CN107137093B - A kind of continuous blood sugar monitoring device comprising abnormal plasma glucose probability alarm - Google Patents
A kind of continuous blood sugar monitoring device comprising abnormal plasma glucose probability alarm Download PDFInfo
- Publication number
- CN107137093B CN107137093B CN201710262105.8A CN201710262105A CN107137093B CN 107137093 B CN107137093 B CN 107137093B CN 201710262105 A CN201710262105 A CN 201710262105A CN 107137093 B CN107137093 B CN 107137093B
- Authority
- CN
- China
- Prior art keywords
- alarm
- value
- blood glucose
- prediction
- probability
- 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.)
- Active
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
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
-
- 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
- A61B5/7225—Details of analog processing, e.g. isolation amplifier, gain or sensitivity adjustment, filtering, baseline or drift compensation
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/74—Details of notification to user or communication with user or patient ; user input means
- A61B5/746—Alarms related to a physiological condition, e.g. details of setting alarm thresholds or avoiding false alarms
Abstract
The invention discloses a kind of continuous blood sugar monitoring devices comprising abnormal plasma glucose probability alarm, the equipment is analyzed for the subcutaneous blood glucose measurement signal of human body, prediction modeling is carried out according to continuous blood sugar monitoring signals, the prediction error distribution situation of prediction technique is fitted, according to the distribution situation of prediction error, the calculating of alarm probabilities is carried out for the sampled point of each blood glucose level data.Due to that can change with external environment in the blood glucose of patient in fluctuation situation, the corresponding distribution situation for predicting error can also change, to need the analytical function of online update prediction error distribution.Abnormal plasma glucose alarm probabilities method employed in the present invention can be according to the overall distribution situation of newest prediction error transfer factor prediction error, so as to more accurately provide the high hypoglycemia alarm probabilities of current blood glucose sampling.Present device can accurately calculate the high hypoglycemia alarm probabilities of blood glucose sampled value, improve the accuracy of high hypoglycemia alarm.
Description
Technical field
The invention belongs to the research fields of blood glucose prediction alarm and analysis, include abnormal plasma glucose probability more particularly to one kind
The continuous blood sugar monitoring device of alarm.
Background technique
In order to realize good glycemic control, it is necessary to be measured to blood glucose level, currently used is continuous blood sugar
Monitor (CGM) equipment.With the fast development of continuous blood sugar monitoring device, so that the control of higher precision may become, it is real
When CGM system play an important role to the detection in advance of high hypoglycemia.However, being collected according to current CGM equipment
Blood glucose value alarm, then take corresponding remedy measures that can postpone the control for blood glucose, one side insulin is in vivo
Generation effect has the regular hour, therefore can postpone the control for hyperglycemia；Another aspect carbohydrate is in vivo
Just generation acts on after a period of time for meeting, therefore can postpone the control for hypoglycemia.So needing to shift to an earlier date a period of time carries out blood
Sugar prediction, suitable therapeutic strategy is selected according to the blood glucose situation of prediction.Traditional prediction and alarm method is directly will prediction
Blood glucose value and high hypoglycemia alarm threshold value be compared, alarm will be generated more than alarm threshold value.However due to prediction technique
There is certain prediction error, therefore predicted value and true blood glucose value have certain deviation, directly with the blood glucose value of prediction
It carries out alarming to generate failing to report and reporting by mistake, seriously affects the control strategy of blood glucose, influence the glycemic control of patient.
Summary of the invention
It is an object of the invention to be directed to the deficiency of existing prediction and alarm method, provide a kind of comprising abnormal plasma glucose probability report
The continuous blood sugar monitoring device of alert device.
The purpose of the present invention is achieved through the following technical solutions: a kind of company comprising abnormal plasma glucose probability alarm
Continuous blood glucose monitoring device, the equipment include: to export the sensor of blood glucose available signal for acquiring blood sugar for human body information；For
The signal amplifier of processing is amplified to the output signal of sensor；Analog signal for exporting to signal amplifier carries out
The single-chip microcontroller of number conversion；Digital signal for exporting to single-chip microcontroller carries out the filter of data processing；For it will filter after
Blood glucose value carry out the calculating of abnormal plasma glucose alarm probabilities abnormal plasma glucose probability alarm；For carrying out numerical value and waveform shows
Display；Memory for data storage；The process that the alarm carries out the calculating of abnormal plasma glucose alarm probabilities includes following
Step:
(1) acquisition of blood glucose prediction error: pass through the blood glucose prediction value of selected prediction modelIt is monitored with continuous blood sugar
The blood glucose value y of equipment acquisition obtains prediction error e:
Wherein yiIt is the sampled value of i-th of moment continuous blood sugar monitoring device,It is i-th of moment blood glucose prediction value, ei
It is the prediction error of i-th of moment blood glucose, N is hits, thus to obtain N number of prediction error E={ e1,e2,…,eN}；
(2) distribution situation of blood glucose prediction error is estimated by the method based on gauss hybrid models (GMM)；
(3) it is distributed and is estimated according to the prediction error of step (2), carry out the high hypoglycemia alarm probabilities of online blood glucose sampled point
Calculating, specifically include following sub-step:
(3.1) the median point of prediction probability of error density function is calculated:
Wherein P (e | θ) is the prediction probability of error density function estimated in step (2), and θ is parameter to be estimated in GMM,
That is θ=(μ, Σ), μ are the mean values of each list Gauss model in GMM, and Σ is the covariance matrix of each list Gauss model in GMM,
C is the median point of probability density function；
(3.2) confidence interval of prediction error is determined according to 95% confidence level:
Wherein σ is confidence radius, and c- σ is confidence lower limit, and c+ σ is confidence upper limit；
(3.3) the abnormal plasma glucose alarm probabilities of each blood glucose prediction value are calculated:
(3.3.1) is if blood glucose prediction value is higher than hyperglycemia alarm threshold value or blood plus the confidence lower limit of prediction error
Sugared predicted value is lower than hypoglycemia alarm threshold value plus the confidence upper limit of prediction error, then alarm probabilities are 1, that is, generates alarm；
(3.3.2) is if high hypoglycemia alarm threshold value will calculate the predicted value in blood glucose prediction value confidence interval
High hypoglycemia alarm probabilities, define alarm distance d according to the following formulalowAnd dhigh:
Wherein HlowIt is hypoglycemia alarm threshold value, HhighIt is hyperglycemia alarm threshold value, dlowIt is that predicted value and hypoglycemia are alarmed
The distance of threshold value, dhighIt is predicted value at a distance from hyperglycemia alarm threshold value；Then the probability of high hypoglycemia is according to following formula meter
It calculates:
It is wherein PlowHypoglycemia alarm probabilities, PhighIt is hyperglycemia alarm probabilities；
(4) the new prediction error update GMM parameter obtained according to new blood glucose sampled value.
Further, the step (2) specifically includes following sub-step:
(2.1) the prediction error E={ e obtained in estimating step (1)1,e2,…,eNDistribution, it is assumed that predict error
Probability density function is made of K single Gauss model, the probability density function of GMM are as follows:
Wherein P (e | θ) is prediction probability of error density function, αkIt is the weight of k-th of single Gauss model, and φ (e | θk) be
The probability density function of k-th of single Gauss model, meets:
Wherein μkIt is the mean value of k-th of single Gauss model, is the covariance matrix Σ of k-th of single Gauss modelk, θkIt is kth
A list Gauss model parameter set, i.e. θk=(μk,Σk), n is the dimension of e, and e is one-dimensional variable here, therefore n=1；
(2.2) using the parameter of EM algorithm estimation GMM, the specific steps are as follows:
(2.2.1) is for αk, μkAnd ΣkAssign initial value, k=1,2 ..., K；
(2.2.2) calculates posterior probability
WhereinIt is j-th of prediction error e in the s times iterationjBelong to the posterior probability of k-th of single Gauss model；
(2.2.3) is calculatedWithThe s+1 times iterative value:
WhereinWithIt is α respectivelyk, μkAnd ΣkThe s+1 times iterative value.
The step (4) specifically includes following sub-step:
(4.1) the prediction error e for newly obtainingN+1, e is calculated separately according to formula (7)N+1Corresponding each single Gauss model
Probability density function φ (eN+1|θk)；
(4.2) single Gauss model M corresponding to the wherein maximum value of probability density function is found outk, judge eN+1Whether in list
In the confidence interval of Gauss model 95%, if in 95% confidence interval of the list Gauss model, the list Gauss model
MkIt is eN+1Matching list Gauss model, using more new algorithm update GMM parameter；If continuous, there are three prediction errors not to exist
In 95% confidence interval of the corresponding single Gauss model of the maximum value of probability density function, then it is assumed that the structure of GMM changes
Become, re-starts GMM parameter Estimation according to formula (8)-(11)；
(4.3) GMM parameter more new algorithm is as follows:
ρ=λ * φ (eN+1|θk) (16)
Wherein, p (Mi|eN+1) it is identifier, only predict error eN+1Matched list Gauss model Mk its identifier p (Mk|
eN+1) it is just 1, the identifier of other list Gauss models is 0；It is the weight of i-th of the N+1 moment single Gauss model, only
Matched list Gauss model MkWeight increase, it is other list Gauss models weight decline；WithIt is to match at the N+1 moment
Single Gauss model MkMean value and covariance matrix, it is other list Gauss models mean value and covariance matrix without update；
φ(eN+1|θk) it is eN+1The probability density function of corresponding each single Gauss model；λ is adjustable parameter.
Compared with prior art, the beneficial effects of the present invention are: proposed by the invention alarms comprising abnormal plasma glucose probability
The continuous blood sugar monitoring device of device can estimate the high hypoglycemia alarm probabilities of blood glucose prediction value according to the distribution of prediction error,
Rather than the blood glucose value of prediction is compared with high hypoglycemia alarm threshold value directly and provides and whether alarms.Therefore, Bing Renke
To formulate suitable treatment plan according to abnormal plasma glucose alarm probabilities, abnormal plasma glucose event is avoided with this.The present invention is easy to real
It applies, the research for handling and analyzing for blood glucose specifies new direction.
Detailed description of the invention
Fig. 1 is the structural block diagram of continuous blood sugar monitoring device of the present invention；
Fig. 2 is the implementation flow chart that abnormal plasma glucose alarm probabilities calculate in continuous blood sugar monitoring device of the present invention；
Fig. 3 be object 7 be respectively adopted AR and ARX model predicted caused by predict the distribution function image of error,
(a) AR prediction model (b) ARX prediction model；
Fig. 4 is the blood glucose prediction value and blood sugar measured and corresponding height blood that object 7 uses AR model to be predicted
The comparison of sugared alarm probabilities and traditional alert method, (a) three days result (b) hypoglycemia alarm probabilities partial enlarged view (c) afterwards
Hyperglycemia alarm probabilities partial enlarged view；
Fig. 5 is the blood glucose prediction value and blood sugar measured and corresponding height that object 7 uses ARX model to be predicted
The comparison of blood glucose alarm probabilities and traditional alert method, (a) three days result (b) hypoglycemia alarm probabilities partial enlarged view afterwards
(c) hyperglycemia alarm probabilities partial enlarged view.
Specific embodiment
Invention is further described in detail in the following with reference to the drawings and specific embodiments.
As shown in Figure 1, a kind of continuous blood sugar monitoring device comprising abnormal plasma glucose probability alarm provided by the invention, packet
It includes: for acquiring blood sugar for human body information, exporting the sensor of blood glucose available signal；It is put for the output signal to sensor
The signal amplifier handled greatly；Analog signal for exporting to signal amplifier carries out the single-chip microcontroller of digital conversion；For right
The digital signal of single-chip microcontroller output carries out the filter of data processing；For filtered blood glucose value to be carried out abnormal plasma glucose alarm
The abnormal plasma glucose probability alarm of probability calculation；The display shown for carrying out numerical value and waveform；For depositing for data storage
Reservoir；The alarm carry out the process of abnormal plasma glucose alarm probabilities calculating the following steps are included:
Step 1: obtaining blood glucose prediction error
It will be that 5 minutes continuous blood sugar monitoring signals obtained are combined into one-dimensional time series data y with the sampling period1×l,
In, y indicates that the blood glucose signal detected, l are sample number, removes spiking therein.In this example, three group objects are shared
Sampled signal, the sampling period be 5 minutes, the 1st group be teenager's group, the 2nd group be adult group, the 3rd group be children's group, every group 10
People, three groups totally 30 people, the sampled signal of each object include five days data.
Here it is predicted using AR model and ARX model, the general formula of ARX are as follows:
Wherein y (k) is blood glucose value, uins(k-kins) and umeal(k-kmeal) it is insulin and diet input quantity, k respectivelyins
It is the input delay of insulin, kmealIt is the input delay of diet, β is deviation, and ε (k) is random perturbation, A (q-1),Bins(q-1) and Bmeal(q-1) respectively represent blood glucose, the coefficient of insulin and diet.Here, for AR model, B is chosenins(q-1)=Bmeal
(q-1)=0, A (q-1) order is 7；For ARX model, A (q is chosen-1), Bins(q-1) and Bmeal(q-1) order be 7,4 respectively
With 4；
First day blood glucose level data is modeled for AR and ARX, is predicted subsequently for second day blood glucose level data, thus
Obtain the prediction error e of blood glucose:
Wherein yiIt is the sampled value of i-th of moment continuous blood sugar monitoring device,It is i-th of moment blood glucose prediction value, ei
It is the prediction error of i-th of moment blood glucose, N is hits, thus to obtain N number of prediction error E={ e1,e2,…,eN, it selects here
Second day data, therefore N=288 are taken.
Step 2: the distribution situation of blood glucose prediction error is estimated by the method based on gauss hybrid models (GMM), specifically
Including following sub-step:
(2.1) the prediction error E={ e obtained in estimating step 11,e2,…,eNDistribution, it is assumed that predict the general of error
Rate density function is made of K single Gauss model, chooses K=4, the probability density function of GMM here are as follows:
Wherein P (e | θ) is prediction probability of error density function, αkIt is the weight of k-th of single Gauss model, and φ (e | θk) be
The probability density function of k-th of single Gauss model, meets:
Wherein μkIt is the mean value of k-th of single Gauss model, is the covariance matrix Σ of k-th of single Gauss modelk, θkIt is kth
A list Gauss model parameter set, i.e. θk=(μk,Σk), n is the dimension of e, and e is one-dimensional variable here, therefore n=1；
(2.2) using the parameter of EM algorithm estimation GMM, the specific steps are as follows:
(2.2.1) is for αk, μkAnd ΣkAssign initial value, k=1,2 ..., K；
(2.2.2) calculates posterior probability
WhereinIt is j-th of prediction error e in the s times iterationjBelong to the posterior probability of k-th of single Gauss model；
(2.2.3) is calculatedWithThe s+1 times iterative value:
WhereinWithIt is α respectivelyk, μkAnd ΣkThe s+1 times iterative value.Fig. 3 illustrates object 7 and divides
Not Cai Yong AR and ARX model predicted caused by predict the distribution function image of error.
Step 3: being distributed and estimated according to the prediction error of step 2, the high hypoglycemia alarm for carrying out online blood glucose sampled point is general
The calculating of rate specifically includes following sub-step:
(3.1) the median point of prediction probability of error density function is calculated:
Wherein P (e | θ) is the prediction probability of error density function estimated in step 2, and θ is parameter to be estimated in GMM, i.e.,
θ=(μ, Σ), μ are the mean values of each list Gauss model in GMM, and Σ is the covariance matrix of each list Gauss model in GMM, C
It is the median point of probability density function；
(3.2) confidence interval of prediction error is determined according to 95% confidence level:
Wherein σ is confidence radius, and c- σ is confidence lower limit, and c+ σ is confidence upper limit；
(3.3) the abnormal plasma glucose alarm probabilities of each blood glucose prediction value are calculated:
(3.3.1) is if blood glucose prediction value is higher than hyperglycemia alarm threshold value or blood plus the confidence lower limit of prediction error
Sugared predicted value is lower than hypoglycemia alarm threshold value plus the confidence upper limit of prediction error, then alarm probabilities are 1, that is, generates alarm；
(3.3.2) is if high hypoglycemia alarm threshold value will calculate the predicted value in blood glucose prediction value confidence interval
High hypoglycemia alarm probabilities, define alarm distance d according to the following formulalowAnd dhigh:
Wherein HlowIt is hypoglycemia alarm threshold value, HhighIt is hyperglycemia alarm threshold value, hypoglycemia alarm threshold value is set as here
70mg/dL, hyperglycemia alarm threshold value are set as 180mg/dL, dlowIt is predicted value at a distance from hypoglycemia alarm threshold value, dhighIt is
Predicted value is at a distance from hyperglycemia alarm threshold value；Then the probability of high hypoglycemia is calculated according to following formula:
It is wherein PlowHypoglycemia alarm probabilities, PhighIt is hyperglycemia alarm probabilities；
Step 4: the new prediction error update GMM parameter obtained according to new blood glucose sampled value specifically includes following son
Step:
(4.1) the prediction error e for newly obtainingN+1, e is calculated separately according to formula (4)N+1Corresponding each single Gauss model
Probability density function φ (eN+1|θk)；
(4.2) single Gauss model M corresponding to the wherein maximum value of probability density function is found outk, judge eN+1Whether in list
In the confidence interval of Gauss model 95%, if in 95% confidence interval of the list Gauss model, the list Gauss model
MkIt is eN+1Matching list Gauss model, using more new algorithm update GMM parameter；If continuous, there are three prediction errors not to exist
In 95% confidence interval of the corresponding single Gauss model of the maximum value of probability density function, then it is assumed that the structure of GMM changes
Become, re-starts GMM parameter Estimation according to formula (5)-(8)；
(4.3) GMM parameter more new algorithm is as follows:
ρ=λ * φ (eN+1|θk) (17)
Wherein, p (Mi|eN+1) it is identifier, only predict error eN+1Matched list Gauss model MkIts identifier p (Mk|
eN+1) it is just 1, the identifier of other list Gauss models is 0；It is the weight of i-th of the N+1 moment single Gauss model, only
Matched list Gauss model MkWeight increase, it is other list Gauss models weight decline；WithIt is to match at the N+1 moment
Single Gauss model MkMean value and covariance matrix, it is other list Gauss models mean value and covariance matrix without update；
φ(eN+1|θk) it is eN+1The probability density function of corresponding each single Gauss model；λ is adjustable parameter, and λ ∈ (0,1) is selected here
Take λ=0.002.
Fig. 4 and Fig. 5 respectively shows the blood glucose prediction value and blood glucose that object 7 uses AR model and ARX model to be predicted
The comparison of measured value and corresponding Gauss blood glucose alarm probabilities and traditional alert method.
Step 5: it is compared according to the alarm probabilities of calculating and traditional alarm method, is alarmed here using correct,
It is compared in terms of the delay four of the number of wrong report, the number failed to report and alarm.Here 40% alarm probabilities conduct is chosen
Alarm threshold value, the probability greater than 40% generate alarm.Table 1 for 3 groups (teenager's group, adult group and children's group) totally 30 it is right
Method of the blood glucose sampled data of elephant for traditional alarm method and based on alarm probabilities is compared, it can be seen that base
There are more correct alarm times in the method for probability alarm, less wrong report and fails to report number, meanwhile, presignal delay also wants small
In traditional alarm method.
Table 1 for 3 groups (teenager's group, adult group and children's group) totally 30 objects blood glucose sampled data for traditional
Alarm method and method based on alarm probabilities are compared (result is indicated with means standard deviation)
* A/B A is the number correctly alarmed or failed to report；B is high/low blood glucose event in total.
Claims (3)
1. a kind of continuous blood sugar monitoring device comprising abnormal plasma glucose probability alarm, which is characterized in that the equipment includes: to be used for
Blood sugar for human body information is acquired, the sensor of blood glucose available signal is exported；Processing is amplified for the output signal to sensor
Signal amplifier；Analog signal for exporting to signal amplifier carries out the single-chip microcontroller of digital conversion；For to single-chip microcontroller
The digital signal of output carries out the filter of data processing；By filtered blood glucose value to be carried out based on abnormal plasma glucose alarm probabilities
The abnormal plasma glucose probability alarm of calculation；The display shown for carrying out numerical value and waveform；Memory for data storage；Institute
State alarm carry out the calculating of abnormal plasma glucose alarm probabilities process the following steps are included:
(1) acquisition of blood glucose prediction error: pass through the blood glucose prediction value of selected prediction modelIt is adopted with continuous blood sugar monitoring device
The blood glucose value y of collection obtains prediction error e:
Wherein yiIt is the sampled value of i-th of moment continuous blood sugar monitoring device,It is i-th of moment blood glucose prediction value, eiIt is i-th
The prediction error of a moment blood glucose, N are hits, thus to obtain N number of prediction error E={ e1,e2,…,eN}；
(2) distribution situation of blood glucose prediction error is estimated by the method based on gauss hybrid models；
(3) it is distributed and is estimated according to the prediction error of step (2), carry out the meter of the high hypoglycemia alarm probabilities of online blood glucose sampled point
It calculates, specifically includes following sub-step:
(3.1) the median point of prediction probability of error density function is calculated:
Wherein P (e | θ) is the prediction probability of error density function estimated in step (2), and θ is to be estimated in gauss hybrid models
Parameter, i.e. θ=(μ, Σ), μ are the mean values of each list Gauss model in gauss hybrid models, and Σ is each in gauss hybrid models
The covariance matrix of single Gauss model, C are the median points of probability density function；
(3.2) confidence interval of prediction error is determined according to 95% confidence level:
Wherein σ is confidence radius, and c- σ is confidence lower limit, and c+ σ is confidence upper limit；
(3.3) the abnormal plasma glucose alarm probabilities of each blood glucose prediction value are calculated:
(3.3.1) is if blood glucose prediction value is higher than hyperglycemia alarm threshold value plus the confidence lower limit of prediction error or blood glucose is pre-
Measured value is lower than hypoglycemia alarm threshold value plus the confidence upper limit of prediction error, then alarm probabilities are 1, that is, generates alarm；
(3.3.2) is if high hypoglycemia alarm threshold value in blood glucose prediction value confidence interval, calculates the height blood of the predicted value
Sugared alarm probabilities define alarm distance d according to the following formulalowAnd dhigh:
Wherein HlowIt is hypoglycemia alarm threshold value, HhighIt is hyperglycemia alarm threshold value, dlowIt is predicted value and hypoglycemia alarm threshold value
Distance, dhighIt is predicted value at a distance from hyperglycemia alarm threshold value；Then the probability of high hypoglycemia is calculated according to following formula:
It is wherein PlowHypoglycemia alarm probabilities, PhighIt is hyperglycemia alarm probabilities；
(4) the new prediction error update gauss hybrid models parameter obtained according to new blood glucose sampled value.
2. a kind of continuous blood sugar monitoring device comprising abnormal plasma glucose probability alarm, feature exist according to claim 1
In the step (2) specifically includes following sub-step:
(2.1) the prediction error E={ e obtained in estimating step (1)1,e2,…,eNDistribution, it is assumed that predict the probability of error
Density function is made of K single Gauss model, the probability density function of gauss hybrid models are as follows:
Wherein P (e | θ) is prediction probability of error density function, αkIt is the weight of k-th of single Gauss model, and φ (e | θk) it is k-th
The probability density function of single Gauss model meets:
Wherein μkIt is the mean value of k-th of single Gauss model, ΣkIt is the covariance matrix of k-th of single Gauss model, θkIt is single k-th
Gauss model parameter set, i.e. θk=(μk,Σk), n is the dimension of e, n=1；
(2.2) using the parameter of EM algorithm estimation gauss hybrid models, the specific steps are as follows:
(2.2.1) is for αk, μkAnd ΣkAssign initial value, k=1,2 ..., K；
(2.2.2) calculates posterior probability
WhereinIt is j-th of prediction error e in the s times iterationjBelong to the posterior probability of k-th of single Gauss model；
(2.2.3) is calculatedWithThe s+1 times iterative value:
WhereinWithIt is α respectivelyk, μkAnd ΣkThe s+1 times iterative value.
3. a kind of continuous blood sugar monitoring device comprising abnormal plasma glucose probability alarm, feature exist according to claim 2
In the step (4) specifically includes following sub-step:
(4.1) the prediction error e for newly obtainingN+1, e is calculated separately according to formula (7)N+1Corresponding each single Gauss model it is general
Rate density function φ (eN+1|θk)；
(4.2) single Gauss model M corresponding to the wherein maximum value of probability density function is found outk, judge eN+1Whether in Dan Gaosi
In the confidence interval of model 95%, if in 95% confidence interval of the list Gauss model, list Gauss model MkIt is
eN+1Matching list Gauss model, using more new algorithm update gauss hybrid models parameter；If continuous, there are three predict error
Not in 95% confidence interval of the corresponding single Gauss model of the maximum value of probability density function, then it is assumed that gauss hybrid models
Structure change, re-start gauss hybrid models parameter Estimation according to formula (8)-(11)；
(4.3) gauss hybrid models parameter more new algorithm is as follows:
ρ=λ * φ (eN+1|θk) (16)
Wherein, p (Mi|eN+1) it is identifier, only predict error eN+1Matched list Gauss model MkIts identifier p (Mk|eN+1)
It is just 1, the identifier of other list Gauss models is 0；It is the weight of i-th of the N+1 moment single Gauss model, only matches
Single Gauss model MkWeight increase, it is other list Gauss models weight decline；WithIt is N+1 moment matched list
Gauss model MkMean value and covariance matrix, it is other list Gauss models mean value and covariance matrix without update；φ
(eN+1|θk) it is eN+1The probability density function of corresponding each single Gauss model；λ is adjustable parameter.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710262105.8A CN107137093B (en) | 2017-04-20 | 2017-04-20 | A kind of continuous blood sugar monitoring device comprising abnormal plasma glucose probability alarm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710262105.8A CN107137093B (en) | 2017-04-20 | 2017-04-20 | A kind of continuous blood sugar monitoring device comprising abnormal plasma glucose probability alarm |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107137093A CN107137093A (en) | 2017-09-08 |
| CN107137093B true CN107137093B (en) | 2019-06-07 |
Family
ID=59774486
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710262105.8A Active CN107137093B (en) | 2017-04-20 | 2017-04-20 | A kind of continuous blood sugar monitoring device comprising abnormal plasma glucose probability alarm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107137093B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200043606A1 (en) * | 2018-08-02 | 2020-02-06 | Cnoga Medical Ltd. | System and method for controlling blood glucose using personalized histograms |
| DE102018121349A1 (en) * | 2018-08-31 | 2020-03-05 | B. Braun Avitum Ag | Self-learning input filter for medical devices |
| CN109682976B (en) * | 2019-02-28 | 2020-05-22 | 东北大学 | Continuous blood glucose monitoring sensor online fault detection method based on multi-model fusion |
| CN111991003A (en) * | 2020-08-12 | 2020-11-27 | 上海萌草科技有限公司 | Savitzky-Golay filtering-based continuous blood glucose smoothing method, device, equipment and storage medium |
| CN114464291B (en) * | 2021-12-22 | 2022-11-25 | 北京理工大学 | MDI dosage suggestion system based on Bayesian optimization |
| CN115670448B (en) * | 2022-10-13 | 2023-09-19 | 安徽医科大学第二附属医院 | Continuous blood glucose monitoring equipment and system |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1644163A (en) * | 2003-08-28 | 2005-07-27 | 生命扫描有限公司 | Analytical device with prediction module and related methods |
| CN101257840A (en) * | 2005-09-09 | 2008-09-03 | 霍夫曼-拉罗奇有限公司 | A system, tools, devices and a program for diabetes care |
| CN103310113A (en) * | 2013-06-24 | 2013-09-18 | 浙江大学 | Universal blood glucose prediction method based on frequency band separation and data modeling |
| CN103400028A (en) * | 2008-04-04 | 2013-11-20 | 海吉雅有限公司 | Device for optimizing patient's insulin dosage regimen |
| CN104814749A (en) * | 2014-01-31 | 2015-08-05 | 精工爱普生株式会社 | Blood sugar level measuring device and blood sugar level measuring method |
| CN104921736A (en) * | 2015-06-08 | 2015-09-23 | 浙江大学 | Continuous blood glucose monitoring device comprising parameter estimation function filtering module |
| CN105403705A (en) * | 2015-12-05 | 2016-03-16 | 浙江大学 | Continuous blood glucose monitoring equipment comprising blood glucose classification function fault detection module |
| WO2017035024A1 (en) * | 2015-08-21 | 2017-03-02 | Medtronic Minimed, Inc. | Data analytics and insight delivery for the management and control of diabetes |
| WO2017061943A1 (en) * | 2015-10-09 | 2017-04-13 | Dianovator Ab | Medical arrangements and a method for determining parameters related to insulin therapy, predicting glucose values and for providing insulin dosing recommendations |
-
2017
- 2017-04-20 CN CN201710262105.8A patent/CN107137093B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1644163A (en) * | 2003-08-28 | 2005-07-27 | 生命扫描有限公司 | Analytical device with prediction module and related methods |
| CN101257840A (en) * | 2005-09-09 | 2008-09-03 | 霍夫曼-拉罗奇有限公司 | A system, tools, devices and a program for diabetes care |
| CN103400028A (en) * | 2008-04-04 | 2013-11-20 | 海吉雅有限公司 | Device for optimizing patient's insulin dosage regimen |
| CN103310113A (en) * | 2013-06-24 | 2013-09-18 | 浙江大学 | Universal blood glucose prediction method based on frequency band separation and data modeling |
| CN104814749A (en) * | 2014-01-31 | 2015-08-05 | 精工爱普生株式会社 | Blood sugar level measuring device and blood sugar level measuring method |
| CN104921736A (en) * | 2015-06-08 | 2015-09-23 | 浙江大学 | Continuous blood glucose monitoring device comprising parameter estimation function filtering module |
| WO2017035024A1 (en) * | 2015-08-21 | 2017-03-02 | Medtronic Minimed, Inc. | Data analytics and insight delivery for the management and control of diabetes |
| WO2017061943A1 (en) * | 2015-10-09 | 2017-04-13 | Dianovator Ab | Medical arrangements and a method for determining parameters related to insulin therapy, predicting glucose values and for providing insulin dosing recommendations |
| CN105403705A (en) * | 2015-12-05 | 2016-03-16 | 浙江大学 | Continuous blood glucose monitoring equipment comprising blood glucose classification function fault detection module |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107137093A (en) | 2017-09-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107137093B (en) | A kind of continuous blood sugar monitoring device comprising abnormal plasma glucose probability alarm | |
| CN104921736A (en) | Continuous blood glucose monitoring device comprising parameter estimation function filtering module | |
| CN105962906B (en) | A kind of body temperature measurement method and device | |
| Tan et al. | Non-invasive continuous blood pressure measurement based on mean impact value method, BP neural network, and genetic algorithm | |
| CN104994535B (en) | Sensor data stream method for detecting abnormality based on Multidimensional Data Model | |
| CN103916896B (en) | Anomaly detection method based on multi-dimensional Epanechnikov kernel density estimation | |
| CN105595990A (en) | Intelligent terminal device for evaluating and distinguishing quality of electrocardiosignal | |
| CN104545870A (en) | Heart rate detection mattress and method | |
| CN105243285A (en) | Big data health forecast system | |
| CN105403705B (en) | Continuous blood glucose monitoring equipment comprising blood glucose classification function fault detection module | |
| CN104462744A (en) | Data quality control method suitable for cardiovascular remote monitoring system | |
| CN106446765A (en) | Health state evaluation system based on multidimensional physiological big data depth learning | |
| CN109564586A (en) | System monitor and system monitoring method | |
| CN110245395A (en) | Monitoring method, system and the medium of cylindrical shape error with spatial coherence | |
| CN107019501B (en) | Remote tumble detection method and system based on genetic algorithm and probabilistic neural network | |
| Shen et al. | Automatic recognition method of cow ruminating behaviour based on edge computing | |
| CN110379503B (en) | Online fault detection and diagnosis system based on continuous blood glucose monitoring system | |
| CN107977730A (en) | A kind of wind measurement method of multisensor Data Fusion technology | |
| CN113842135A (en) | BilSTM-based sleep breathing abnormality automatic screening method | |
| KR20140016559A (en) | Apparatus and method for monitoring disease status based on time-series data modeling | |
| CN111278353A (en) | Method and system for detecting vital sign signal noise | |
| CN116712099A (en) | Fetal heart state detection method based on multi-mode data, electronic equipment and storage medium | |
| CN114468996B (en) | Method for analyzing breast signs based on order, multi-mode and symmetry deficiency | |
| CN114903475A (en) | Health detector based on PCANet neural network and multi-sensor information fusion and method thereof | |
| CN112071434B (en) | Abnormal body temperature sequence detection method |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |