| Diabetes,as one of the major diseases threatening human health in this century,still has no effective cure and only can be controlled by drugs or other medical means.Therefore,continuous monitoring and control of a patient’s blood glucose level are important means in the treatment of diabetes.Photoacoustic non-invasive blood glucose detection can avoid the damage and infection risk of frequent blood collection to patients,which is one of the current research hotspots.In this dissertation,the photoacoustic non-invasive blood glucose detection was studied and analyzed from the aspects of hardware design and software system,and a photoacoustic signal acquisition system was designed.At the same time,the timefrequency features,waveform features of the photoacoustic blood glucose signal and multi-feature fusion had been mainly studied.The main research contents are as follows:(1)A photoacoustic blood-glucose signal-acquisition system is designed and built.A corresponding data set is built to provide data support for subsequent algorithm testing and verification.This system used a single-wavelength pulsed laser with a wavelength of1064 nm as the excitation source to excite the photoacoustic signal of the target and used a focused ultrasound probe to detect the generated photoacoustic signal.In addition,for the designed signal acquisition system,the parameters such as peak power density,thermal relaxation time and pressure relaxation time are used to theoretically analyze and verify,and a set of methods for theoretical verification of the system are given.In the actual experiment,the rationality of the system was verified,and a sufficient amount of photoacoustic data was obtained.(2)The basic theories of photoacoustic signal detection,classification and recognition are studied.This part mainly studies two aspects: 1)the model of photoacoustic signal generation and transmission process;2)the basic theories of photoacoustic signal classification and recognition.The generation and transmission of photoacoustic signals can be modelled as three equations,namely the radiation transfer equation,the photothermal conversion equation and the wave equation,but these studies have a little discussion on the transmission medium.Aiming at the characteristics of human skin in photoacoustic blood-glucose detection,a multi-layer skin model and a blood model were proposed to describe the properties of the conductive medium during the signal transmission process,and to establish the numerical correlation between the blood glucose concentration and the mediums’ physical properties.Finally,these models were utilized for the numerical simulation of photoacoustic signals.In this model,the blood glucose concentration mainly affected the light absorption coefficient.Moreover,the greater the blood glucose concentration,the greater the light absorption coefficient.In addition,the basic theories of classification and recognition of various types of photoacoustic signals were investigated,including classification methods based on evidence theory.(3)A blood glucose detection method based on time-frequency features is proposed.Since the photoacoustic blood glucose signal is a typical non-stationary signal,timefrequency analysis and corresponding attribute analysis are important techniques for analyzing non-stationary signals.The studies in this part are divided into two parts: 1)the influence of different time-frequency analysis methods on the time-frequency spectrum of the photoacoustic blood glucose signal is studied.It was found that the time-frequency spectrum of the photoacoustic blood glucose signal based on the S transform did not contain cross-term interference,and has good time resolution and frequency resolution.The differences in the time-frequency spectrum caused by the different blood glucose concentrations could be clearly seen;2)In-depth analysis of the linear relationship between different time-frequency attributes and blood glucose concentration is made.And it is found that the Teager-Kaiser main energy attribute has the ability to describe the blood glucose concentration.And it had strong prediction performance in a low signal-to-noise ratio.Combining the above two parts,a blood glucose detection method based on Teager-Kaiser main energy is proposed,which can predict the blood glucose concentration of photoacoustic signals of unknown concentration,and achieve a good prediction performance.(4)A blood glucose detection method based on waveform features is proposed.The signal features commonly used in blood glucose detection mainly focus on the basic characteristics of the signal,such as amplitude,peak-to-peak value,mean value,variance value and other energy features.These basic features could often achieve good prediction results when the experimental conditions are excellent,but they often failed in poor experimental conditions with large interference or measurement errors.Therefore,in response to this problem,this dissertation started with the vibration mode of the photoacoustic signal and deeply studied the forced vibration equation and its characteristics.A waveform feature is proposed to describe the amplitude decay rate of the photoacoustic signal and is applied in the prediction of blood glucose concentration.I modelled the overall shape of the photoacoustic signal using the forced vibration equation and took the appropriate model parameters as the feature to describe the amplitude attenuation characteristics of the photoacoustic signal,which provides a new feature description for the modelling of the photoacoustic signal.In addition,a loss function was constructed with the signal envelope,and it was minimized by gradient descent to obtain the most suitable model parameters as waveform features.(5)A blood glucose detection method based on evidence fusion and self-similar focal element structure is proposed.In previous studies,multiple signal features of photoacoustic signals were often fused in parallel,but this construction method will gradually increase the degree of nonlinearity when the feature dimension gradually increases,making it difficult to handle.Therefore,this dissertation combining the evidence theory transforms the regression problem of blood-glucose concentration prediction into a classification problem by means of quantitative segmentation to deal with the problem of blood-glucose concentration prediction.This section improved the traditional focal element structure to make it easy to handle the features from many different sources.At the same time,the self-similar focal element construction made the evidence have a certain degree of the fractal feature that is self-similarity.This method predicted blood glucose concentration from the perspective of classifier design,which provides a new research idea for subsequent research.According to the generation,transmission and signal characteristics of photoacoustic blood glucose signals,a photoacoustic signal acquisition system and three methods were proposed for key technologies of blood glucose detection.Blood glucose detection solved the deficiencies in the existing research work to a certain extent,improved the performance of the photoacoustic non-invasive blood glucose detection system,and enhanced the detection performance of the system.In addition,the research on photoacoustic non-invasive blood glucose detection in this dissertation contributed to the clinical application of the system. |
PDF Full Text Request