Blood glucose concentration estimation by Raman spectroscopy based on particle swarm optimized SVR

Blood glucose level has important significance for medical diagnosis. Blood glucose measurement in traditional methods requires collecting blood samples several times a day, which causes discomfort, environmental pollution and so on. As a "fingerprint" spectrum for molecular recognition, Raman spectroscopy has attracted attention in blood glucose measurement. However, blood glucose level is low and spectral signal of glucose is easy to be influenced by noise and other components. To improve accuracy of blood glucose concentration estimation by Raman spectroscopy, we carried out the Raman blood glucose measurement in vitro, the interferograms of blood samples in different glucose concentrations were measured by the self-developed Spatial Heterodyne Raman Spectrometer (SHRS), and converted the interferograms to one-dimensional spectroscopic data using Fourier transform. In order to get data with higher quality, we used wavelet decomposition to remove the noise and sparse representation to remove the signal baseline. Then, selected the spectroscopy at 500-2500 cm^-1 as input, and the corresponding blood glucose concentration value as label, use particle swarm optimization-support vector regression (PSO-SVR) algorithm to construct the blood glucose concentration estimation model. The results show that the R² of test set is 0.8041 and the RMSE is 1.8580. And the accuracy of blood glucose concentration estimation was evaluated by the Clark Error Grid. The model based on PSO-SVR can achieve accurate estimation of blood glucose concentration. This method has important research significance and application potential for blood glucose measurement.