Fingertip Hyperlycemia Pulse Signal Analysis Based On Non-Invasive Detection Methods

As a unique medical theoretical system of the Chinese nation,traditional Chinese medicine has a relatively long history.Pulse diagnosis has always been a very important component of traditional Chinese medicine due to its convenient and accurate detection methods.However,compared to other diagnostic methods,the accuracy of pulse diagnosis results depends heavily on the relevant experience of practitioners,which poses a significant obstacle to the promotion of pulse diagnosis in clinical medicine.Therefore,how to introduce modern technology into traditional pulse diagnosis,so that pulse can adapt to the development of the times,has become a very important research direction in modern medicine.With the development of society and the progress of science and technology,more and more people are suffering from high blood pressure,hyperlipidemia,and diabetes related diseases of wealth and honour.Currently,there are three main methods for measuring blood glucose concentration on the market:invasive detection,minimally invasive detection,and non-invasive detection.Although the first two methods have high accuracy,due to the need to collect blood from patients,it is inevitable that they will bring some physiological and psychological pain to patients during the testing process.Therefore,proposing a noninvasive detection method has become a very important research direction in blood glucose research.This study proposes a method for detecting whether the human body belongs to a hyperglycemic population based on fingertip pulse signals,with fasting blood glucose greater than 6.1mmol/l and postprandial blood glucose greater than 7.8mmol/l as the criteria for hyperglycemia.This paper proposes a non invasive blood glucose detection method based on deep learning.Through preprocessing the collected pulse signal,mainly including wavelet denoising,reasonable period segmentation,removal of abnormal values,removal of baseline drift,and quality detection of the pulse signal,and other processes,and then through wavelet transform to extract the characteristics of the pulse signal and ShuffleNet neural network training,we can discover the information of hyperglycemia in the pulse.Compared to the mainstream blood glucose detection methods on the market,this high blood glucose concentration detection method has a lower detection cost and does not cause physical pain to the tester.While possessing a high accuracy rate for detecting hyperglycemia.Can be reused.There are two sources of pulse data for hyperglycemia used in this topic.One is to collect pulse signals from people whose blood sugar meets the hyperglycemia standard by participating in free blood sugar testing activities regularly held by offline pharmacies.After more than one month of collection,a total of over 40 pulse signals of hyperglycemia from different individuals were collected.Another source of data is the regular detection of pulse signals from relatives with high blood sugar in the home.In this sampling,a total of more than 60 pulse signals were collected.A total of 102 pulse signals of hyperglycemia and the same number of pulse signals of normal blood glucose concentration were obtained through statistics.After obtaining enough hyperglycemic pulse data,this article conducted a total of three experiments.The first experiment was to distinguish between people with high blood sugar.In this experiment,75 pulse data of hyperglycemia and 75 pulse data of people with normal blood sugar were used as training sets,and other data were used as test sets.Through calculation,we can know that the accurate calculation of hyperglycemia is 85.2%.The second experiment is the pulse wave classification experiment of diabetic patients.The fingertip pulse of 36 diabetics and 34 people with normal blood sugar were used as the training set,and 9 diabetics and 9 people with normal blood sugar were used as the test set.Through calculation,the prediction accuracy of this experiment is 83.3%.The third experiment was to choose a better network model for identifying hyperglycemic people by comparing different convolutional neural networks.Therefore,the control group in this experiment used a more classical network model,Resnet network model.Through the analysis of the recognition results of the experimental group and the control group,it can be seen that ShuffleNet network has a higher recognition accuracy rate on the premise of satisfying similar computational complexity.This experiment shows that ShuffleNet network has a better advantage over Resnet network in the experiment of measuring the classification of people with high blood sugar.