Study on a modified baseline correction method based on wavelet transform

A modified baseline correction method based on wavelet transform is investigated and applied to the human blood glucose non-invasive measurement. The selection method of optimum decomposition level using wavelet transform is researched, and a new method to determine the optimum decomposition level is proposed which is simple, fast, and it can effectively improve the result of baseline correction based on wavelet transform. The spectra of human finger skin and the corresponding blood glucose value are measured in human oral glucose tolerance test (OGTT) which can make the concentration of human blood glucose vary in a certain range in a short time. A regression model is established and evaluated by cross validation method. The correlation coefficient between prediction value and true value is 0.75, and the root mean square error of prediction is 1.36 mmol/L. The prediction accuracy of calibration model with this modified baseline correction method is greatly improved compared with the predictions of the original spectra and the corrected spectra at other decomposition level, the root mean square error approximately decreases by 39%, and the correlation coefficient increases by 0.64.