| The main characteristic of diabetes is that the blood glucose concentration remains continuously elevated,and it is a chronic metabolic disease that cannot be cured by current clinical medical technology.The magnitude of blood glucose concentration is an important indicator of diabetes.Patients with diabetes require frequent measurement of blood glucose levels and,if necessary,insulin injections to maintain blood glucose stability.At present,clinical blood glucose testing is measured by intravenous blood sampling.Although this measurement method is more accurate,it is very painful for diabetic patients to have frequent and repeated blood collection.Most of the minimally invasive blood glucose measurement products on the market today use probes or laser pins to pierce the fingertips or other parts of the body to obtain tissue fluid for blood glucose level measurement.Compared with traditional measurement methods,this kind of testing method can reduce the pain of blood collection from veins,but the detection of blood glucose has a lag,and it is easy to cause infection in the wound.Continuous Glucose Monitoring(CGM)can achieve long time dynamic glucose testing and facilitate patients’ self-testing.However,CGM is very expensive,the equipment needs to be calibrated from time to time,and the probe needs to be replaced frequently.Moreover,for patients who are allergic,the material of implantable probe may cause skin allergy.Therefore,non-invasive blood glucose testing technology,which is non-invasive,can be continuously measured and can observe changes in blood glucose levels,has become a popular research topic in the international academic community.In the past decade or so,research teams in related fields have been striving to explore the methods of non-invasive blood glucose measurement,and so far,the international research results have been increasing,and many cutting-edge innovative technologies have started to be applied to non-invasive blood glucose testing.However,the technology of non-invasive blood glucose testing is not yet mature,and the technical difficulty involves knowledge in various fields,including data acquisition,data processing and calculation methods.Due to the limitation of technical difficulty,it is very rare to be able to perform high precision noninvasive blood glucose testing technology and complete productization so far.Thus,this study uses big data thinking to model multiple features extracted from the electrocardiogram(ECG)signal,proposes a portable noninvasive blood glucose detection system based on deep learning of ECG information,and designs a noninvasive blood glucose detection system with both long duration and high accuracy.The model prediction results were compared and statistically analyzed,and the best model for noninvasive blood glucose estimation was convolutional neural network combined with long and short time memory network(CNN+LSTM).The research of this paper combines the knowledge of human physiological weak signal detection and processing,embedded software and hardware design,machine learning and deep learning,etc.,and mainly accomplishes the following work.1.Made a detailed research on blood glucose level measurement methods and compared the advantages and disadvantages;made a systematic overview of the research status of non-invasive blood glucose detection technology at home and abroad;and analyzed the market demand of non-invasive blood glucose detection devices.2.Improve the hardware circuit design and realize the hardware circuit platform where ECG signal and pulse wave can be measured simultaneously,which includes pulse wave and ECG signal synchronous acquisition circuit,charging and voltage regulator circuit,STM32 control circuit,and give a detailed introduction of the design ideas and circuit principles,and also give the design diagram of this paper.3.The ECG signal filtering algorithm is improved,focusing on the detection methods of ECG signal features such as P-wave descent,QRS wave groups,and T-wave.The use of several machine learning algorithms and the design rationale are elaborated.4.Machine learning and deep learning modeling using ECG signal features,after several iterations of tuning experiments,and finally analysis and comparison to get the optimal model for blood glucose level prediction,the highest accuracy of CNN+LSTM model test reached 90.13%.5.To realize the prediction of blood glucose concentration by machine learning and deep learning models,analyze and compare the prediction results of the models,and prove that the machine learning and deep learning algorithms based on the multi-feature information fusion of ECG signal can get a better prediction effect of blood glucose level.It provides a powerful algorithm reserve for a non-invasive blood glucose testing device with both portability and accuracy. |
PDF Full Text Request