Photoacoustic qualitative classification of blood glucose with multiple factors based on BP neural network

Abstract

In this work, the photoacoustic detection of blood glucose with the interference of multiple factors was studied. A set of photoacoustic detection system of blood glucose was established, in which the interference of multiple factors including the laser energy, concentration, temperature, flow velocity and detection distance were combined into. Under different conditions of multiple factors, the time-resolved photoacoustic signals and peak-to-peak values of blood samples were all obtained. To accurately classify the concentration of blood glucose samples, back propagation (BP) neural network was employed to train the photoacoustic peak-to-peak values and the multiple factors. In BP neural network, five different Arabic numerals from 1 to 5 were labeled to denote five kinds of blood glucose levels ranged from2mmol/Lto14mmol/L. The photoacoustic peak-to-peak values, laser energy, temperature, flow velocity and detection distance were used as the input data, the labels denoted different concentrations were used as the output data. Meanwhile, the effects of neurons number in hidden layer and learning factor on the classification accuracy of blood glucose level were investigated. Under the optimal parameters of BP neural network, the accuracy of classifying concentration of blood glucose level reached 85.6% for the test blood glucose samples. Compared with the classification accuracy (71.2%) of blood glucose level based on support vector machine (SVM) algorithm, it is demonstrated that the photoacoustic spectroscopy combined with BP neural network has a good performance in qualitative classification of blood glucose under the interference of multiple factors.