基于可见光图像的无创血糖测量仿体实验验证

Non-invasive blood glucose detection based on optical measurement is a research hotspot in the biomedical field at present. However, due to the problems of low signal-to-noise ratio, background noise interference and low accuracy, the non-invasive blood glucose detection method is still in the experimental stage and cannot be applied in clinical practice. To solve these problems, a non-invasive blood glucose detection method based on visible image is proposed. By using the collected scattering images and the gradient boosting decision tree algorithm, the regression model of the relationship between the characteristic parameters of the scattering images and the blood glucose concentration is established, and the accuracy of the model is verified by the phantom experiment. Experimental results show that the relationship between visible scattering images and glucose concentration can be modeled by the gradient boosting regression model, with a consistency determination coefficient up to 0.929 and an average absolute error of glucose detection accuracy of 0.156 g•L^-1.