient ofblood sugar, blood lipids, cholesterol and the absorbance of other components inserum. Results show that, the absorption coefficient of the blood fat is larger than zeroin the whole waveb and and appear large values in the200-300nm band, indicatingthat blood fat has strong absorption effect in UV-light band of200-300nm. Theabsorption coefficient of blood sugar in the200-300nm band presents positive andnegative alternate distribution. The absorption coefficient of cholesterol is positivenear414nm and negative in200-300nm. The absorption coefficient of blood fatnears414nm is almost zero, while that of blood sugar and cholesterol present a smallpeak. The absorption effect of other components is significantly greater than bloodsugar, blood fat and cholesterol near414nm.2. We have proposed the method of multi-scale wavelet composing andderivation to preprocess the serum spectra basing on the complexity of the spectralcomposition of blood, and the multiple correlations of various serum spectra arestudied. Results shows that serum spectra have serious multiple correlation. Thecorrelation coefficient of absorption spectra is as high as0.8. The correlationcoefficients of the original fluorescence spectra and filtering fluorescence spectra arehigher than of0.96and0.98, respectively. The noise，derivation and waveletdecomposition of spectra has effect on the inhibitory multiple correlation.3. We have research the correlations of serum spectra and blood sugar, blood fatand cholesterol concentration. Results show that there is a greater correlationcoefficient between blood fat concentration and absorption spectra in200-300nmrange. In the200-800nm range, the derivatives spectroscopy of three compositionsand the third and fourth layer spectrum presents some wavelength point with biggercorrelation coefficient. In the300-700nm range, blood fat concentration and bloodsugar respectively show a higher positive and negative correlation, and cholesterolreveals a very weak negative correlation.4. The ADF software based on the density functional is used to calculate ultraviolet absorption spectrum and fluorescence spectrum of the blood sugar. Thecalculation results show that, there are absorption peaks of blood sugar molecules at185nm,199nm,216nm,221nm,242nm,246nm,290nm and there arefluorescence peak at295nm,315nm,342nm,392nm,427nm,502nm,730nm,1058nm. The blood sugar fluorescence peak appears at730nm, in good agreementwith the experimental results.5. Based on the window-genetic algorithm, the modeling software supported bythe nonlinear partial least squares method, BP artificial neural network and leastsquares support vector machine are designed, and spectrum prediction models ofblood sugar, blood fat and cholesterol concentration are established. Research showsthat genetic algorithm can suppress noise and improve the accuracy of the model.Overall, fluorescence spectrum model is better than the absorption spectrum, themodeling effect of blood fat is more prominent. The high accuracy prediction modelsof blood glucose concentration are established based on the PLS method, using220-240nm blood sugar absorption spectrum and near730nm blood sugarfluorescence peak nearby band. The high accuracy prediction models of cholesterolconcentration is established based on the BP neural network, using near560nmcholesterol fluorescence peak nearby band, which indicates that the window methodcombined with genetic algorithm to select the best modeling wave band is feasible,the prediction model established for blood sugar, blood fat and cholesterol is reliable.These models are expected to provide a theoretical support to develop the fast,convenient and high precise serum ultraviolet-visible absorption and fluorescencespectrum instrument, and these models have important application value.