CN106709225A - Early-warning method, device and system for diabetic retinopathy - Google Patents
Early-warning method, device and system for diabetic retinopathy Download PDFInfo
- Publication number
- CN106709225A CN106709225A CN201510770592.XA CN201510770592A CN106709225A CN 106709225 A CN106709225 A CN 106709225A CN 201510770592 A CN201510770592 A CN 201510770592A CN 106709225 A CN106709225 A CN 106709225A
- Authority
- CN
- China
- Prior art keywords
- patient
- data
- pvr
- diabetes
- time
- 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
Landscapes
- Investigating Or Analysing Biological Materials (AREA)
Abstract
The invention provides an early-warning method, device and system for diabetic retinopathy. The method includes the steps that diabetes historical data of sample patients of a preset number are obtained; the diabetes historical data of sample patients of the preset number are preprocessed to obtain medical characteristic data; the incidence relation between the medical characteristic data and retinopathy is set up; the early-warning information whether a patient not having retinopathy suffers from retinopathy or not is obtained according to the incidence relation. The diabetes historical data of sample patients are preprocessed, the incidence relation with occurrence of retinopathy is set up, the early-warning information of retinopathy is fed back to the diabetes patient, the diabetes patient is reminded to see a doctor in time, and the possibility of illness state exacerbation is reduced or avoided.
Description
Technical field
The present invention relates to mobile health field, more particularly to a kind of method for early warning of diabetic retinopathy,
Apparatus and system.
Background technology
Diabetic retinopathy is a series of caused allusion quotations of retinal microvascular infringement caused by diabetes
Type lesion.In China, illness rate of the PVR in people with diabetes is 24.7%-37.5%.
According to《China's diabetic retinopathy clinic diagnosis guide (2014)》, blood sugar,
Blood pressure, blood fat are 3 important risk factors that PVR occurs, and diabetic duration is most heavy
The occurrence factor wanted, the blood sugar level of diabetic, glycosylated hemoglobin (HbA1c) concentration
Level has direct relation with PVR.
In early days, patient may be asymptomatic, and eyesight is unaffected for PVR, but with disease development, depending on
Power declines even blinds.Early detection, in time correct treatment are very crucial.At present, lack a kind of personalized
PVR method for early warning, be only limitted to remind diabetic that (such as every half a year once) goes to hospital on time
See a doctor, check PVR situation.If patient compliance is poor, could not go to a doctor on time, then may
Appearance state of an illness discovery is later, delay the adverse consequences such as treatment.
The content of the invention
It is an object of the invention to provide a kind of method for early warning of diabetic retinopathy, apparatus and system,
To diabetic feed back PVR early warning information, remind diabetic go to a doctor in time, reduce or
Avoid the possibility that sb.'s illness took a turn for the worse.
To achieve these goals, the pre- police of a kind of diabetic retinopathy provided in an embodiment of the present invention
Method, including:
Obtain the diabetes historical data of the sample patient of predetermined quantity;
Diabetes historical data to the sample patient of predetermined quantity is pre-processed, and obtains medical characteristics data;
Set up the incidence relation that the medical characteristics data occur with PVR;
According to the incidence relation, whether the patient for not occurred PVR there is retinopathy
The early warning information of change.
Wherein, the diabetes historical data includes:Diabetes diagnosis time, dynamic retinoscopy history and blood
The Monitoring Data of pressure, blood sugar, Glycohemoglobin HbA1c and blood fat.
Wherein, the medical characteristics data include:Diabetic duration, dysarteriotony ratio, pathoglycemia ratio
Example, HbA1c unnatural proportions and dyslipidemia ratio;The unnatural proportions control mesh by the personalized of patient
Mark determines.
Wherein, the diabetes historical data to the patient of predetermined quantity is pre-processed, and obtains medical characteristics number
According to the step of, including:
The patient categories of the sample patient are determined according to the dynamic retinoscopy history;
The diabetes historical data is pre-processed according to the patient categories, obtains medical characteristics data.
Wherein, the step of patient categories of the sample patient being determined according to the dynamic retinoscopy history, bag
Include:
All the time the sample patient for not occurring PVR in the dynamic retinoscopy history is defined as
One class patient;The sample patient that PVR occurs in midway is defined as Equations of The Second Kind patient.
Wherein, the diabetes historical data is pre-processed according to the patient categories, obtains medical treatment special
The step of levying data, including:
The time checked for the last time in dynamic retinoscopy history according to the first kind patient, it is determined that described
The diabetic duration of first kind patient;
Obtain in the dynamic retinoscopy history it is last check twice between interval time in all of blood pressure,
The Monitoring Data of blood sugar, HbA1c and blood fat, determines the unnatural proportions of the Monitoring Data.
Wherein, the diabetes historical data is pre-processed according to the patient categories, obtains medical treatment special
The step of levying data, also includes:
Checked for the first time in dynamic retinoscopy history according to the Equations of The Second Kind patient PVR when
Between, determine the diabetic duration of the Equations of The Second Kind patient;
Obtain check for the last time in the dynamic retinoscopy history without lesion with check lesion for the first time
Between time interval in all of blood pressure, blood sugar, the Monitoring Data of HbA1c and blood fat, determine institute
State the unnatural proportions of Monitoring Data.
Wherein, the step of setting up the incidence relation that the medical characteristics data occur with PVR, including:
The medical characteristics data are obtained into disaggregated model parameter by a support vector machines algorithm;
The medical characteristics data and associating that PVR occurs are set up according to the disaggregated model parameter
The svm classifier model of relation.
Wherein, according to the incidence relation, whether the patient for not occurred PVR regards
The step of early warning information of retinopathy, including:
By in the medical characteristics data input to the svm classifier model, do not occurred described in generation
Whether the patient of PVR there is the early warning and alert result of PVR.
The embodiment of the present invention also provides a kind of prior-warning device of diabetic retinopathy, including:
Acquisition module, the diabetes historical data of the sample patient for obtaining predetermined quantity;
Data preprocessing module, the diabetes historical data for the sample patient to predetermined quantity carries out pre- place
Reason, obtains medical characteristics data;
Relation sets up module, for setting up the incidence relation that the medical characteristics data occur with PVR;
Lesion warning module, for according to the incidence relation, not occurred the trouble of PVR
Whether person there is the early warning information of PVR.
Wherein, the diabetes historical data includes:Diabetes diagnosis time, dynamic retinoscopy history and blood
The Monitoring Data of pressure, blood sugar, Glycohemoglobin HbA1c and blood fat.
Wherein, the medical characteristics data include:Diabetic duration, dysarteriotony ratio, pathoglycemia ratio
Example, HbA1c unnatural proportions and dyslipidemia ratio;The unnatural proportions control mesh by the personalized of patient
Mark determines.
Wherein, the data preprocessing module includes:
Classification determination sub-module, the patient for determining the sample patient according to the dynamic retinoscopy history
Classification;
Data prediction submodule, it is pre- for being carried out to the diabetes historical data according to the patient categories
Treatment, obtains medical characteristics data.
Wherein, the classification determination sub-module includes:
Classification determination unit, for PVR will not all the time occur in the dynamic retinoscopy history
Sample patient is defined as first kind patient;The sample patient that PVR occurs in midway is defined as Equations of The Second Kind trouble
Person.
Wherein, the data prediction submodule includes:
First data determination unit, in the dynamic retinoscopy history according to the first kind patient last
The time of secondary inspection, determine the diabetic duration of the first kind patient;
First data acquisition process unit, finally check it twice in the dynamic retinoscopy history for obtaining
Between interval time in all of blood pressure, blood sugar, HbA1c and blood fat Monitoring Data, determine the prison
Survey the unnatural proportions of data.
Wherein, the data prediction submodule also includes:
Second data determination unit, for first time in the dynamic retinoscopy history according to the Equations of The Second Kind patient
The time of PVR is checked, the diabetic duration of the Equations of The Second Kind patient is determined;
Second data acquisition process unit, check for the last time in the dynamic retinoscopy history for obtaining
All of blood pressure, blood sugar, the HbA1c in the time interval between lesion are checked without lesion and first time
And the Monitoring Data of blood fat, determine the unnatural proportions of the Monitoring Data.
Wherein, the relation is set up module and is included:
Model parameter determination sub-module, for the medical characteristics data to be passed through into a support vector machines
Algorithm obtains disaggregated model parameter;
Disaggregated model sets up module, for according to the disaggregated model parameter set up the medical characteristics data with
The svm classifier model of the incidence relation that PVR occurs.
Wherein, the lesion warning module includes:
Early warning generates submodule, for by the medical characteristics data input to the svm classifier model,
Whether the patient for not occurred PVR described in generation there is the early warning and alert result of PVR.
The embodiment of the present invention also provides a kind of early warning system of diabetic retinopathy, including diabetes backstage
Self-management system and mobile terminal;Wherein,
The diabetes backstage self-management system includes the diabetic retinopathy as described in above-mentioned embodiment
The prior-warning device of change;
The mobile terminal is used to gather the diabetes historical data of patient, receives diabetes backstage self-management
The early warning information of the PVR that system is returned.
Above-mentioned technical proposal of the invention has the beneficial effect that:
In the scheme of the embodiment of the present invention, by the pretreatment of the diabetes historical data to sample patient, obtain
To medical characteristics data, the incidence relation that the medical characteristics data occur with PVR is set up, so as to
Diabetic feeds back the early warning information of PVR, reminds diabetic to go to a doctor in time, reduces or keeps away
Exempt from the possibility that sb.'s illness took a turn for the worse.
Brief description of the drawings
Fig. 1 is the basic step schematic diagram of the method for early warning of the diabetic retinopathy of the embodiment of the present invention;
Fig. 2 is the composition structural representation of the prior-warning device of the diabetic retinopathy of the embodiment of the present invention;
Fig. 3 is the basic composition frame chart of the early warning system of the diabetic retinopathy of the embodiment of the present invention.
Specific embodiment
To make the technical problem to be solved in the present invention, technical scheme and advantage clearer, below in conjunction with attached
Figure and specific embodiment are described in detail.
The present invention fails to be found in time, causes to suffer from for diabetic complication PVR in the prior art
Person fails to go to a doctor in time, delays the problem for the treatment of, there is provided a kind of method for early warning of diabetic retinopathy,
Apparatus and system, by the pretreatment of the diabetes historical data to sample patient, sets up and PVR
The incidence relation of generation, so as to feed back the early warning information of PVR to diabetic, reminds diabetes
Patient goes to a doctor in time, lowers or avoid the possibility that sb.'s illness took a turn for the worse.
First embodiment
As shown in figure 1, the embodiment of the present invention provides a kind of method for early warning of diabetic retinopathy, including:
Step 11, obtains the diabetes historical data of the sample patient of predetermined quantity;
Here, the diabetes historical data of the sample patient of predetermined quantity is diabetes backstage self-management system
The diabetes historical data of a large amount of patients stored in database, the quantity of its patient can flexibly be set by system.
Step 12, the diabetes historical data to the sample patient of predetermined quantity is pre-processed, and obtains doctor
Treat characteristic;
Step 13, sets up the incidence relation that the medical characteristics data occur with PVR;
Step 14, according to the incidence relation, whether the patient for not occurred PVR sends out
The early warning information of raw PVR.
In the scheme of the embodiment of the present invention, by the pretreatment of the diabetes historical data to sample patient, obtain
To medical characteristics data, the incidence relation that the medical characteristics data occur with PVR is set up, so as to
Diabetic feeds back the early warning information of PVR, reminds diabetic to go to a doctor in time, reduces or keeps away
Exempt from the possibility that sb.'s illness took a turn for the worse.
Specifically, diabetes historical data described in the embodiment of the present invention include:The diabetes diagnosis time, depending on
Nethike embrane checks the Monitoring Data of history and blood pressure, blood sugar, Glycohemoglobin HbA1c and blood fat.
Here it should be noted that, above-mentioned six diabetes historical datas are most direct with having for PVR
Relation, be effective foundation of early warning PVR.
Specifically, medical characteristics data described in the embodiment of the present invention include:Diabetic duration, dysarteriotony
Ratio, pathoglycemia ratio, HbA1c unnatural proportions and dyslipidemia ratio;The unnatural proportions are by suffering from
The personalized control targe of person determines.
Preferably, step 12 described in the embodiment of the present invention can be further included:
Step 121, the patient categories of the sample patient are determined according to the dynamic retinoscopy history;
Step 122, pre-processes according to the patient categories to the diabetes historical data, obtains doctor
Treat characteristic.
Specifically, the step 121 can be further included:
Step 1211, the sample for not occurring PVR in the dynamic retinoscopy history all the time is suffered from
Person is defined as first kind patient;The sample patient that PVR occurs in midway is defined as Equations of The Second Kind patient.
Specifically, the step 122 can be further included:
Step 1221, checked for the last time in the dynamic retinoscopy history according to the first kind patient when
Between, determine the diabetic duration of the first kind patient;
Here, the diabetic duration of first kind patient be check for the last time in its dynamic retinoscopy history when
The spacing diabetes mellitus make a definite diagnosis the duration of time.
Step 1222, in the interval time in the acquisition dynamic retinoscopy history between last inspection twice
The Monitoring Data of all of blood pressure, blood sugar, HbA1c and blood fat, determines the anomaly ratio of the Monitoring Data
Example.
In this step, the Monitoring Data in the interval time between last inspection twice is selected because the time gets over
The late, course of disease is more long, it is bigger that the probability of PVR occurs, so closer to first kind patient and Equations of The Second Kind
The distinguishing limit of patient, the accurate foundation for association relation model is also more favourable.
Specifically, the step 122 can also be further included:
Step 1223, view is checked in the dynamic retinoscopy history according to the Equations of The Second Kind patient for the first time
The time of film lesion, determine the diabetic duration of the Equations of The Second Kind patient;
Here, the diabetic duration of Equations of The Second Kind patient be its dynamic retinoscopy history in check view for the first time
The time interval of the film lesion diabetes mellitus make a definite diagnosis the duration of time.
Step 1224, checked without lesion and first time for the last time in the acquisition dynamic retinoscopy history
Check the monitoring number of all of blood pressure in the time interval between lesion, blood sugar, HbA1c and blood fat
According to determining the unnatural proportions of the Monitoring Data.
In this step, selection last time is checked without between lesion and the time for checking between lesion for the first time
Every interior Monitoring Data because need observation from it is disease-free change to occur patient between lesion to blood pressure,
The control situation of the Monitoring Data of blood sugar, HbA1c and blood fat, these directly influence the generation of lesion.
Preferably, step 13 described in the embodiment of the present invention can be further included:
The medical characteristics data are obtained mould of classifying by step 131 by a support vector machines algorithm
Shape parameter;
Step 132, sets up the medical characteristics data and is sent out with PVR according to the disaggregated model parameter
The svm classifier model of raw incidence relation.
Here, SVM algorithm is a sorting algorithm of the prior art, by seeking structuring least risk
To improve learning machine generalization ability, the minimum of empiric risk and fiducial range is realized, so as to reach in statistics
In the case that sample size is less, the purpose of good statistical law can be also obtained.
Citing simple declaration below determines the handling process of svm classifier model using SVM algorithm.Specifically
The step of it is as follows:
S01, to medical characteristics data normalization;
Here it should be noted that, normalization of this step to medical characteristics data be specially to diabetic duration,
The normalization of dysarteriotony ratio, pathoglycemia ratio, HbA1c unnatural proportions and dyslipidemia ratio.
Here, dysarteriotony ratio, pathoglycemia ratio, HbA1c unnatural proportions and dyslipidemia ratio,
Itself is ratio value, in the interval in 0-1, therefore without normalization;
Mainly to diabetic duration, the parameter is normalized in this step, specially:
Maximum diabetic duration T in record sample patientmaxWith minimum diabetic duration Tmin, then glycosuria
The normalized computing formula of the course of disease is:T '=(T-Tmin)/(Tmax-Tmin)。
S02, reduces normalized diabetic duration, dysarteriotony ratio, pathoglycemia ratio, HbA1c
The dimension of five dimensional vectors of unnatural proportions and dyslipidemia ratio composition;
Here, although having very between the parameter such as only five parameters, blood sugar, HbA1c after normalization
Big correlation, data redundancy can be reduced by reducing dimension, improve efficiency.
Principal component analysis PCA algorithms can be taken in this step, five dimensional vector is mapped as smaller dimension
Vector, and preserve covariance matrix C and dimension N, can make after the input of new patient medical characteristic
With.
S03, the grid optimizing of disaggregated model parameter;
Here, because the RBF RBF kernel functions used in the svm classifier model are related to c, γ
Two parameters, by cross validation, it is determined that optimal c and γ so that svm classifier model can be correct
Prediction unknown data (the diabetes historical data of i.e. new patient), there is classification accurate rate higher.
S04, svm classifier model is set up.
Here specifically, carrying out two classification using RBF kernel functions, optimal c and γ values are substituted into,
Probability Estimation is set simultaneously, it is final to determine disaggregated model parameter model, complete the foundation of disaggregated model.
Preferably, step 14 described in the embodiment of the present invention can be further included:
Step 141, by the medical characteristics data input to the svm classifier model, generation is described
Whether the patient for not occurred lesion there is the early warning and alert result of PVR.
Here, the medical characteristics data are that the patient for not occurred lesion is gathered by mobile terminal
Diabetes historical data be sent in diabetes backstage self-management system, data prediction is obtained.
Be may include in the early warning and alert object information:Patient categories belonging to patient and suffer from the general of PVR
Rate.
In the scheme of the embodiment of the present invention, by the pretreatment of the diabetes historical data to sample patient, obtain
To medical characteristics data, the incidence relation that the medical characteristics data occur with PVR is set up, so as to
Diabetic feeds back the early warning information of PVR, reminds diabetic to go to a doctor in time, reduces or keeps away
Exempt from the possibility that sb.'s illness took a turn for the worse.
Second embodiment
As shown in Fig. 2 the embodiment of the present invention provides a kind of prior-warning device of diabetic retinopathy, including:
Acquisition module 21, the diabetes historical data of the sample patient for obtaining predetermined quantity;
Here, the diabetes historical data of the sample patient of predetermined quantity is diabetes backstage self-management system
The diabetes historical data of a large amount of patients stored in database, the quantity of its patient can flexibly be set by system.
Data preprocessing module 22, the diabetes historical data for the sample patient to predetermined quantity is carried out
Pretreatment, obtains medical characteristics data;
Relation sets up module 23, for setting up the medical characteristics data and associating that PVR occurs
Relation;
Lesion warning module 24, for according to the incidence relation, not occurred PVR
Patient whether there is the early warning information of PVR.
Specifically, diabetes historical data described in the embodiment of the present invention include:The diabetes diagnosis time, depending on
Nethike embrane checks the Monitoring Data of history and blood pressure, blood sugar, Glycohemoglobin HbA1c and blood fat.
Here it should be noted that, above-mentioned six diabetes historical datas are most direct with having for PVR
Relation, be effective foundation of early warning PVR.
Specifically, medical characteristics data described in the embodiment of the present invention include:Diabetic duration, dysarteriotony
Ratio, pathoglycemia ratio, HbA1c unnatural proportions and dyslipidemia ratio;The unnatural proportions are by suffering from
The personalized control targe of person determines.
Data preprocessing module 22 described in the embodiment of the present invention may particularly include:
Classification determination sub-module, the patient for determining the sample patient according to the dynamic retinoscopy history
Classification;
Data prediction submodule, it is pre- for being carried out to the diabetes historical data according to the patient categories
Treatment, obtains medical characteristics data.
Specifically, the classification determination sub-module may include:
Classification determination unit, for PVR will not all the time occur in the dynamic retinoscopy history
Sample patient is defined as first kind patient;The sample patient that PVR occurs in midway is defined as Equations of The Second Kind trouble
Person.
Specifically, the data prediction submodule may include:
First data determination unit, in the dynamic retinoscopy history according to the first kind patient last
The time of secondary inspection, determine the diabetic duration of the first kind patient;
Here, the diabetic duration of first kind patient be check for the last time in its dynamic retinoscopy history when
The spacing diabetes mellitus make a definite diagnosis the duration of time.
First data acquisition process unit, finally check it twice in the dynamic retinoscopy history for obtaining
Between interval time in all of blood pressure, blood sugar, HbA1c and blood fat Monitoring Data, determine the prison
Survey the unnatural proportions of data.
In this unit, the Monitoring Data in the interval time between last inspection twice is selected because the time gets over
The late, course of disease is more long, it is bigger that the probability of PVR occurs, so closer to first kind patient and Equations of The Second Kind
The distinguishing limit of patient, the accurate foundation for association relation model is also more favourable.
The data prediction submodule may also include:
Second data determination unit, for first time in the dynamic retinoscopy history according to the Equations of The Second Kind patient
The time of PVR is checked, the diabetic duration of the Equations of The Second Kind patient is determined;
Here, the diabetic duration of Equations of The Second Kind patient be its dynamic retinoscopy history in check view for the first time
The time interval of the film lesion diabetes mellitus make a definite diagnosis the duration of time.
Second data acquisition process unit, check for the last time in the dynamic retinoscopy history for obtaining
All of blood pressure, blood sugar, the HbA1c in the time interval between lesion are checked without lesion and first time
And the Monitoring Data of blood fat, determine the unnatural proportions of the Monitoring Data.
In this unit, selection last time is checked without between lesion and the time for checking between lesion for the first time
Every interior Monitoring Data because need observation from it is disease-free change to occur patient between lesion to blood pressure,
The control situation of the Monitoring Data of blood sugar, HbA1c and blood fat, these directly influence the generation of lesion.
Relation described in the embodiment of the present invention is set up module 23 and be may particularly include:
Model parameter determination sub-module, for the medical characteristics data to be passed through into a support vector machines
Algorithm obtains disaggregated model parameter;
Disaggregated model sets up module, for according to the disaggregated model parameter set up the medical characteristics data with
The svm classifier model of the incidence relation that PVR occurs.
Here, SVM algorithm is a sorting algorithm of the prior art, by seeking structuring least risk
To improve learning machine generalization ability, the minimum of empiric risk and fiducial range is realized, so as to reach in statistics
In the case that sample size is less, the purpose of good statistical law can be also obtained.
The process that the SVM algorithm is the foundation for realizing svm classifier model is applied in the embodiment of the present invention,
Determined using SVM algorithm in first embodiment svm classifier model handling process citing in it is specific
Illustrate, repeat no more here.
Specifically, the lesion warning module 24 of the embodiment of the present invention may include:
Early warning generates submodule, for by the medical characteristics data input to the svm classifier model,
Whether the patient for not occurred PVR described in generation there is the early warning and alert result of PVR.
Here, the medical characteristics data are that the patient for not occurred lesion is gathered by mobile terminal
Diabetes historical data be sent in diabetes backstage self-management system, data prediction is obtained.
Be may include in the early warning and alert object information:Patient categories belonging to patient and suffer from the general of PVR
Rate.
The embodiment of the present invention also provides a kind of early warning system of diabetic retinopathy, including:After diabetes
Platform self-management system and mobile terminal;Wherein,
The diabetes backstage self-management system includes the diabetic retinopathy as described in above-mentioned embodiment
The prior-warning device of change;
The mobile terminal is used to gather the diabetes historical data of patient, receives diabetes backstage self-management
The early warning information of the PVR that system is returned.
Here, the diabetes historical data of patient is gathered, regarding for urine platform self-management after being ill system return is received
The early warning information of retinopathy can be by installing self-management in diabetes application software come real on mobile terminals
Existing, certainly, other can realize that the method for this function is applicable.
In the scheme of the embodiment of the present invention, by the pretreatment of the diabetes historical data to sample patient, obtain
To medical characteristics data, the incidence relation that the medical characteristics data occur with PVR is set up, so as to
Diabetic feeds back the early warning information of PVR, reminds diabetic to go to a doctor in time, reduces or keeps away
Exempt from the possibility that sb.'s illness took a turn for the worse.
3rd embodiment
As shown in figure 3, the basic composition frame chart of the early warning system of diabetic retinopathy.Should scheme below
And combine the implementation process that a specific embodiment describes the method for early warning of diabetic retinopathy in detail.Specifically
It is as follows:
S ' 01, the sugar comprising great amount of samples patient in the database in diabetes backstage self-management system
The sick historical data of urine;The sample patient in database is classified according to diabetes historical data;And
Diabetes historical data to sorted patient is pre-processed, and obtains medical characteristics data.
It is illustrated below:
Time (2004-10-1), glycemic target, blood that the data of patient A make a definite diagnosis diabetes including it
Sign data and the inspection of pressure control targe, HbA1c control targes, blood fat control targe and time series
Data are looked into, as shown in table 1:
Table 1
Date | Blood sugar | Blood pressure | HbA1c | Blood fat | PVR inspection result |
2015-1-1 | 7 | 130/90 | 7 | 5.5 | Nothing |
... | ... | ... | ... | ... | ... |
2015-3-1 | 8 | 130/90 | 8 | 6 | Nothing |
2015-3-2 | 7 | 140/94 | - | - | - |
2015-3-3 | 7.4 | 135/90 | - | - | - |
... | ... | ... | ... | ... | ... |
2015-6-1 | 7.2 | 135/95 | 7.5 | 6.2 | Nothing |
In all previous PVR inspection results of patient A, do not occur PVR, therefore be marked as
First kind patient.
Diabetes historical data to patient A is pre-processed, and first parameter is last time retina
Diabetic duration during inspection, is 11.6;Second parameter be 2015-3-1 to 2015-6-1 between,
Abnormal data proportion in blood glucose level data, is 25%;3rd parameter is abnormal data institute in blood pressure data
Accounting example, is 20%;Fourth, fifth parameter is similar to, and respectively 15%, 20%.Thus, by patient A
Medical characteristics data be expressed as:11.6,0.25,0.20,0.15,0.20, it is labeled as the first kind.
Time (2000-4-1), glycemic target, blood that the data of patient B make a definite diagnosis diabetes including it
Sign data and the inspection of pressure control targe, HbA1c control targes, blood fat control targe and time series
Data are looked into, as shown in table 2:
Table 2
Date | Blood sugar | Blood pressure | HbA1c | Blood fat | PVR inspection result |
2014-12-1 | 8 | 130/90 | 7 | 5 | Nothing |
... | ... | ... | ... | ... | ... |
2015-1-1 | 8 | 130/90 | 7.5 | 5.5 | Nothing |
2015-1-2 | 8.7 | 150/100 | - | - | - |
2015-1-3 | 8.4 | 155/110 | - | - | - |
... | ... | ... | ... | ... | ... |
2015-4-1 | 8.7 | 155/105 | 8.5 | 6.5 | Have |
In patient's B PVR inspection results, do not occurred PVR once, but occur afterwards
PVR, therefore it is marked as Equations of The Second Kind patient.
Diabetes historical data to patient B is pre-processed, and first parameter checks disease for first time
Diabetic duration during change, is 15 years;Second parameter be 2015-1-1 to 2015-4-1 between, blood sugar
Abnormal data proportion in data, is 70%;3rd parameter is abnormal data institute accounting in blood pressure data
Example, is 50%;Fourth, fifth parameter is similar to, and respectively 75%, 60%.Thus, by patient B's
Medical characteristics data are expressed as:15,0.70,0.50,0.75,0.60, it is labeled as Equations of The Second Kind.
Time (2000-5-1), glycemic target, blood that the data of patient C make a definite diagnosis diabetes including it
Sign data and the inspection of pressure control targe, HbA1c control targes, blood fat control targe and time series
Data are looked into, as shown in table 3:
Table 3
Date | Blood sugar | Blood pressure | HbA1c | Blood fat | PVR inspection result |
2013-6-1 | 7 | 140/100 | 7.2 | 7.2 | Have |
... | ... | ... | ... | ... | ... |
2013-12-1 | 8 | 145/105 | 7.5 | 7 | Have |
... | ... | ... | ... | ... | ... |
In all previous PVR inspection results of patient C, there is PVR always, therefore this is suffered from
Person's data foreclose, not labeled as the first kind or Equations of The Second Kind patient.
After the diabetes historical data pretreatment of all sample patients in pending data storehouse is finished, obtain such as table
Form shown in 4.
Table 4
Parameter 1 | Parameter 2 | Parameter 3 | Parameter 4 | Parameter 5 | Classification |
11.6 | 0.25 | 0.20 | 0.15 | 0.20 | 1 |
15 | 0.70 | 0.50 | 0.75 | 0.60 | 2 |
5 | 0.30 | 0.35 | 0.30 | 0.20 | 1 |
10 | 0.90 | 0.60 | 0.80 | 0.70 | 2 |
... | ... | ... | ... | ... | ... |
Explanation is needed exist for, parameter 1 represents diabetic duration, and parameter 2 represents blood glucose level data anomaly ratio
Example, parameter 3 represents blood pressure data unnatural proportions, and parameter 4 represents HbA1c data exception ratios, parameter
5 represent blood fat data exception ratio.
S ' 02, sets up the svm classifier model of the incidence relation that medical characteristics data occur with PVR;
Here, gone on to say with the example above.Based on the data in above-mentioned table 4, svm classifier mould is set up
Type.Comprise the following steps that:
First, parameter 1 is normalized, maximum is 20, minimum value is 1, is calculated using normalization
Formula:T '=(T-Tmin)/(Tmax-Tmin), then first parameter of patient 1 is updated to:(11.6-1)/
(20-1)=0.56.
Secondly dimension is reduced, tieing up parameter by originally 5 by PCA algorithms is reduced to 3-dimensional;
It should be noted that, the 80% of 5 dimension parameters originally information content can be expressed used here as 3-dimensional.
Finally, find optimal c and γ, and substitute into order cmd=['-c ', num2str (c), '
–g’,num2str(γ),’–b1’]。
Finally, disaggregated model parameter model is determined.
Model=svmtrain (label, data, cmd), wherein label are classification, and data is the three-dimensional after dimensionality reduction
Parameter.
S ' 03, according to svm classifier model, whether the patient for not occurred PVR occurs
The early warning information of PVR.
It is illustrated below:
New patient D, time (2010-10-1) that data make a definite diagnosis diabetes including it, glycemic target,
The sign data of controlling of blood pressure target, HbA1c control targes, blood fat control targe and time series and
Data are checked, as shown in table 5:
Table 5
Date | Blood sugar | Blood pressure | HbA1c | Blood fat | PVR inspection result |
2015-5-1 | 7 | 136/96 | 7 | 6 | Nothing |
2015-5-2 | 6 | 130/85 | - | - | - |
2015-5-3 | 6.4 | 120/80 | - | - | - |
... | ... | ... | ... | ... | ... |
2015-7-1 | 6 | 130/95 | 6.5 | 6 | Nothing |
Diabetes historical data to patient D is pre-processed, and first parameter is the glycosuria of current time
The course of disease, is 4.75;Second parameter be 2015-5-1 to 2015-7-1 between, it is different in blood glucose level data
Regular data proportion, is 10%;Third and fourth, five parameters be similar to, respectively 14%, 12%, 10%.
Thus, the medical characteristics data of patient D are expressed as:4.75,0.10,0.14,0.12,0.10.
Utilize an algorithm to judge that the patient (has closer to first kind patient (without lesion) or Equations of The Second Kind patient
Lesion), so as to decide whether to provide early warning for it.Specially:
Patient D data are further normalized and dimensionality reduction:Diabetic duration=(4.75-1)/(20-1)=0.20,
Hereafter, five dimension parameters are multiplied by covariance matrix and take preceding three-dimensional.Finally, svmpredict functions are substituted into:
[label, acc, esti]=svmpredict (labelc, datac, model, '-b 1 '), wherein labelc is to be arbitrarily designated
Classification, datac be dimensionality reduction after three-dimensional parameter, esti be retinopathy changeable probability.Label is patient D
Category label, be the first kind, illustrate not for the patient provides PVR early warning.
New patient E, time (2000-8-1) that data make a definite diagnosis diabetes including it, glycemic target,
The sign data of controlling of blood pressure target, HbA1c control targes, blood fat control targe and time series and
Data are checked, as shown in table 6:
Table 6
Date | Blood sugar | Blood pressure | HbA1c | Blood fat | PVR inspection result |
2015-3-1 | 8 | 140/100 | 8 | 7 | Nothing |
2015-3-2 | 7 | 150/90 | - | - | - |
2015-3-3 | 7.5 | 140/90 | - | - | - |
... | ... | ... | ... | ... | ... |
2015-8-1 | 7.8 | 150/95 | 7.5 | 7 | Nothing |
Diabetes historical data to patient E is pre-processed, and first parameter is the glycosuria of current time
The course of disease, is 15 years;Second parameter is between 2015-3-1 to 2015-8-1, abnormal in blood glucose level data
Data proportion, is 60%;Third and fourth, five parameters be similar to, respectively 65%, 70%, 75%.
Thus, the medical characteristics data of patient E are expressed as:15,0.60,0.65,0.70,0.75.
Utilize an algorithm to judge that the patient (has closer to first kind patient (without lesion) or Equations of The Second Kind patient
Lesion), so as to decide whether to provide early warning for it.Specially:
Patient E data are further normalized and dimensionality reduction:Diabetic duration=(15-1)/(20-1)=0.74,
Hereafter, five dimension parameters are multiplied by covariance matrix and take preceding three-dimensional.Finally, svmpredict functions are substituted into:
[label, acc, esti]=svmpredict (labelc, datac, model, '-b 1 '), wherein labelc is to be arbitrarily designated
Classification, datac be dimensionality reduction after three-dimensional parameter, esti be retinopathy changeable probability.Label is patient E
Category label, be Equations of The Second Kind, illustrate to provide PVR early warning for the patient, esti is 70%, represent
The probability that it suffers from PVR is 70%.
In the scheme of the embodiment of the present invention, by the pretreatment of the diabetes historical data to sample patient, obtain
To medical characteristics data, the incidence relation that the medical characteristics data occur with PVR is set up, so as to
Diabetic feeds back the early warning information of PVR, reminds diabetic to go to a doctor in time, reduces or keeps away
Exempt from the possibility that sb.'s illness took a turn for the worse.
The above is the preferred embodiment of the present invention, it is noted that for the common skill of the art
For art personnel, on the premise of principle of the present invention is not departed from, some improvements and modifications can also be made,
These improvements and modifications also should be regarded as protection scope of the present invention.
Claims (19)
1. a kind of method for early warning of diabetic retinopathy, it is characterised in that including:
Obtain the diabetes historical data of the sample patient of predetermined quantity;
Diabetes historical data to the sample patient of predetermined quantity is pre-processed, and obtains medical characteristics data;
Set up the incidence relation that the medical characteristics data occur with PVR;
According to the incidence relation, whether the patient for not occurred PVR there is retinopathy
The early warning information of change.
2. the method for early warning of diabetic retinopathy according to claim 1, it is characterised in that
The diabetes historical data includes:Diabetes diagnosis time, dynamic retinoscopy history and blood pressure, blood sugar,
The Monitoring Data of Glycohemoglobin HbA1c and blood fat.
3. the method for early warning of diabetic retinopathy according to claim 1, it is characterised in that
The medical characteristics data include:Diabetic duration, dysarteriotony ratio, pathoglycemia ratio, HbA1c
Unnatural proportions and dyslipidemia ratio;The unnatural proportions are determined by the personalized control targe of patient.
4. the method for early warning of the diabetic retinopathy according to Claims 2 or 3, its feature exists
In the diabetes historical data to the patient of predetermined quantity is pre-processed, and obtains the step of medical characteristics data
Suddenly, including:
The patient categories of the sample patient are determined according to the dynamic retinoscopy history;
The diabetes historical data is pre-processed according to the patient categories, obtains medical characteristics data.
5. the method for early warning of diabetic retinopathy according to claim 4, it is characterised in that
The step of patient categories of the sample patient are determined according to the dynamic retinoscopy history, including:
All the time the sample patient for not occurring PVR in the dynamic retinoscopy history is defined as
One class patient;The sample patient that PVR occurs in midway is defined as Equations of The Second Kind patient.
6. the method for early warning of diabetic retinopathy according to claim 5, it is characterised in that
The diabetes historical data is pre-processed according to the patient categories, obtains the step of medical characteristics data
Suddenly, including:
The time checked for the last time in dynamic retinoscopy history according to the first kind patient, it is determined that described
The diabetic duration of first kind patient;
Obtain in the dynamic retinoscopy history it is last check twice between interval time in all of blood pressure,
The Monitoring Data of blood sugar, HbA1c and blood fat, determines the unnatural proportions of the Monitoring Data.
7. the method for early warning of diabetic retinopathy according to claim 5, it is characterised in that
The diabetes historical data is pre-processed according to the patient categories, obtains the step of medical characteristics data
Suddenly, also include:
Checked for the first time in dynamic retinoscopy history according to the Equations of The Second Kind patient PVR when
Between, determine the diabetic duration of the Equations of The Second Kind patient;
Obtain check for the last time in the dynamic retinoscopy history without lesion with check lesion for the first time
Between time interval in all of blood pressure, blood sugar, the Monitoring Data of HbA1c and blood fat, determine institute
State the unnatural proportions of Monitoring Data.
8. the method for early warning of the diabetic retinopathy according to claim 6 or 7, its feature exists
In, the step of set up the incidence relation that the medical characteristics data and PVR occur, including:
The medical characteristics data are obtained into disaggregated model parameter by a support vector machines algorithm;
The medical characteristics data and associating that PVR occurs are set up according to the disaggregated model parameter
The svm classifier model of relation.
9. the method for early warning of diabetic retinopathy according to claim 8, it is characterised in that
According to the incidence relation, whether the patient for not occurred PVR there is PVR
The step of early warning information, including:
By in the medical characteristics data input to the svm classifier model, do not occurred described in generation
Whether the patient of PVR there is the early warning and alert result of PVR.
10. a kind of prior-warning device of diabetic retinopathy, it is characterised in that including:
Acquisition module, the diabetes historical data of the sample patient for obtaining predetermined quantity;
Data preprocessing module, the diabetes historical data for the sample patient to predetermined quantity carries out pre- place
Reason, obtains medical characteristics data;
Relation sets up module, for setting up the incidence relation that the medical characteristics data occur with PVR;
Lesion warning module, for according to the incidence relation, not occurred the trouble of PVR
Whether person there is the early warning information of PVR.
The prior-warning device of 11. diabetic retinopathy according to claim 10, it is characterised in that
The diabetes historical data includes:Diabetes diagnosis time, dynamic retinoscopy history and blood pressure, blood sugar,
The Monitoring Data of Glycohemoglobin HbA1c and blood fat.
The prior-warning device of 12. diabetic retinopathy according to claim 10, it is characterised in that
The medical characteristics data include:Diabetic duration, dysarteriotony ratio, pathoglycemia ratio,
HbA1c unnatural proportions and dyslipidemia ratio;The unnatural proportions are true by the personalized control targe of patient
It is fixed.
The prior-warning device of 13. diabetic retinopathy according to claim 11 or 12, its feature
It is that the data preprocessing module includes:
Classification determination sub-module, the patient for determining the sample patient according to the dynamic retinoscopy history
Classification;
Data prediction submodule, it is pre- for being carried out to the diabetes historical data according to the patient categories
Treatment, obtains medical characteristics data.
The prior-warning device of 14. diabetic retinopathy according to claim 13, it is characterised in that
The classification determination sub-module includes:
Classification determination unit, for PVR will not all the time occur in the dynamic retinoscopy history
Sample patient is defined as first kind patient;The sample patient that PVR occurs in midway is defined as Equations of The Second Kind trouble
Person.
The prior-warning device of 15. diabetic retinopathy according to claim 14, it is characterised in that
The data prediction submodule includes:
First data determination unit, in the dynamic retinoscopy history according to the first kind patient last
The time of secondary inspection, determine the diabetic duration of the first kind patient;
First data acquisition process unit, finally check it twice in the dynamic retinoscopy history for obtaining
Between interval time in all of blood pressure, blood sugar, HbA1c and blood fat Monitoring Data, determine the prison
Survey the unnatural proportions of data.
The prior-warning device of 16. diabetic retinopathy according to claim 14, it is characterised in that
The data prediction submodule also includes:
Second data determination unit, for first time in the dynamic retinoscopy history according to the Equations of The Second Kind patient
The time of PVR is checked, the diabetic duration of the Equations of The Second Kind patient is determined;
Second data acquisition process unit, check for the last time in the dynamic retinoscopy history for obtaining
All of blood pressure, blood sugar, the HbA1c in the time interval between lesion are checked without lesion and first time
And the Monitoring Data of blood fat, determine the unnatural proportions of the Monitoring Data.
The prior-warning device of 17. diabetic retinopathy according to claim 15 or 16, its feature
It is that the relation sets up module to be included:
Model parameter determination sub-module, for the medical characteristics data to be passed through into a support vector machines
Algorithm obtains disaggregated model parameter;
Disaggregated model sets up module, for according to the disaggregated model parameter set up the medical characteristics data with
The svm classifier model of the incidence relation that PVR occurs.
The prior-warning device of 18. diabetic retinopathy according to claim 17, it is characterised in that
The lesion warning module includes:
Early warning generates submodule, for by the medical characteristics data input to the svm classifier model,
Whether the patient for not occurred PVR described in generation there is the early warning and alert result of PVR.
19. a kind of early warning systems of diabetic retinopathy, it is characterised in that including diabetes backstage certainly
My management system and mobile terminal;Wherein,
The diabetes backstage self-management system includes the diabetes as described in claim any one of 10-18
The prior-warning device of PVR;
The mobile terminal is used to gather the diabetes historical data of patient, receives diabetes backstage self-management
The early warning information of the PVR that system is returned.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510770592.XA CN106709225A (en) | 2015-11-12 | 2015-11-12 | Early-warning method, device and system for diabetic retinopathy |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510770592.XA CN106709225A (en) | 2015-11-12 | 2015-11-12 | Early-warning method, device and system for diabetic retinopathy |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106709225A true CN106709225A (en) | 2017-05-24 |
Family
ID=58929635
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510770592.XA Pending CN106709225A (en) | 2015-11-12 | 2015-11-12 | Early-warning method, device and system for diabetic retinopathy |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106709225A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109350006A (en) * | 2018-08-10 | 2019-02-19 | 北京大学深圳研究生院 | Diabetic complication retinopathy data detection method, instrument and Medical Devices |
CN109903838A (en) * | 2019-02-22 | 2019-06-18 | 爱尔眼科医院集团股份有限公司 | Diabetes and diabetic retinopathy network prevent and treat system |
CN110097966A (en) * | 2018-01-30 | 2019-08-06 | 中国移动通信有限公司研究院 | A kind of information prompting method, device and terminal device |
CN112786193A (en) * | 2021-01-20 | 2021-05-11 | 上海市第六人民医院 | Monitoring and early warning system and method for hypoglycemia patient |
CN113035352A (en) * | 2021-04-27 | 2021-06-25 | 河南科技大学 | Diabetic retinopathy early warning method based on BP neural network |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102178511A (en) * | 2011-04-11 | 2011-09-14 | Tcl集团股份有限公司 | Disease prevention warning system and implementation method |
CN103458772A (en) * | 2011-04-07 | 2013-12-18 | 香港中文大学 | Method and device for retinal image analysis |
-
2015
- 2015-11-12 CN CN201510770592.XA patent/CN106709225A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103458772A (en) * | 2011-04-07 | 2013-12-18 | 香港中文大学 | Method and device for retinal image analysis |
CN102178511A (en) * | 2011-04-11 | 2011-09-14 | Tcl集团股份有限公司 | Disease prevention warning system and implementation method |
Non-Patent Citations (7)
Title |
---|
K.R.ANANTHAPADMANABAN ET AL: "Prediction of Chances-Diabetic Retinopathy using Data Mining Classification Techniques", 《INDIAN JOURAL OF SCIENCE AND TECHNOLOGY》 * |
中华医学会眼科学会眼底病学组: "我国糖尿病视网膜病变临床诊疗指南(2014年)", 《中华眼科杂志》 * |
中华医学会糖尿病学分会: "中国2型糖尿病防治指南(2013年版)", 《中国糖尿病杂志》 * |
朱治人等: "电子计算机筛选糖尿病性视网膜病变的防止指标——Bayes逐步判别分析", 《眼科新进展》 * |
董萍: "《建设监理人员压力管理研究》", 31 December 2012, 合肥工业大学出版社 * |
陈云: "《金融大数据》", 31 January 2015 * |
陈向: "糖尿病视网膜病变图像的渗出物自动检测算法研究", 《中国优秀硕士学位论文全文数据库 医药卫生科技辑》 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110097966A (en) * | 2018-01-30 | 2019-08-06 | 中国移动通信有限公司研究院 | A kind of information prompting method, device and terminal device |
CN110097966B (en) * | 2018-01-30 | 2021-09-14 | 中国移动通信有限公司研究院 | Information reminding method and device and terminal equipment |
CN109350006A (en) * | 2018-08-10 | 2019-02-19 | 北京大学深圳研究生院 | Diabetic complication retinopathy data detection method, instrument and Medical Devices |
CN109903838A (en) * | 2019-02-22 | 2019-06-18 | 爱尔眼科医院集团股份有限公司 | Diabetes and diabetic retinopathy network prevent and treat system |
CN112786193A (en) * | 2021-01-20 | 2021-05-11 | 上海市第六人民医院 | Monitoring and early warning system and method for hypoglycemia patient |
CN113035352A (en) * | 2021-04-27 | 2021-06-25 | 河南科技大学 | Diabetic retinopathy early warning method based on BP neural network |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US20220076420A1 (en) | Retinopathy recognition system | |
US11967074B2 (en) | Method and system for computer-aided triage | |
US11213197B2 (en) | Artificial neural network and system for identifying lesion in retinal fundus image | |
CN106709225A (en) | Early-warning method, device and system for diabetic retinopathy | |
Ganesan et al. | Computer-aided diabetic retinopathy detection using trace transforms on digital fundus images | |
US20210391056A1 (en) | Health big data service method and system based on remote fundus screening | |
US11488299B2 (en) | Method and system for computer-aided triage | |
US20120275677A1 (en) | Image Analysis System and Related Methods and Computer Program Products | |
US11526762B2 (en) | Method and system of training a machine learning neural network system for patient medical states | |
WO2022082004A1 (en) | Systems and methods using ensemble machine learning techniques for future event detection | |
Odeh et al. | Diabetic retinopathy detection using ensemble machine learning | |
Uppamma et al. | Deep learning and medical image processing techniques for diabetic retinopathy: a survey of applications, challenges, and future trends | |
Reddy et al. | Discovering optimal algorithm to predict diabetic retinopathy using novel assessment methods | |
Sambyal et al. | Aggregated residual transformation network for multistage classification in diabetic retinopathy | |
Wehkamp et al. | The quality and utility of artificial intelligence in patient care | |
Qaraqe et al. | AI and machine learning in diabetes management: opportunity, status, and challenges | |
WO2023242016A1 (en) | Detecting user infection using wearable sensor data cross-reference to related applications | |
Dai et al. | Detecting Mental Disorders with Wearables: A Large Cohort Study | |
CN112418290B (en) | ROI (region of interest) region prediction method and display method of real-time OCT (optical coherence tomography) image | |
CN115206494A (en) | Film reading system and method based on fundus image classification | |
Mohan et al. | Using artificial intelligence in diabetic retinopathy | |
Dorali et al. | Utilizing Simulation to Update Routine Diabetic Retinopathy Screening Policies | |
US20220207732A1 (en) | Systems and methods for processing of fundus images | |
Hakeem et al. | Inception V3 and CNN Approach to Classify Diabetic Retinopathy Disease | |
US11766223B1 (en) | Systems and methods for processing of fundus images |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170524 |