Research On Component Measurement Of Blood And Urine Based On Multiple Optical Path Length Spectroscopy

Spectral analysis method, with sensitive, rapid and accurate features, is widely used in various fields. The concentration level of blood and urine components is the main judgment basis of human health condition. Low-cost and rapid detection of blood and urine components has important academic significance and social value. In theory, the multiple optical path length spectra, with the nonlinear feature, can provide more information of the sample than single optical path length spectra and can improve the accuracy of measurement, using the abundant information. Based on this theory, the preliminary research and exploration on the measurement of blood and urine components was carried out in this work.To obtain the multiple optical path length spectra data, the automatic detection system was established. The system consists of an automatic micro-displacement device, two spectrometers and a PC. By the cooperative work of the software designed based on VB.net language and the micro-controller, the minute optical length can be changed accurately and stably and the data obtaining processing can be performed automatically.In the measuring experiment of blood components, near-infrared multiple optical path length spectra of 200 serum samples were collected. The models of glucose, total cholesterol, total protein and albumin were established. The r2 of prediction are 0.8686, 0.9432, 0.8953 and 0.8993, and the RMSEP are 0.6792 mmol/L, 0.2064 mmol/L, 2.8469 g/L and 1.7942 g/L, respectively. The result of the preliminary study on the feature extraction shows that picking up the special wavelengths from group of optical path length has the best performance. The dimension of spectral data decreases to 1/40 of the original data. This method has the positive significance to the modeling of multiple optical path length spectra.In the measuring experiment of urine components, UMALB, the sensitive indicator of renal injury, was selected as the measuring object. 27 standard samples of urine and 207 human urine samples were acquired. The modeling results of standard samples show that the model established by both visible and near-infrared spectral information is better than that using a single band. Its r2 is up to 0.9905, and the maximum absolute prediction error is 7.56mg/L. Considering the nonlinear factors of spectral data, the PLS-ANN modeling method was involved to increase the accuracy of quantitative analysis. The r2 of the human urine samples is 0.9511, and RMSEP is 5.02 mg/L.This detection system designed in this work reduced the probability of misoperation, and improved the efficiency and accuracy of experiments greatly. The prediction accuracy of glucose, total cholesterol, total protein, albumin and UMALB concentration is improved according to the preliminary study of data mining method, such as model establishment and feature extraction, which is basis of the follow-up blood and urine multiple optical path length spectra collection and analysis.