CN109171756A - Diabetes index prediction technique and its system based on depth confidence network model - Google Patents
Diabetes index prediction technique and its system based on depth confidence network model Download PDFInfo
- Publication number
- CN109171756A CN109171756A CN201811016543.7A CN201811016543A CN109171756A CN 109171756 A CN109171756 A CN 109171756A CN 201811016543 A CN201811016543 A CN 201811016543A CN 109171756 A CN109171756 A CN 109171756A
- Authority
- CN
- China
- Prior art keywords
- data
- network model
- depth confidence
- confidence network
- skin
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/145—Measuring characteristics of blood in vivo, e.g. gas concentration, pH value; Measuring characteristics of body fluids or tissues, e.g. interstitial fluid, cerebral tissue
- A61B5/14532—Measuring characteristics of blood in vivo, e.g. gas concentration, pH value; Measuring characteristics of body fluids or tissues, e.g. interstitial fluid, cerebral tissue for measuring glucose, e.g. by tissue impedance measurement
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/01—Measuring temperature of body parts ; Diagnostic temperature sensing, e.g. for malignant or inflamed tissue
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/02—Detecting, measuring or recording pulse, heart rate, blood pressure or blood flow; Combined pulse/heart-rate/blood pressure determination; Evaluating a cardiovascular condition not otherwise provided for, e.g. using combinations of techniques provided for in this group with electrocardiography or electroauscultation; Heart catheters for measuring blood pressure
- A61B5/0205—Simultaneously evaluating both cardiovascular conditions and different types of body conditions, e.g. heart and respiratory condition
- A61B5/02055—Simultaneously evaluating both cardiovascular condition and temperature
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/103—Detecting, measuring or recording devices for testing the shape, pattern, colour, size or movement of the body or parts thereof, for diagnostic purposes
- A61B5/11—Measuring movement of the entire body or parts thereof, e.g. head or hand tremor, mobility of a limb
- A61B5/1116—Determining posture transitions
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/44—Detecting, measuring or recording for evaluating the integumentary system, e.g. skin, hair or nails
- A61B5/441—Skin evaluation, e.g. for skin disorder diagnosis
- A61B5/443—Evaluating skin constituents, e.g. elastin, melanin, water
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
Abstract
The invention belongs to intelligent monitor equipment technical fields, more particularly to diabetes index prediction technique and its system based on depth confidence network model.The present invention after the data of the heart rate signal data of several first time points, temperature data, skin electrical signal data, moisture of skin numeric data, body posture information data, predicts the default body blood viscosity of first time point, the prediction data of blood glucose and glycosylated hemoglobin value in the first preset time period by the depth confidence network model completed using training in the first preset time period by acquiring the electrocardio wave of default body.Therefore, the present invention effectively solves the technological deficiency for bringing the accuracy of the uncomfortable and current diabetes index prediction technique of blood sampling low to patient existing for intrusive diabetes Evaluation of blood test technology traditional at present.
Description
Technical field
The invention belongs to intelligent monitor equipment technical fields, more particularly to the diabetes based on depth confidence network model refer to
Mark prediction technique and its system.
Background technique
Diabetes are to be only second to the high-risk chronic disease of cancer, and the crowd of the not all right having diabetes in China has been up to 5,000
Ten thousand, and the situation gradually increased is presented in the patient groups of diabetes in recent years, and diabetes and its complication are not only to patient's
Quality of life impacts, and the family for returning patient brings heavy economy and psychological burden, becomes serious public health and asks
Topic.
Blood glucose target is the major criterion for diagnosing diabetes, can be to tested by detection fasting blood-glucose and 2h-plasma glucose value
Whether person, which suffers from diabetes, diagnoses.There is obvious " three-many-one-little " symptom, as long as an abnormal plasma glucose value, that is, diagnosable.Without disease
Shape person diagnoses diabetes and needs abnormal plasma glucose value twice.
Glycosylated hemoglobin (GHb) is the product that the hemoglobin in red blood cell is combined with the carbohydrate in serum.It is
Reacted and formed by slow, lasting and irreversible saccharification, content number depend on blood sugar concentration and blood glucose with it is blood red
Albumen time of contact, and with blood drawing time, patient whether on an empty stomach, it is whether unrelated using the factors such as insulin.Therefore, GHb can have
The case where reflection diabetic in effect ground goes over glycemic control in 1~2 month.GHb is made of HbA1a, HbA1b, HbA1c,
Middle HbA1c accounts for about 70%, and stable structure, therefore is used as the monitoring index of blood glucose control.
Total cholesterol refers to the summation of cholesterol contained by all lipoprotein in blood, the higher liver for illustrating human body of total cholesterol
Substantive lesion takes place with lung.Crowd's total cholesterol level depends primarily on inherent cause and the total gallbladder of life style is solid
Alcohol includes free cholesterol and cholesteryl ester, and liver is the major organs of synthesis and storage.Cholesterol is synthesis adrenal cortex
The important source material of the physiological activators such as hormone, sex hormone, bile acid and vitamin D, and the main component of cell membrane is constituted,
Its serum-concentration can be used as the index of lipid metaboli.
Existing blood sugar test equipment needs blood taking needle being pierced into the skin of patient or blood vessel takes blood sample based on intrusive mood
This, blood sampling can be brought to patient does not accommodate infection risk；In addition, heart rate variability rate (heart rate variability,
HRV) be Assessment of Cardiac Autonomic Nerve regulatory function situation sensitive, non-invasive objective indicator, in the past 10 years with HRV predict blood
Sugar, glycosylated hemoglobin index have breakthrough, it is found that there is correlativity, the HRV phases of diabetic for HRV and blood glucose
It is higher than in normal person.But the correlation of HRV and blood glucose are very poor, r only has 0.2~0.4 or so, therefore is directly predicted with HRV
The precision of blood glucose is very poor.And it is many studies have shown that HRV is also influenced by several factors, such as the age, gender, constitutional index,
Environmental factor, movement, temperature, mood etc. so while having breakthrough in the field relevant to diabetes HRV, but can not also
Reach application.In addition measurement HRV still leads measuring method using 12 now, although measurement is accurate, to operator
The specialized capability of member requires height, and needs special measurement place, can not carry out large-scale application.
Therefore, the non-intrusion type diabetes blood that a forecasting accuracy is high, easy to use, portable, based on HRV is researched and developed to refer to
Mark the technical issues of detection device is those skilled in the art's urgent need to resolve.
Summary of the invention
In view of this, the diabetes index prediction technique that the invention discloses one based on depth confidence network model and its being
System, effectively solve traditional intrusive diabetes Evaluation of blood test technology at present it is existing to patient bring blood sampling it is uncomfortable with
And the low technological deficiency of accuracy of diabetes index prediction technique at present.
The invention discloses a kind of diabetes index prediction techniques based on depth confidence network model, comprising:
Acquire the electrocardio wave of default body in the first preset time period the heart rate signal data of several first time points,
Temperature data, skin electrical signal data, moisture of skin numeric data, the data of body posture information data；
The heart rate signal data of data and the first time point to the electrocardio wave of each first time point, body temperature number
Correlation analysis is carried out according to, the data of skin electrical signal data, moisture of skin numeric data, body posture information data, by phase
Close the heart rate signal data, the temperature data, the skin electrical signal data, the skin that property coefficient is greater than preset value
Moisture figure data, the data of the body posture information data as the first time point diabetes index it is crucial because
Element；
By the heart rate signal data of each first time point, the temperature data, the skin electrical signal data, institute
State moisture of skin numeric data, the data of the body posture information data, the key factor of the diabetes index and the electrocardio
The acquisition data of the data of wave form the data set of the first time point, and the data set of each first time point is formed data
Library；
Depth confidence network model is established, the data set of first time points multiple in the database is inputted into the depth
Confidence network model, and the depth confidence network model is trained；
The depth confidence network model completed using training predicts the default body first in the first preset time period
Blood viscosity, the prediction data of blood glucose and glycosylated hemoglobin value at time point.
Specifically, acquiring the heart rate letter of electrocardio wave several first time points in the first preset time period of default body
Number, temperature data, skin electrical signal data, moisture of skin numeric data, the data of body posture information data are specifically wrapped
It includes: collected data is pre-processed, electrocardio wave be then segmented to ecg wave form, except making an uproar to every section of heart
Electrical waveform and the body movement state being detected simultaneously by with this section of ecg wave form, skin body temperature, skin pricktest and humidity of skin index
The primary features for inputting depth confidence network are extracted jointly；After ecg wave form segmentation, d dimension is extracted to every section of ecg wave form
Feature, by dimension to body movement state, skin body temperature, skin pricktest and humidity of skin feature carry out mean value be 0, variance 1
Standardization, obtain being eventually used for the d dimension primary features of input depth confidence network；It tags to each sample, label is inspection
Measure blood viscosity, blood glucose and the glycosylated hemoglobin value of subject
It is specifically included specifically, establishing depth confidence network model: primary features obtained above is inputted into depth confidence
Network, layer-by-layer unsupervised pre-training depth confidence network, the depth confidence network model optimized, for extract collect by
The abstract characteristics of examination person's body data, specific steps are as follows:
Depth confidence network for extracting subject's body acquisition data abstraction feature is stacked by limited Boltzmann machine
It forms, wherein first layer is input layer.Another layer is the depth confidence that data abstraction feature is acquired for extracting subject's body
The implicit number of plies of network.Each limited Boltzmann machine is made of one layer of visible layer and one layer of hidden layer, wherein described can
See that layer is the input layer of each limited Boltzmann machine, the input layer of first limited Boltzmann machine is the subject without label
Body acquires the primary features of data, and being limited Boltzmann machine visible layer number of nodes, rule of thumb information and test result are manually adjusted
The unit number of whole respective layer determines: the hidden layer of limited Boltzmann machine is the output layer of limited Boltzmann machine, each limited
Input of the output of Boltzmann machine as next limited Boltzmann machine, the node in hidden layer of each limited Boltzmann machine
For the visible layer number of nodes of next limited Boltzmann machine.
Activation primitive is chosen to hidden layer, by contrast divergence algorithm and gibbs sampler to the every of limited Boltzmann machine
One layer of progress successive ignition training obtains parameter preferably depth confidence network model, each parameter to update model parameter
The calculation of update is as follows:
Δwij=ε (< vihj>data-<vihj>recom)
Δbi=ε (< vi>data-<vi>recom)
Δaj=ε (< hj>data-<hj>recom)；
Wherein Δ wijIndicate the renewal amount of the weight matrix of i-th layer of connection and jth layer, Δ biIndicate that i-th of visible layer is inclined
The renewal amount set, Δ ajIndicate that the renewal amount of j-th of hidden layer biasing, ε indicate learning rate,<>dataIndicate real data distribution
Situation,<>reconIndicate model distribution situation after reconstructing.
After the depth confidence network model after above-mentioned pre-training plus one layer of softmax is returned, and is carried out to whole network anti-
To fine tuning, the abstract characteristics of subject's body acquisition data of the last one hidden layer of depth confidence network output are input to
Softmax returns layer, and the result of output is that each subject's body acquires the corresponding blood viscosity of data, blood glucose and HbAle
Protein value label, and by prediction blood viscosity, blood glucose and the glycosylated hemoglobin of output and actual blood viscosity, blood glucose and sugar
Change Hemoglobin Value to be compared, obtains the predictablity rate of the index.Specific steps are as follows:
Increase by one softmax layers after the last one hidden layer of depth confidence network after pre-training as output
Layer calculates the probability that each subject's body acquisition data sample is divided into certain section physiochemical indice, subject's body is acquired
Data sample is divided into the label of corresponding maximum probability, then the physiochemical indice of subject prediction is the corresponding actual measurement of label
Physiochemical indice.For input sample x, it is divided into the new probability formula of corresponding physiochemical indice are as follows:
Then the corresponding prediction label of the input is the corresponding label of maximum probability value, formula are as follows:
ypred=argmaxmP (Y=m | x, W, b)
Wherein W indicates that softmax layers of weight matrix, b indicate softmax layers of biasing, and Y indicates that each probability is corresponding
Identification (RFID) tag, ypredIndicate the corresponding prediction label of maximum probability.
The data of physiochemical indice label are led to using the parameter of backpropagation and gradient descent method fine tuning whole network
It crosses and successively minimizes loss function minimum reconstructed error, acquire the optimized parameter of whole network, and then obtain extracting subject
Body acquires the optimum depth confidence network model of data abstraction feature, and tests the prediction blood of each subject based on this model
Fluid viscosity, blood glucose and glycosylated hemoglobin value, and be compared with the blood viscosity of actual measurement, blood glucose and glycosylated hemoglobin value,
Obtain the predictablity rate of blood viscosity, blood glucose and glycosylated hemoglobin.
Preferably, the depth confidence network model includes an input layer, an output layer and is arranged described defeated
Enter multiple hidden layers between layer and the output layer, any two adjacent layer in each hidden layer and the input layer forms one
A limited Boltzmann machine, the output layer and adjacent hidden layer form a BP neural network module；Each layer is provided with
The node of preset quantity, and connection weight matrix is provided between adjacent layer.
Preferably, described be trained the depth confidence network model, comprising:
A training objective function, the instruction are established in each limited Boltzmann machine of the depth confidence network model
The variable for practicing objective function is the state value of each node in the limited Boltzmann machine, bias and the limited Boltzmann machine
In connection weight matrix between two layers, the training objective of the training objective function be the limited Boltzmann machine energy value most
It is small；
The optional connection weight matrix and each section in the limited Boltzmann machine are calculated using maximum- likelihood estimation
The optional bias of point；
Using comparison hash degree algorithm progress optimizing, in the optional of the optional connection weight matrix and each node
The optimal bias of optimal connection weight matrix and each node is filtered out in bias.
Preferably, predicting that the default body is pre- first in the depth confidence network model completed using training
If in the period before the prediction data of the blood viscosity of first time point, blood glucose and glycosylated hemoglobin value, further includes:
The error of the bias of each node and every is calculated in the depth confidence network model using back-propagation algorithm
The error of one connection weight matrix；
Using obtained error to the bias and each connection weight square of each node of depth confidence network model
Battle array is modified.
Preferably, determining the number of nodes of the input layer using following formula:
N1=m1 × m2；
Wherein, N1 is the number of nodes of input layer, and m1 is the number of first time point, and m2 is the of each first time point
The sum of the number of key factor of the diabetes index at one time point.
Preferably, determining the number of nodes of each hidden layer using following formula:
N2=N1+N3+a1；
Wherein, N2 is the number of nodes of each hidden layer, and N1 is the number of nodes of the input layer, and N3 is the output layer
Number of nodes, a1 be more than or equal to 0 and be less than or equal to 10 integer.
Preferably, the output layer is softmax function, the softmax function formula are as follows:
Wherein output layer data formula are as follows:
ypred=argmaxmP (Y=m | x, W, b) (2)；
The wherein weight matrix that W is softmax layers, the biasing that b is softmax layers, Y are the corresponding mark mark of each probability
Label, ypredIndicate the corresponding output layer data of maximum probability.
Preferably, determining the number of nodes of the output layer using following formula:
N3=m1 × m4
Wherein, N3 is the number of nodes of output layer, and m1 is the number of first time point in first preset time period, m4
For the binary coding digit of the key factor of diabetes index in the data set of each first time point.
Preferably, the hidden layer chooses activation primitive, by contrast divergence algorithm and gibbs sampler to limited glass
The each layer of progress successive ignition training of the graceful machine of Wurz obtains parameter to update the parameter of the depth confidence network model
Preferably depth confidence network model.
The invention also discloses a kind of diabetes index forecasting systems based on depth confidence network model, comprising:
Data acquisition module, for acquiring the electrocardio wave of default body, several are at the first time in the first preset time period
Heart rate signal data, temperature data, skin electrical signal data, moisture of skin numeric data, the body posture information data of point
Data；
Correlating module, described in the data and the first time point for the electrocardio wave to each first time point
Heart rate signal data, temperature data, skin electrical signal data, moisture of skin numeric data, the data of body posture information data
Correlation analysis is carried out, relative coefficient is greater than to the heart rate signal data, the temperature data, the skin of preset value
Electrical signal data, the moisture of skin numeric data, the data of the body posture information data are as the first time point
The key factor of diabetes index；
Database forms module, for by the heart rate signal data of each first time point, the temperature data, institute
State skin electrical signal data, the moisture of skin numeric data, the data of the body posture information data, the diabetes index
Key factor and the acquisition data of data of the electrocardio wave form the data set of the first time point, by each first time point
Data set formed database；
Model building module, for establishing depth confidence network model, by first time points multiple in the database
Data set inputs the depth confidence network model, and is trained to the depth confidence network model；
Index prediction module, the depth confidence network model for being completed using training predict the default body first
The blood viscosity of first time point, the prediction data of blood glucose and glycosylated hemoglobin value in preset time period.
As can be seen from the above technical solutions, the invention has the following advantages that
In the present invention, pass through the potential change of human heart, heart rate signal data, the temperature data of human body, skin telecommunications
After number, the moisture of skin numeric data of human body, the body posture information data of human body, built-in in microprocessor into crossing
The predicted value of blood viscosity, blood glucose and glycosylated hemoglobin is calculated in preset program, finally by blood viscosity, blood glucose and saccharification
The predicted value of hemoglobin shows over the display, thus realize it is round-the-clock, measure to non-intrusion type in real time it is each in blood
Kind index.The present invention is based on depth confidence network models to obtain the data of diabetes physiochemical indice label, using backpropagation and
Gradient descent method finely tunes the parameter of whole network, minimizes reconstructed error by successively minimizing loss function, acquires entire
The optimized parameter of network, and then obtain extracting the optimum depth confidence network model of subject's body acquisition data abstraction feature,
And test prediction fasting blood-glucose, glycosylated hemoglobin and the total cholesterol value of each subject based on this model, and with actual measurement
Fasting blood-glucose, glycosylated hemoglobin and total cholesterol value are compared, and obtain fasting blood-glucose, glycosylated hemoglobin and total cholesterol
The predictablity rate of value.The present invention can to the physiochemical indice of user carry out in real time assessment monitoring, have the characteristics that it is portable easy-to-use,
Both bedside diagnosis can be carried out, the use of household custodial care facility is also can be used as, also has the function of easy to spread.And non-intrusion type
Detection method can allow patient to have more comfortable medical treatment to experience, and can also allow patient that can carry out self to condition of blood at home
Detection, is preferably managed disease.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described.
Fig. 1, which shows in the structure of portable blood detection device provided by the invention a kind of, cuts open figure；
Fig. 2 shows a kind of portable blood detection device provided by the invention；
Fig. 3 shows a kind of circuit diagram of portable blood detection device provided by the invention；
Fig. 4 shows the correlation of detection blood glucose value provided by the invention with actual measurement blood glucose value；
Fig. 5 shows the detection blood glucose value and actual measurement blood glucose value correlation point of portable blood detection device provided by the invention
Analysis, * * * * indicate that the detection blood glucose value of portable blood detection device provided by the invention and actual measurement blood glucose value correlation are significant, P
<0.05；
Fig. 6 shows the detection blood viscosity of portable blood detection device provided by the invention and the pass of actual measurement blood viscosity
System, wherein survey 3 times for blood viscosity one day, just get up morning, 1h before 1h and supper after lunch；
Fig. 7 shows the detection blood viscosity value and actual measurement blood viscosity value phase of portable blood detection device provided by the invention
The analysis of closing property, * * * * indicate that the detection blood viscosity value of portable blood detection device and actual measurement blood viscosity value correlation are aobvious
It writes, P < 0.05；
Fig. 8 shows the detection glycosylated hemoglobin and actual measurement blood HbAle of portable blood detection device provided by the invention
The relationship of albumen；
Fig. 9 shows the detection glycosylated hemoglobin value and actual measurement HbAle of portable blood detection device provided by the invention
Protein value correlation analysis, * * * * indicate the detection glycosylated hemoglobin value and actual measurement HbAle of portable blood detection device
Protein value correlation is significant, P < 0.05；
Wherein, appended drawing reference, body temperature detector 1, moisture of skin detector 2, skin resistance detector 3, Electrocardiography
Instrument 4, gravity accelerometer 5, microprocessor 6, power supply 7, heart rate detector 8, display 9, button A.
Specific embodiment
The invention discloses diabetes index prediction techniques and its system based on depth confidence network model, effectively solve
Blood viscosity method traditional at present can not carry out round-the-clock real-time monitoring to blood viscosity, and can bring to detected person
The technological deficiency of pain.
The technical scheme in the embodiments of the invention will be clearly and completely described below, it is clear that described implementation
Example is only a part of the embodiment of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, this field is common
Technical staff's every other embodiment obtained without making creative work belongs to the model that the present invention protects
It encloses.
The embodiment of the invention provides the specific implementations of the diabetes index prediction technique based on depth confidence network model
Mode (or can detect the heart rate signal data, the temperature data, the skin electrical signal number using current bracelet
According to, the physical signs detector of the moisture of skin numeric data, the body posture information data and heart wave data) it can examine
Survey HRV (SDNN, SDANN, RMSSD, PNN50), heart rate, body temperature, humidity of skin, skin electrical signal and body movement state number
According to, and the data that above-mentioned data are inputted and measured are calculated by blood glucose, glycosylated hemoglobin and total gallbladder by computing module and are consolidated
The predicted value of alcohol index.
In overseas Chinese hospital, 100 diabetic subjects made a definite diagnosis are randomly choosed from endocrine clinic, and from families of patients
150 volunteers are recruited, totally 250 people.Wherein 52 people of male, 98 people of women, the age of subject was at 30~70 years old.Use life
Reason Indexs measure device detects to obtain the heart rate of 150 volunteers, HRV, body temperature, humidity of skin, electrodermal activity, movement shape
The data informations such as state, resting heart rate, heart rate volatility and heart rate resume speed.It is detected using existing blood glucose and glycosylated hemoglobin
Method measures the blood glucose of subject and the numerical value of glycosylated hemoglobin, and the equipment used is the saccharification detection of Bo Tangping blood glucose
Instrument, specific implementation method according to equipment operation specification, using existing blood sugar detecting method to the total cholesterol of subject into
Row measurement, the equipment used are people's health lipids detection instrument, and specifically used method uses existing survey according to equipment operation specification
The method (capillary viscosity mensuration) of blood viscosity measures the blood viscosity numerical value of subject, and the equipment used is section
6000 blood rheological instrument of West Germany, specific implementation method are to carry out using match section in blood sampling 5ml, 2h on an empty stomach to subject using green head tube
6000 blood rheological instrument of West Germany is detected.
Blood was taken using three days respectively, fasting plasma glucose concentration, total cholesterol level, glycosylated hemoglobin is measured, makes even respectively
Mean value, as a result as described in Table 1.
Table 1
Prediction model is built
Depth confidence learning model is built, using the numerical value measured above, training pattern, and exports predicted value.
Input and output setting:
This time 250 volunteers are as data in selection above-described embodiment, after sample is upset at random, wherein the number of 200 people
According to as training data, the data of 50 volunteers are as inspection data.According to present studies have shown that research sample size reaches
To 100 or more, diabetes blood index of correlation has universality, is able to reflect the feature of universal sample.Because in this model
Reflection predicted value (fasting plasma glucose concentration, total cholesterol level, glycosylated hemoglobin) trend characteristic value have fasting plasma glucose concentration,
The characteristic value of total cholesterol level, glycosylated hemoglobin, and have there are three types of different conditions (sleep, empty stomach tranquillization, motion state),
Every kind of state measurement time is 1h, calculates measured value every 1min and is averaged, can collect 60 data points.It selects first 10 minutes
Information data predicts the 11st minute trend, so the input layer of this paper prediction model is (5 × 10) a input altogether, and presses every point
The time sequencing of clock is sequentially input.Output layer neuron type has 3 seed types (fasting plasma glucose concentration, total cholesterol level, saccharification
Hemoglobin)；
Hidden layer setting:
Three depth confidence network models are set separately to train the hidden layer number of plies, first network model it is implicit layer by layer
Number is arranged to 3 layers, and the neuron number of hidden layer is successively set to 150,100,60；Second depth confidence network mould
The hidden layer number of plies of type is arranged to 4 layers, and the neuron number of every layer of hidden layer is successively set to 150,100,60,40；The
The hidden layer number of plies of three depth confidence network models is arranged to 5 layers, and every layer of neuron number is successively set to 150,
100,60,40,20；
Depth confidence network model parameter setting:
Depth confidence network model parameter includes the weight coefficient ω between every layer of each neuron, visual layers neuron
The bias vector bj of bias vector ai, hidden layer neuron.The process that parameter determines is before this by parameter initialization, then by instructing
Practice constantly adjustment to update.The initial weight matrix ω of the embodiment of the present invention is set to the random number of normal distribution N (0,0.01),
If the bias vector of visual layers neuron is taken as ai, the bias vector of hidden layer neuron is chosen for bj, wherein initial hidden
The bias vector b0=0 of the neuron containing layer, the θ initial value of the depth confidence network model of the embodiment of the present invention are definite value, value
It is 0.1, training cut-off flag bit learning efficiency θ < 0.001.According to training sample data directly to connection weight ω, visual layers and
Hidden layer biasing is trained update adjustment, and training sample data are divided by the present embodiment in advance is according to according to every 10 samples
1 group of Small Sample Database collection is divided.
Bibliography " research of diabetes heart rate variability map feature and its clinical correlation " uses measurement HRV
Method predicts blood glucose, and HRV predicted value and actual value (blood drawing determines) are compared.
Predict error=abs (predicted value-actual value)/actual value.
Setting prediction error≤0.15 is that prediction is accurate, and it is pre- to be below that previous methods export 50 inspection datas
It is as follows to survey accuracy rate situation:
| Fasting blood-glucose | Previous methods |
| Model prediction accuracy rate | 40.56% |
The Accuracy Analysis of diabetes index prediction technique based on depth confidence network model: the blood that the present invention is predicted
Sugar, glycosylated hemoglobin and total cholesterol (predicted value of the invention) and blood glucose, the HbAle egg measured using conventional method
The numerical value (blood drawing actual value) of bletilla total cholesterol compares, and comparing result is as follows:
Predict error=abs (predicted value-actual value)/actual value.
Setting predicts that error≤0.15 is that prediction is accurate, is prediction of the model to 50 inspection output after training below
Accuracy rate situation.
It is as follows that model implies number of plies setting example predictablity rate contrast table:
Wherein, the fasting blood sugar of normal blood glucose value normal person is 3.89~6.1mmol/L；Such as larger than 6.1mmol/L
And being less than 7.0mmol/L is impaired fasting glucose；As fasting blood-glucose is more than or equal to 7.0mmol/L consideration diabetes twice；It is recommended that
Check fasting blood-glucose, carbohydrate tolerance test.If random blood sugar, which is more than or equal to 11.1mmol/L, can make a definite diagnosis diabetes.
Therefore, as seen from the above analysis, prediction technique of the invention is higher than existing prediction technique accuracy.
The embodiment of the invention also provides a kind of specific embodiment of portable blood detection device, embodiment 2 includes:
Electrocardiography instrument 4, heart rate detector 8, body temperature detector 1, skin resistance detector 3, moisture of skin detector 2, gravity add
Velocity sensor 5, microprocessor 6, display 9 and power supply 7；Electrocardiography instrument 4, heart rate detector 8, body temperature detector 1, skin
Skin resistance detector 3, moisture of skin detector 2, gravity accelerometer 5, microprocessor 6 and display 9 connect with power supply 7
It connects；Electrocardiography instrument 4, heart rate detector 8, body temperature detector 1, skin resistance detector 3, moisture of skin detector 2 and again
Power acceleration transducer 5 is electrically connected with microprocessor 6；Microprocessor 6 and display 9 are electrically connected.
After starting detection, the device of the invention collects human ecg signal (P wave number evidence, QRS by Electrocardiography instrument 4
Wave number evidence, T wave number evidence, PR interval data, QT interval data, PR segment data and ST segment data), and the electrocardiosignal of collection is passed
Defeated to arrive microprocessor 6, microprocessor 6 carries out analysis quantization to electrocardiosignal；The device of the invention is collected by heart rate detector 8
Human heart rate's signal；The local volume temperature value of human body is collected by body temperature detector 1；Human body is collected by skin resistance detector 3
Skin electrical signal (skin resistance, skin voltage, skin potential SP, skin potential level SPL, skin potential response SPR, skin
The horizontal SCL of conductance, skin conductivity reaction SCR), human body moisture of skin numerical value is collected by moisture of skin detector 2, passes through gravity
Acceleration transducer 5 detects the acceleration of human body to obtain the motion state and body gesture of human body, and above-mentioned detection is tied
Fruit transmits microprocessor 6.Microprocessor 6 is by heart rate, body temperature, skin electrical signal, moisture of skin, body movement state and body appearance
Gesture information carries out comprehensive analysis, and the ECG Signal Analysis for combining Electrocardiography instrument 4 to transmit is as a result, by preset meter
Calculate the detected value that blood viscosity, blood glucose and glycosylated hemoglobin is calculated in formula.In portable equipment measurement through the invention
Index is stated, multiple indexs including electrocardiogram can be measured round-the-clock, in real time, and can incite somebody to action according to Promethean mathematical model
Measured index is converted to blood index of correlation.
The embodiment of the invention provides a kind of specific embodiment of portable blood detection device, embodiment 2 includes: the heart
Electrograph detector 4, heart rate detector 8, body temperature detector 1, skin resistance detector 3, moisture of skin detector 2, gravity accelerate
Spend sensor 5, microprocessor 6, display 9 and power supply 7；Electrocardiography instrument 4, heart rate detector 8, body temperature detector 1, skin
Resistance detector 3, moisture of skin detector 2, gravity accelerometer 5, microprocessor 6 and display 9 connect with power supply 7
It connects；Electrocardiography instrument 4, heart rate detector 8, body temperature detector 1, skin resistance detector 3, moisture of skin detector 2 and again
Power acceleration transducer 5 is electrically connected with microprocessor 6；Microprocessor 6 and display 9 are electrically connected.
Specifically, Electrocardiography instrument 4 is specially conducting wire Electrocardiography instrument, conducting wire Electrocardiography instrument be will test
Obtained P wave number evidence, QRS wave data, T wave number evidence, PR interval data, QT interval data, PR segment data and ST segment data is sent
To microprocessor.
Specifically, heart rate detector 8 is specially infrared ray heart rate detector, heart rate is detected by infrared ray.
Specifically, body temperature detector 1 is specially liquid crystal thermometer.
Specifically, the current potential of the specific resistance of skin for obtaining human body of skin resistance detector 3, the voltage of skin, skin
SP, the potential level SPL of skin, the potentiometric response SPR of skin, the horizontal SCL of conductance of skin, skin conductance react SCR.
Specifically, moisture of skin detector 2 specifically obtains human skin water content.
Specifically, gravity accelerometer 5 obtains the acceleration information of human body.
Specifically, microprocessor 6 is specially central microprocessor CPU.
Specifically, display 9 specifically: cathode-ray tube display, plasma display or liquid crystal display.
Specifically, power supply 7 is specially external battery.
More specifically, power supply 7 can power from household socket, and battery power supply also can be used.
Electrocardio that the embodiment of the present invention 2 is detected by sensor, heart rate, body temperature, skin pricktest, moisture of skin, body fortune
Dynamic state and body gesture signal data.After the completion of test, blood viscosity, blood glucose and the sugar of detection can be checked by button A
Change Hemoglobin Value.
Studies have shown that blood viscosity is related with multiple indexs that can be detected by external non-invasive manner, index of correlation is retouched
State as follows: electrocardiogram (ECG) refer to heart in each cardiac cycle, it is in succession excited by pacemaker, atrium, ventricle, along with
The figure of the potential change of diversified forms is drawn in bioelectric variation by EGC sensor from body surface.Early-stage study discovery, blood
Fluid viscosity with cardiac myocytes contractility be it is related, blood viscosity is higher, and the myocardial contractive power that body blood supply needs is got over
Greatly, it will lead to electrocardio variation.Therefore, the variation of blood viscosity can be showed by electrocardio, it can by detecting the heart
Electricity detects blood viscosity.In addition to electrocardio, early-stage study discovery, heart rate, body temperature, moisture of skin, skin pricktest and machine
Also there is relationships with blood viscosity for body motion state.When body is in different sound states, such as it is static, move and sleep
It sleeps, the speed of blood flow is different.Therefore, in the case where blood viscosity is constant, cardiac muscle needs the diastole and contraction exported
Power can be changed because of the difference of body status, therefore electrocardio current potential and waveform can all change.Therefore, it is necessary to pass through
The motion state of gravity sensor measurement body can just obtain more accurate blood to be corrected to the electrocardiogram (ECG) data measured
Viscosity value；Body is in different body temperatures and can also have an impact to the state of blood viscosity, the liquid of same material concentration
Body, the viscosity of substance can be varied at different temperature.Body temperature increases the mobility that will increase blood, reduces viscosity；
Conversely, hypothermia blood viscosity can relative increase.Therefore, accurately measurement blood viscosity need to take into account the factor of body temperature；It is different
Blood viscosity microcirculation can be had an impact, the variation that can be reflected in skin pricktest index, thus detect skin pricktest index energy
Auxiliary judges blood viscosity.
The present invention provides a kind of detection embodiments of portable blood detection device, in the endocrine clinic of overseas Chinese hospital
200 subjects are chosen, are detected using the present apparatus, subject uses present apparatus measurement index data, obtains subject's
Blood viscosity value, blood glucose value and glycosylated hemoglobin value, wherein the grouping of the abscissa of Fig. 5, Fig. 7 and Fig. 9 refers to by random
The detection data of patient has been divided into 5 groups by forest tree-model.
As shown in Fig. 4 to Fig. 9, the result of Fig. 4 and Fig. 5 illustrate testing result, actual measurement blood glucose value provided by the invention and inspection
The correlation for surveying blood glucose value is significant；The result explanation of Fig. 6 and Fig. 7, the detection of portable blood detection device provided by the invention
Blood viscosity value and actual measurement blood viscosity value correlation are significant；The result explanation of Fig. 8 and Fig. 9, Portable blood provided by the invention
The detection glycosylated hemoglobin value of liquid detection device and actual measurement glycosylated hemoglobin value correlation are significant.This is sent out in the present embodiment
The bright method for surveying blood glucose value, blood viscosity and glycosylated hemoglobin with traditional blood sampling compares, and comparing result display is originally
It invents the blood glucose value, blood viscosity and the glycosylated hemoglobin that measure and conventional method is close, it was demonstrated that blood glucose value that the present invention measures,
The accuracy of blood viscosity and glycosylated hemoglobin is high.
The above is only a preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art
For member, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications are also answered
It is considered as protection scope of the present invention.
Claims (10)
1. a kind of diabetes index prediction technique based on depth confidence network model characterized by comprising
Acquire the electrocardio wave for presetting body heart rate signal data of several first time points, body temperature in the first preset time period
Data, skin electrical signal data, moisture of skin numeric data, the data of body posture information data；
The heart rate signal data of data and the first time point to the electrocardio wave of each first time point, temperature data,
Skin electrical signal data, moisture of skin numeric data, the data of body posture information data carry out correlation analysis, by correlation
Coefficient is greater than the heart rate signal data, the temperature data, the skin electrical signal data, the moisture of skin of preset value
The key factor of numeric data, the data of the body posture information data as the diabetes index of the first time point；
By the heart rate signal data of each first time point, the temperature data, the skin electrical signal data, the skin
Skin moisture figure data, the data of the body posture information data, the key factor of the diabetes index and the electrocardio wave
The acquisition data of data form the data set of the first time point, and the data set of each first time point is formed database；
Depth confidence network model is established, the data set of first time points multiple in the database is inputted into the depth confidence
Network model, and the depth confidence network model is trained；
The depth confidence network model completed using training predicts the default body in the first preset time period at the first time
Blood viscosity, the prediction data of blood glucose and glycosylated hemoglobin value of point.
2. the diabetes index prediction technique according to claim 1 based on depth confidence network model, which is characterized in that
The depth confidence network model include an input layer, an output layer and setting the input layer and the output layer it
Between multiple hidden layers, any two adjacent layer in each hidden layer and the input layer forms a limited Boltzmann machine,
The output layer and adjacent hidden layer form a BP neural network module；Each layer is provided with the node of preset quantity, and
Connection weight matrix is provided between adjacent layer.
3. the diabetes index prediction technique according to claim 2 based on depth confidence network model, which is characterized in that
It is described that the depth confidence network model is trained, comprising:
A training objective function, the trained mesh are established in each limited Boltzmann machine of the depth confidence network model
The variable of scalar functions is two in the state value of each node in the limited Boltzmann machine, bias and the limited Boltzmann machine
Connection weight matrix between layer, the training objective of the training objective function are that the limited Boltzmann machine energy value is minimum；
Optional connection weight matrix in the limited Boltzmann machine and each node are calculated using maximum- likelihood estimation
Optional bias；
Optimizing is carried out using comparison hash degree algorithm, in the optional biasing of the optional connection weight matrix and each node
The optimal bias of optimal connection weight matrix and each node is filtered out in value.
4. the diabetes index prediction technique according to claim 2 based on depth confidence network model, which is characterized in that
When the depth confidence network model completed using training predicts that the default body is first in the first preset time period
Between before the prediction data of blood viscosity, blood glucose and glycosylated hemoglobin value put, further includes:
The error of the bias of each node and each company in the depth confidence network model are calculated using back-propagation algorithm
Connect the error of weight matrix；
Using obtained error to the bias of each node of depth confidence network model and each connection weight matrix into
Row amendment.
5. the diabetes index prediction technique according to claim 2 based on depth confidence network model, which is characterized in that
The number of nodes of the input layer is determined using following formula:
N1=m1 × m2；
Wherein, N1 is the number of nodes of input layer, and m1 is the number of first time point, when m2 is the first of each first time point
Between the sum of the number of key factor of diabetes index put.
6. the diabetes index prediction technique according to claim 2 based on depth confidence network model, which is characterized in that
The number of nodes of each hidden layer is determined using following formula:
N2=N1+N3+a1；
Wherein, N2 is the number of nodes of each hidden layer, and N1 is the number of nodes of the input layer, and N3 is the section of the output layer
Point quantity, a1 are the integer more than or equal to 0 and less than or equal to 10.
7. the diabetes index prediction technique according to claim 2 based on depth confidence network model, which is characterized in that
The output layer is softmax function, and the softmax function formula is
Wherein output layer data formula are as follows:
ypred=argmaxmP (Y=m | x, W, b) (2)；
The wherein weight matrix that W is softmax layers, the biasing that b is softmax layers, Y are the corresponding identification (RFID) tag of each probability,
ypredIndicate the corresponding output layer data of maximum probability.
8. the diabetes index prediction technique according to claim 2 based on depth confidence network model, which is characterized in that
The number of nodes of the output layer is determined using following formula:
N3=m1 × m4；
Wherein, N3 is the number of nodes of output layer, and m1 is the number of first time point in first preset time period, and m4 is every
The binary coding digit of the key factor of diabetes index in the data set of one first time point.
9. the diabetes index prediction technique according to claim 1 based on depth confidence network model, which is characterized in that
The hidden layer chooses activation primitive, by contrast divergence algorithm and gibbs sampler to each layer of limited Boltzmann machine into
The training of row successive ignition obtains parameter preferably depth confidence network to update the parameter of the depth confidence network model
Model.
10. a kind of diabetes index forecasting system based on depth confidence network model characterized by comprising
Data acquisition module, for acquiring the electrocardio wave of default body several first time points in the first preset time period
Heart rate signal data, temperature data, skin electrical signal data, moisture of skin numeric data, the data of body posture information data；
Correlating module, for the data of the electrocardio wave to each first time point and the heart rate of the first time point
Signal data, temperature data, skin electrical signal data, moisture of skin numeric data, the data of body posture information data carry out
Relative coefficient is greater than the heart rate signal data, the temperature data, the skin telecommunications of preset value by correlation analysis
The glycosuria of number, the moisture of skin numeric data, the data of the body posture information data as the first time point
The key factor of sick index；
Database forms module, for by the heart rate signal data of each first time point, the temperature data, the skin
The pass of skin electrical signal data, the moisture of skin numeric data, the data of the body posture information data, the diabetes index
The acquisition data of key factor and the data of the electrocardio wave form the data set of the first time point, by the number of each first time point
Database is formed according to collection；
Model building module, for establishing depth confidence network model, by the data of first time points multiple in the database
Collection inputs the depth confidence network model, and is trained to the depth confidence network model；
Index prediction module, the depth confidence network model for being completed using training predict that the default body is default first
The blood viscosity of first time point, the prediction data of blood glucose and glycosylated hemoglobin value in period.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810625965 | 2018-06-18 | ||
| CN2018106259658 | 2018-06-18 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109171756A true CN109171756A (en) | 2019-01-11 |
Family
ID=64917785
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811016543.7A Pending CN109171756A (en) | 2018-06-18 | 2018-09-03 | Diabetes index prediction technique and its system based on depth confidence network model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109171756A (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109887599A (en) * | 2019-02-25 | 2019-06-14 | 中国计量大学 | A kind of tcm prescription curative effect deduction method neural network based |
| CN111223568A (en) * | 2019-04-18 | 2020-06-02 | 岭南师范学院 | Diabetes early warning model for improving k-means clustering |
| CN111631729A (en) * | 2020-05-14 | 2020-09-08 | 中国科学院深圳先进技术研究院 | Hypoglycemia prediction method and system based on multimode fusion |
| CN111631704A (en) * | 2020-05-29 | 2020-09-08 | 中国科学院深圳先进技术研究院 | Diabetes early-stage detection system and method based on combination of electrocardio information and electroencephalogram information |
| CN111920410A (en) * | 2020-08-12 | 2020-11-13 | 上海诺诚电气股份有限公司 | Diabetes early-stage screening system and method based on big data |
| CN112741623A (en) * | 2020-12-09 | 2021-05-04 | 中国科学院深圳先进技术研究院 | Electrocardio-based glycosylated hemoglobin detection system and method |
| CN113035352A (en) * | 2021-04-27 | 2021-06-25 | 河南科技大学 | Diabetic retinopathy early warning method based on BP neural network |
| CN113229793A (en) * | 2021-06-23 | 2021-08-10 | 杭州市临安区第一人民医院 | Diabetes risk early detection management system |
| US11488038B2 (en) | 2019-03-29 | 2022-11-01 | Sony Network Communications Europe B.V. | Method and device for monitoring |
| CN116671906A (en) * | 2023-08-01 | 2023-09-01 | 亿慧云智能科技（深圳）股份有限公司 | Noninvasive blood glucose measurement method and noninvasive blood glucose measurement system for smart watch |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20040122296A1 (en) * | 2002-12-18 | 2004-06-24 | John Hatlestad | Advanced patient management for triaging health-related data |
| CN101107025A (en) * | 2005-01-17 | 2008-01-16 | 诺和诺德公司 | Fluide delivery device with integrated monitoring of physiological characteristics |
| CN105825230A (en) * | 2016-03-15 | 2016-08-03 | 中国农业大学 | Soil moisture status prediction method and system based on DBN (Deep Belief Network) model |
| CN106328121A (en) * | 2016-08-30 | 2017-01-11 | 南京理工大学 | Chinese traditional musical instrument classification method based on depth confidence network |
| CN106384013A (en) * | 2016-09-28 | 2017-02-08 | 湖南老码信息科技有限责任公司 | Incremental neural network model-based type-II diabetes prediction method and prediction system |
| WO2017054715A1 (en) * | 2015-09-30 | 2017-04-06 | 齐心 | Apparatus and method for measuring vital signs |
| CN106725518A (en) * | 2016-08-03 | 2017-05-31 | 赵宏杰 | Noninvasive infrared system for detecting blood sugar |
| CN107174258A (en) * | 2017-06-02 | 2017-09-19 | 北京信息科技大学 | Blood sugar concentration Forecasting Methodology |
-
2018
- 2018-09-03 CN CN201811016543.7A patent/CN109171756A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20040122296A1 (en) * | 2002-12-18 | 2004-06-24 | John Hatlestad | Advanced patient management for triaging health-related data |
| CN101107025A (en) * | 2005-01-17 | 2008-01-16 | 诺和诺德公司 | Fluide delivery device with integrated monitoring of physiological characteristics |
| WO2017054715A1 (en) * | 2015-09-30 | 2017-04-06 | 齐心 | Apparatus and method for measuring vital signs |
| CN105825230A (en) * | 2016-03-15 | 2016-08-03 | 中国农业大学 | Soil moisture status prediction method and system based on DBN (Deep Belief Network) model |
| CN106725518A (en) * | 2016-08-03 | 2017-05-31 | 赵宏杰 | Noninvasive infrared system for detecting blood sugar |
| CN106328121A (en) * | 2016-08-30 | 2017-01-11 | 南京理工大学 | Chinese traditional musical instrument classification method based on depth confidence network |
| CN106384013A (en) * | 2016-09-28 | 2017-02-08 | 湖南老码信息科技有限责任公司 | Incremental neural network model-based type-II diabetes prediction method and prediction system |
| CN107174258A (en) * | 2017-06-02 | 2017-09-19 | 北京信息科技大学 | Blood sugar concentration Forecasting Methodology |
Non-Patent Citations (1)
| Title |
|---|
| 李东明，贾书海: "基于多光谱应用BP人工神经网络预测血糖", 《激光与光电子学进展》 * |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109887599A (en) * | 2019-02-25 | 2019-06-14 | 中国计量大学 | A kind of tcm prescription curative effect deduction method neural network based |
| US11488038B2 (en) | 2019-03-29 | 2022-11-01 | Sony Network Communications Europe B.V. | Method and device for monitoring |
| CN111223568A (en) * | 2019-04-18 | 2020-06-02 | 岭南师范学院 | Diabetes early warning model for improving k-means clustering |
| CN111223568B (en) * | 2019-04-18 | 2024-03-26 | 岭南师范学院 | Diabetes early warning model for improving k-means clustering |
| CN111631729A (en) * | 2020-05-14 | 2020-09-08 | 中国科学院深圳先进技术研究院 | Hypoglycemia prediction method and system based on multimode fusion |
| CN111631729B (en) * | 2020-05-14 | 2023-06-06 | 中国科学院深圳先进技术研究院 | Hypoglycemia prediction method and system based on multimode fusion |
| CN111631704A (en) * | 2020-05-29 | 2020-09-08 | 中国科学院深圳先进技术研究院 | Diabetes early-stage detection system and method based on combination of electrocardio information and electroencephalogram information |
| CN111631704B (en) * | 2020-05-29 | 2021-06-25 | 中国科学院深圳先进技术研究院 | Diabetes early-stage detection system and method based on combination of electrocardio information and electroencephalogram information |
| CN111920410A (en) * | 2020-08-12 | 2020-11-13 | 上海诺诚电气股份有限公司 | Diabetes early-stage screening system and method based on big data |
| CN112741623A (en) * | 2020-12-09 | 2021-05-04 | 中国科学院深圳先进技术研究院 | Electrocardio-based glycosylated hemoglobin detection system and method |
| CN113035352A (en) * | 2021-04-27 | 2021-06-25 | 河南科技大学 | Diabetic retinopathy early warning method based on BP neural network |
| CN113229793A (en) * | 2021-06-23 | 2021-08-10 | 杭州市临安区第一人民医院 | Diabetes risk early detection management system |
| CN116671906A (en) * | 2023-08-01 | 2023-09-01 | 亿慧云智能科技（深圳）股份有限公司 | Noninvasive blood glucose measurement method and noninvasive blood glucose measurement system for smart watch |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109171756A (en) | Diabetes index prediction technique and its system based on depth confidence network model | |
| Joshi et al. | iGLU 2.0: A new wearable for accurate non-invasive continuous serum glucose measurement in IoMT framework | |
| CN103959291B (en) | Glucose predictions device based on the regularization network with adaptively selected core and regularization parameter | |
| RU2757048C1 (en) | Method and system for assessing the health of the human body based on the large-volume sleep data | |
| ES2659945T3 (en) | Waste based monitoring of human health | |
| CN101500475B (en) | Method and device for monitoring a physiological parameter | |
| US20090033333A1 (en) | Apparatus and method for measuring parameters associated with electrochemical processes | |
| US20150150476A1 (en) | Electrophysiological analysis system and method | |
| Alić et al. | Classification of metabolic syndrome patients using implemented expert system | |
| JPH0260626A (en) | Apparatus for obtaining a body impedence data | |
| CN109065162A (en) | A kind of comprehensive intelligent diagnostic system | |
| CN107705848A (en) | A kind of method and system for recommending coordinating program according to user health situation | |
| JP2001221803A (en) | Apparatus for determining saccharometabolic capacity | |
| Orha et al. | Automated system for evaluating health status | |
| Kim et al. | Advanced prediction model for individual thermal comfort considering blood glucose and salivary cortisol | |
| Gan et al. | Human-computer interaction based interface design of intelligent health detection using PCANet and multi-sensor information fusion | |
| CN110325115A (en) | System and method for determining window of becoming pregnant not loftyly | |
| CN107516019A (en) | Noninvasive health forecast system and method | |
| CN110428893A (en) | A kind of tele-medicine vital sign data acquisition device | |
| Nakajima et al. | Systems health care: Coevolutionary integration of smart devices and smart services | |
| CN117116462A (en) | Method, system and device for predicting glycosylated hemoglobin value | |
| Tamura | Progress of home healthcare sensor in our experience: Development of wearable and unobtrusive monitoring | |
| Scagliusi et al. | Bioimpedance Spectroscopy-based Edema Supervision Wearable System for Non-invasive Monitoring of Heart Failure | |
| CN206080503U (en) | Healthy all -in -one of intelligence | |
| CN110415778A (en) | A kind of individual health data management method and system based on cloud data |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190111 |
|
| WD01 | Invention patent application deemed withdrawn after publication |