The Method Of FTIR/ATR Spectroscopy Analysis For Thalassemia Screening Indicators

Thalassemia is a single-gene genetic disease, which is extremely dangerous but quite popular in south China region including Guangxi and Guangdong provinces. Currently there is no effective treatment. The main method of the disease prevention is to build up a massive thalassemia screening. The major indicators for the screening include HGB, MCV and MCH, which presently require different chemical agents for the chemical reactions, therefore is both time-consuming and costly, which impedes the massive screening. In this project, Fourier transform infrared (FTIR) spectrometer and attenuated total reflection (ATR) techniques were applied, in order to build up a direct, accurate and rapid quantification method of analyzing those indicators without any chemical reagent.We firstly tested the FTIR/ATR spectrum of HGB, MCV and MCH for the whole blood sample group,2 times,3 times,4 times,5 times and 6 times diluted hemolytic solution sample groups. The FTIP/ATR spectrum was analyzed to check its feasibility in detecting and quantifying those indicators. Then based upon the 2times diluted hemolytic solution, we started the major experiment on applying FTIR/ATR for screening the thalassemia indicators. In the major experiment, multiple models were compared, which included single-point regression model, direct PLS model, PLS model based on Savitzky-Golary smoothing, equivalent combination moving window MLR(ECMWMLR) model based on discrete wavenumbers and continuous moving window PLS(MWPLS) model. We compared those spectrum analysis models, optimized the models and tested their stability, then retested the models after all the procedures.The major conclusion of the thesis includes the following:1) Since both the predictive effect and the parameters of the model will vary with the different segmentation of objective and predictive group, we applied the optimized stable model based on the average results from 50 times segmentation to get the most stable results. We set up corresponding platform to calculate the average and standard deviation of the corresponding root mean squareed error of prediction (RMSEP) and correlation coefficient of prediction (RP) for each of the 50 segementations (as RMSEPAve,RMSEPStd,RP,Ave,RP,Std), then choose the most reliable model based on RMSEPAve.2) Based on 540 various Savitzky-Golary smoothing models, we set up a SG smoothing calculating platform to calculate the smoothing parameter for each of the model and built up the database. The large scale screening based on our SG smoothing model produced a better smoothing results than direct PLS model, with a higher number of smoothing points than traditonal parameter table, indicating that it's quite necessary to extend the smoothing point numbers as we do in the thesis.3) Based on different starting points, the continuous characteristic spectrum with high signal-to-noise ratio was chosen by varying the combination of number of adopted wavenumbers and factors. This had be done by setting up correspongind MWPLS calculation platform. The width of the chosen optimal spectrum window is narrower than the full scope spectrum. Thus this method greatly decreases the complexity of the model and effectively increases the prediction effect, which could replace the full scope spectrum as a more stable and simpler model. The optimal MWPLS model for HGB has a starting point of 1178 cm-1, number of adopted wavenumbers of 12, factor of 2, with its RMSEPAve and RP,Ave to be 5.334 g/L and 0.968 respectively. The optimal MWPLS model for MCV has a starting point of 3098 cm-1, number of adopted wavenumbers of 310, factor of 5, with its RMSEPAve and Rp,Ave to be 4.299 fl and 0.869 respectively. Last, the optimal MWPLS model for MCH has a starting point of 1549 cm-1, number of adopted wavenumbers of 9, factor of 6, with its RMSEPavc and RP,Ave to be 1.794 pg and 0.910 respectively.4) We proposed the ECMWMLR (Equivalent Combination Moving Window Multiple Linear Regression) model based on the moving window and multiple linear regression method. The corresponding calculating platform for a large scale screening was set up, which enabled us to build a highly accurate discrete prediction model with a fewer discreate wavenumbers instead of the full spectrum, thus decreses the complexity of the model. The optimal ECMWMLR model for HGB has a starting point of 1952 cm-1, a gap of 68 and point number 4, with its RMSEPAve and Rp,Ave to be 5.036 g/L and 0.972 respectively. The optimal ECMWMLR model for MCV has a starting point of 3964 cm-1, a gap of 109 and point number 13, with its RMSEPAve and RP,Ave to be 3.929 fl and 0.892 respectively. Last, the optimal ECMWMLR model for MCH has a starting point of 3568 cm-1, a gap of 69 and point number 8, with its RMSEPAve and Rp,Ave to be 1.662 pg and 0.919 respectively.The thesis firstly set up the FTIR/ATR spectrum method for screening the thalassemia indicators and quantitively analyzing. We further optimized the model and analyzed its stability and reported that this method has an advantage of affording proper accuracy directly and rapidly. The discrete and continuous model we set up in the thesis provides a stable quantifying model for screening the thalassemia indicators, which can act as the theory basis for designing the spectrum system with high signal-to-noise ratio for a mini-infrared spectroscopy with both discrete and continuous analyzing abilities. Thus the model we set up and described in the thesis has important value in both research and practice.