Research On Techniques Of Human Blood Composition Analysis Based On Near-infrared Spectroscopy

The composition and state of blood is an important indication of human physiological state,which has important guiding significance for the diagnosis of various human diseases.Spectral detection and analysis technology has the advantages of convenient,rapid,non-destructive and high accuracy,which has a well application prospect in blood non-destructive determination.But the nature of blood itself brings a lot of interference to spectral analysis,such as serious spectral overlap,low signal-to-noise ratio and so on.Aiming at the defects and many unstable factors of near-infrared spectroscopy,this paper studies the selection and division of samples,data processing and the construction of detection and prediction model for blood component analysis by near-infrared spectroscopyThe main research work is as follows:(1)The feasibility of near infrared spectroscopy for quantitative hemoglobin concentration determination was studied.Near-infrared transmission spectrum was used to analyze and predict the hemoglobin concentration of whole blood samples The established calibration model has achieved good results in predicting hemoglobin concentration.Partner wavelength selection(PWM)method uses only a few wavelengths to establish the model,which can achieve or even slightly exceed the prediction effect of full-spectrum model with only a few wavelengths,and improve the efficiency of near-infrared spectrum detection(2)Tracing the root of the accuracy of spectral model,SPXY method was improved,and a weighted SPXY(WSPXY)method was proposed,which can set the weight distribution of the distance between spectral data and the concentration of the component to be measured.WSPXY method was compared with other calibration sample selection methods in modeling and prediction of hemoglobin concentration in experiments.It proved that model with samples selected by WSPXY method can achieve better prediction effect than other calibration sample selection methods,and further improve the accuracy of hemoglobin concentration detection model with near-infrared spectroscopy(3)Based on the preceding research of hemoglobin concentration by near infrared spectroscopy,we further testify the influence of the ratio of the sample number of prediction set to the sample number of calibration set(W)on the modeling with experimental methods.In the processing of choosing samples for calibration set,the ratio should be larger than 0.4 to ensure the stability and reliability of the established model,and we achieved the ratio(W)being 0.4 as a gold ratio of sample division.(4)For spectral modeling method,the difference of the prediction results by models with linear and nonlinear methods for blood component detection was studied.The prediction performance of three methods was compared,including principal component analysis combined with multivariate linear regression(PCA+MLR)method,partial least square regression(PLSR)method and BP-ANN method.The shortcoming of these methods and complementary approachs were analyzed.Specificly,the prediction results of all three models can be more accurate and the quality of the models can be further promoted,when samples in calibration set were selected by WSPXY method.The best prediction model with near-infrared spectroscopy was achieved as combining WSPXY method with BP-ANN method.(5)Combined with the detection methods and theories of wavelength optimization,WSPXY method and BP-ANN method,the feasibility of visible and near-infrared spectroscopy for non-destructive detection of blood bag free hemoglobin concentration is further studied.The results show that near-infrared band(900-1700 Nm)has a good prediction effect for detection of free hemoglobin concentration in plasma,while the prediction effect in whole blood needs to be improved.Meanwhile,"M+N" theory is used to study the influence of the red blood cell count and hemoglobin concentration in whole blood on the nondestructive free hemoglobin concentration detection in blood bags.According to "M+N" theory,the accuracy origin of blood concentration with spectral methods is analysed,and the relationship between the noise and non-target components in blood and the accuracy of target components is established.The results prove that the red blood cell count and hemoglobin concentration in whole blood have an positive effect on the prediction od free hemoglobin concentration.The reseach can promote the application of achievements in blood composition detection with near infrared spectroscopy.In this paper,many subjects were intergrated to the research,including spectroscopy,intelligent control,computer science and so on.The research results can effectively improve detection capability and analytical skills of blood components with near infrared spectroscopy.It is conducive to the clinical application of spectroscopic blood component analysis,and has important theoretical and application valueThis paper includes 49 pictures,17 tables and 158 references.