Study On Key Technology Of Glucose Concentration Prediction In Serum Solution Based On Continuous Photoacoustic Spectroscopy

With the increasing number of patients with diabetes worldwide,conventional blood glucose detection methods will bring pain.Non invasive,portable,economical and dynamic blood glucose monitoring equipment has become an urgent need for diabetes patients.The photoacoustic method has its unique advantages,not only has no strong scattering and attenuation problems compared with traditional optical methods,but also has the potential to achieve non-invasive blood glucose detection with high sensitivity.At present,the non-invasive blood glucose photoacoustic scheme with good effect on human body uses the mid-infrared quantum cascade laser,but because of its expensive price and huge system volume,it has not gone out of the laboratory.Near infrared continuous laser is cheap and is expected to be a good substitute.In this paper,the glucose concentration prediction of glucose solution and serum solution has been completed using a near-infrared continuous tunable laser.In this paper,a set of near-infrared continuous photoacoustic spectrum signal detection system was built,a photoacoustic cell was designed to meet the future blood glucose detection scheme,and relevant software was developed for the control system,which can collect photoacoustic signals with high signal-to-noise ratio.Then the relationship between photoacoustic signal and laser power,laser modulation frequency,temperature and laser wavelength was studied.Through the standardized photoacoustic spectrum of high concentration glucose solution and deionized water,it is clear that the area around 1571 nm in the 1500-1630 nm band is the most sensitive to glucose.Subsequently,the system parameters for the subsequent collection of data sets were determined.The modulation frequency was 437 k Hz at room temperature.The laser wavelength was selected from 1540 to 1600 nm,which was sensitive to glucose.Six wavelengths with high power were selected,and the relevant software for real-time glucose concentration prediction was completed.The glucose solution data set of human glucose concentration range was constructed in the experiment,and seven machine learning models were used for training and prediction.The best effect was the Gaussian process regression model of quadratic rational kernel function.The experimental results show that the near-infrared photoacoustic multispectral diagnostic system can be used to predict glucose levels.The Clarke error grid area A accounts for 95.99% in the data set’s 5-fold cross-validation,and the area A accounts for 92.38% in the real-time glucose concentration prediction.Subsequently,the glucose concentration in serum solution was predicted using the glucose solution training model.With the increase of serum glucose content,the prediction results of the model also showed a high linear relationship,indicating that the photoacoustic method is sensitive to the detection of glucose concentration changes.At last,the photoacoustic signal acquisition of human body has been preliminarily carried out,and high-quality photoacoustic signals can be collected from the earlobe.The research results in this paper are not only possible to better develop the photoacoustic blood glucose meter,but also possible to apply it to the detection of other components.