Low glucose concentration estimation based on reaction with 4,4'-biphenyl boronic acid using deep learning

Abstract

Minimally invasive blood glucose level estimation with Raman spectroscopy is an important research field and attracts great attention. However, glucose concentration in blood is low and is difficult to be accurately measured. In this paper, we creatively proposed applying the 4,4'-biphenyl boronic acid to react with different concentrations of glucose to obtain the complex—(C36H40O18B4) n. We performed a regression of the Raman spectral data of (C36H40O18B4) n and the glucose solution separately to compare their estimation results. We applied a deep learning network, ResNet, and compared it with regression models of conventional machine learning, uniformly using ten-fold cross-validation. The experimental results show that the generated (C36H40O18B4) n can effectively improve the estimation performance of glucose. The results showed, the ResNet model does not require explicit feature extraction and can achieve fast and accurate estimation. Its performance is significantly better than the traditional linear analysis method, and the R square can reach 0.93. The method in the article can effectively improve the estimation effect of low-concentration glucose.