CN1757375A - Lamination estimate method for cardiovascular danger of hyperpietic based artificial nervous network - Google Patents
Lamination estimate method for cardiovascular danger of hyperpietic based artificial nervous network Download PDFInfo
- Publication number
- CN1757375A CN1757375A CNA2005100611523A CN200510061152A CN1757375A CN 1757375 A CN1757375 A CN 1757375A CN A2005100611523 A CNA2005100611523 A CN A2005100611523A CN 200510061152 A CN200510061152 A CN 200510061152A CN 1757375 A CN1757375 A CN 1757375A
- Authority
- CN
- China
- Prior art keywords
- index
- hyperpietic
- layering
- cardiovascular
- cardiovascular danger
- 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
- Measuring Pulse, Heart Rate, Blood Pressure Or Blood Flow (AREA)
Abstract
A method for evaluating the cardiovascular damage degree of hypertension patient on the basis of artificial nerve network features that after 10 parameters including systolic pressure, diastolic pressure, body weight index, ejection period, pulse output, cardiac output, etc are input, the cardiovascular damage degree can be calculated out by use of the nerve network equation Y=f2 (W2f1(W1x+B1)+B2). Its advantage is high correctness.
Description
Technical field
The present invention relates to a kind of hyperpietic's cardiovascular danger layering appraisal procedure based on artificial neural network.
Background technology
Cardiovascular danger abswolute level layering (hereinafter to be referred as risk stratification) is an international cardiovascular status characteristic parameter, and its meaning is to understand influence and the prognosis to cardiovascular system of patient's blood pressure level and other risk factors." Chinese hypertension prevention and control guide ", " 1999WHO/ISH hypertension is handled guide " (hereinafter to be referred as WHO99), " European hypertension therapeutic guide 2003 " (hereinafter to be referred as Europe 2003), American National hypertension prevention, detection, evaluation and treatment joint committee the 6th, the computational methods of reporting that standards such as (hereinafter to be referred as JNC6, JNC7) has all been issued risk stratification the 7th time.Yet in clinical, because the restriction of level own, too subjectivity and dependent patient master state, China clinician is lower to the grasp degree of risk stratification, so that has influence on follow-up clinical decision effect.
Artificial neural network (artificial neural network) is a black-box model, can express those also can't accurately describe with mechanism, but the definitiveness that really exists between the input and output or the problem of ambiguity objective law, these characteristics are highly suitable in this situation that can't write out conventional algorithm and use.The error anti-pass (Back-Propagation, BP) network is as a kind of classic algorithm, and the risk stratification information that various standards can be provided is stored in the network parameter with the form of weights and biasing just.
Summary of the invention
The purpose of this invention is to provide a kind of hyperpietic's cardiovascular danger layering appraisal procedure based on artificial neural network, the cardio-vascular parameters that obtains from clinical examination, objective evaluation hyperpietic cardiovascular danger is eliminated the subjective uncertainty of introducing in doctor's practical operation.
Hyperpietic's cardiovascular danger layering appraisal procedure based on artificial neural network may further comprise the steps:
1) input measured's systolic pressure, diastolic pressure, Body Mass Index, ejection time, stroke volume, cardiac output, sclerosis of blood vessels index, pulse wave conduction speed, system's vascular resistance index, left heart performance index;
2) utilize neutral net equation Y=f
2(W
2f
1(W
1X+B
1)+B
2), with each amount of step 1) input cardiovascular danger layering Y as the calculation of parameter measured; X is the vector that parameter constitutes in the step 1) in the formula, f
1Be S type transfer function, f
2Be linear transfer function, W
1, W
2, B
1, B
2Be constant matrices.
The present invention is divided into four layers of temporary nothing or general danger, low degree of hazard, poor risk, height or R4s with hyperpietic's cardiovascular danger layering.
Constant matrices W in the above-mentioned neutral net equation
1, W
2, B
1, B
2Can be in advance by the input of the cardio-vascular parameters in the existing database of patient information of utilization as the neutral net equation, the cardiovascular danger layering of standard is exported as target, adopts error back propagation algorithm to obtain.Here the cardio-vascular parameters in the said information database comprises: systolic pressure, diastolic pressure, Body Mass Index, ejection time, stroke volume, cardiac output, sclerosis of blood vessels index, pulse wave conduction speed, system's vascular resistance index, left heart performance index.
Determine W
1, W
2, B
1, B
2Concrete steps are as follows:
1. determine the input of neutral net equation
Adopt database of patient information as the input source.In the data base, choose the sample of sufficient amount, after its clinical examination data process cleaning and pretreatment, be summarised as three parts: measured's essential information, cardiac functional parameter and peripheral blood vessel parameter.Again it is carried out determining behind multi-variate statistical analysis, the principal component analysis: systolic pressure, diastolic pressure, Body Mass Index, ejection time, stroke volume, cardiac output, the sclerosis of blood vessels index, pulse wave conduction speed, system's vascular resistance index, ten parameters of left heart performance index are as the input of neutral net equation.
2. determine the target output of neutral net equation
According to " Chinese hypertension prevention and control guide ",, and work out with reference to associated guidelines such as WHO99, JNC7, Europe 2003 in conjunction with China's present situation.Wherein the blood pressure classification is with reference to the standard simplified of JNC7.Other risk factors are determined according to " Chinese hypertension prevention and control guide ".As Data Source, take all factors into consideration blood pressure classification, other risk factors are divided into four layers with danger level with database of patient information.
3. carrying out the error anti-pass calculates
Adopt three layers of (comprising input layer, hidden layer, output layer) error back-propagation network, select the hidden layer node number, use Levenberg-Marquardt Back-Propagation (anti-pass of LM error) algorithm to upgrade the weights W of network according to trial-and-error method
1, W
2With biasing B
1, B
2Each renewal is as follows:
Wb wherein
kBe to represent current weight W
1, W
2With biasing B
1, B
2Vector, wb
K+1Be the value after upgrading, A
kBe learning rate, provide by following formula:
A
k=J
T(x
k)J(x
k)+μ
kI (2)
g
kBe current gradient:
g
k=J
T(x
k)e(x
k) (3)
Wherein J is that network error is to weights W
1, W
2With biasing B
1, B
2The Jacobian matrix of one subdifferential, e are the network error vectors, and μ is a scalar that influences learning rate, and I is a unit matrix.
Adopt mean square error (MSE) to come the evaluating network performance, it is defined as follows:
Wherein N is the number of training of input, t
iBe by target output, a
iBe network output.When MSE reaches default standard, network just stops training.Introduced the conclusive evidence collection in addition as another one standard arranged side by side, the mean square error of conclusive evidence collection (is designated as MSE
v) after each training, all will check, if MSE
vDiminish, training process continues, otherwise stops training, uses the current network parameter as final result.So the comprehensive criterion that stops to train is as follows:
K represents the k time training, MSE in the formula
v kAnd MSE
v K-1The MSE of conclusive evidence collection when representing current and last the training respectively.
On the whole, training process comprises three steps:
The first, select training set and conclusive evidence collection in the sample general collection at random.
The second, for the MSE between the output that reduces the neutral net equation and the target output, network weight W is also adjusted in training
1, W
2With biasing B
1, B
2
The 3rd, if criterion formula (5) satisfies, training stops; Otherwise repeat second, third step.After the training of neutral net equation is finished, W
1, W
2, B
1, B
2Determine.
The beneficial effect of appraisal procedure of the present invention:
1. the physiological parameter of utilizing objective measurement to obtain mainly is that cardio-vascular parameters calculates an international cardiovascular status characteristic parameter---cardiovascular danger abswolute level layering, has eliminated the subjective uncertainty of introducing in doctor's practical operation.
2. carry out easyly, only need measured's cardiac function, peripheral blood vessel and some essential information data, saved other loaded down with trivial details inspections,, provide cost savings such as biochemical analysis etc.
3. having used the BP network is the data mining technology of core, has proved to have contact between risk stratification and the cardio-vascular parameters really, has found out the bigger some cardio-vascular parameters of risk stratification influence.
Description of drawings
Fig. 1 is the comparison diagram of this appraisal procedure test result.
The specific embodiment
Further specify the present invention below in conjunction with embodiment.
Neutral net equation Y=f
2(W
2f
1(W
1X+B
1)+B
2) middle constant matrices W
1, W
2, B
1, B
2Determine:
A) utilize certain 261 routine patient's of hospital information database, put out wherein essential information, cardiac function and peripheral blood vessel parameter etc. in order, these parameters are carried out multi-variate statistical analysis, determine following parameter behind the principal component analysis: systolic pressure (SBP), diastolic pressure (DBP), Body Mass Index, ejection time, stroke volume, cardiac output, sclerosis of blood vessels index, pulse wave conduction speed, system's vascular resistance index, left heart performance index is as the input of neutral net equation.
B),,, and formulate the cardiovascular danger layered sheet (table 1) of simplification with reference to associated guidelines such as WHO99, JNC7, Europe 2003 in conjunction with China's present situation according to " Chinese hypertension prevention and control guide " at 261 routine patients.Wherein blood pressure classification is divided into 4 grades with reference to the JNC7 standard simplified, and other risk factors are defined as following five according to " Chinese hypertension prevention and control guide ": smoking, hyperlipidemia, diabetes, male greater than 55 years old or women greater than 65 years old, the angiopathy family history of early making up one's mind.Take all factors into consideration above-mentioned blood pressure classification and other risk factors, it is dangerous that the cardiovascular danger layering is divided into temporary nothings, general dangerous, low degree of hazard, and poor risk, highly dangerous and R4, the target of Here it is this neural network equation is exported.Consider the simplicity of clinical execution, the output of neutral net equation is reduced to four layers, and 1 layer of expression do not have danger or general dangerous temporarily, 2 layers of expression low degree of hazard, and 3 layers of expression poor risk are represented highly dangerous or R4 for 4 layers.
The cardiovascular danger layered sheet that table 1 is simplified
Other risk factors | Blood pressure (mmHg) | |||
Normal arterial pressure SBP<120 and DBP<80 | Hypertension SBP:120-139 in early stage or DBP:80-89 | First phase hypertension SBP:140-159 or DBP:90-99 | Second phase hypertension SBP>=160 or DBP>=100 | |
There are not other risk factors | Do not have dangerous temporarily | General dangerous | Low degree of hazard | Poor risk |
1 to 2 risk factor | General dangerous | Low degree of hazard | Poor risk | Highly dangerous |
3 above risk factors or diabetes are arranged | Low degree of hazard | Poor risk | Highly dangerous | R4 |
C) determine that according to trial-and-error method the hidden layer node number is 5, mean square error (MSE) preset standard is 10%.Select 174 examples at random as training set in the 261 routine samples, 87 examples are as the conclusive evidence collection.After training is finished, determine neutral net equation Y=f
2(W
2f
1(W
1X+B
1)+B
2) in constant matrices W
1, W
2, B
1, B
2It may be noted that a bit because the stratified result of standard is 1 layer to 4 layers, and the output of neutral net equation is the mixed decimal value, so done following processing when calculating layering: the output of neutral net equation is less than 1 layer of 1.5 correspondence; 1.5 to 2 layers of 2.5 correspondences; 2.5 to 3 layers of 3.5 correspondences; 3.5 above corresponding 4 layers.
Embodiment 1
Systolic pressure 174 with certain hyperpietic, diastolic pressure 91, Body Mass Index 21.78, ejection time 0.3662, stroke volume 54.33, cardiac output 4.167, sclerosis of blood vessels index 33.93, pulse wave conduction speed 3.280, system's vascular resistance index 3345,4.489 ten parameters of left heart performance index (unit is slightly) input computer; The risk stratification result who utilizes the neutral net Equation for Calculating to obtain is 4.0038, corresponding 4 layers.
It is more that this hyperpietic master states uncomfortable symptom, and do not carry out Drug therapy, judges that according to table 1 risk stratification of standard also belongs to 4 layers, is height or R4.
Embodiment 2
Systolic pressure 109 with certain hyperpietic, diastolic pressure 73.5, Body Mass Index 21.01, ejection time 0.3092, stroke volume 53.33, cardiac output 3.533, sclerosis of blood vessels index 29.88, pulse wave conduction speed 3.329, system's vascular resistance index 3403,2.611 ten parameters of left heart performance index (unit is slightly) input computer; The risk stratification result who utilizes the neutral net Equation for Calculating to obtain is 0.9964, corresponding 1 layer.
It is light that this hyperpietic master states diet, and treatment actively takes regular exercise, and judges that according to table 1 risk stratification of standard also belongs to 1 layer, is temporary nothing or general dangerous.
Figure 1 shows that the result that 87 routine hyperpietics adopt the present invention to assess to obtain, the risk stratification of transverse axis for obtaining among the figure according to table 1, the risk stratification that the longitudinal axis obtains for the present invention's assessment, 87 routine samples represented in asterisk, solid line is represented two kinds of layering results' best linear fit, both coefficient R=0.993.The result shows that the present invention assesses the risk stratification that obtains and is 82.76% according to the risk stratification that table 1 the is judged rate of coincideing.
Claims (4)
1. based on hyperpietic's cardiovascular danger layering appraisal procedure of artificial neural network, it is characterized in that may further comprise the steps:
1) input measured's systolic pressure, diastolic pressure, Body Mass Index, ejection time, stroke volume, cardiac output, sclerosis of blood vessels index, pulse wave conduction speed, system's vascular resistance index, left heart performance index;
2) utilize neutral net equation Y=f
2(W
2f
1(W
1X+B
1)+B
2), with each amount of step 1) input cardiovascular danger layering Y as the calculation of parameter measured; X is the vector that parameter constitutes in the step 1) in the formula, f
1Be S type transfer function, f
2Be linear transfer function, W
1, W
2, B
1, B
2Be constant matrices.
2. the hyperpietic's cardiovascular danger layering appraisal procedure based on artificial neural network according to claim 1 is characterized in that constant matrices W
1, W
2, B
1, B
2By the input of the cardio-vascular parameters in the existing database of patient information of utilization as the neutral net equation, the cardiovascular danger layering of standard is exported as target, adopts error back propagation algorithm to obtain in advance.
3. the hyperpietic's cardiovascular danger layering appraisal procedure based on artificial neural network according to claim 2, it is characterized in that the cardio-vascular parameters in the information database comprises: systolic pressure, diastolic pressure, Body Mass Index, ejection time, stroke volume, cardiac output, sclerosis of blood vessels index, pulse wave conduction speed, system's vascular resistance index, left heart performance index.
4. the hyperpietic's cardiovascular danger layering appraisal procedure based on artificial neural network according to claim 1 is characterized in that the cardiovascular danger layering is divided into four layers of temporary nothing or general danger, low degree of hazard, poor risk, height or R4s.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CNA2005100611523A CN1757375A (en) | 2005-10-18 | 2005-10-18 | Lamination estimate method for cardiovascular danger of hyperpietic based artificial nervous network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CNA2005100611523A CN1757375A (en) | 2005-10-18 | 2005-10-18 | Lamination estimate method for cardiovascular danger of hyperpietic based artificial nervous network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN1757375A true CN1757375A (en) | 2006-04-12 |
Family
ID=36703048
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CNA2005100611523A Pending CN1757375A (en) | 2005-10-18 | 2005-10-18 | Lamination estimate method for cardiovascular danger of hyperpietic based artificial nervous network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN1757375A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109935327A (en) * | 2019-03-15 | 2019-06-25 | 南方医科大学顺德医院(佛山市顺德区第一人民医院) | Hypertensive patient's cardiovascular risk grading appraisal procedure based on intelligence decision support system |
CN110688421A (en) * | 2018-06-20 | 2020-01-14 | 南京网感至察信息科技有限公司 | Intelligent customizable data management and analysis method |
CN111248879A (en) * | 2020-02-20 | 2020-06-09 | 电子科技大学 | Hypertension old people activity analysis method based on multi-mode attention fusion |
CN113712524A (en) * | 2021-09-14 | 2021-11-30 | 北京大学人民医院 | Data processing device, system and kit for auxiliary evaluation of cardiovascular risk |
-
2005
- 2005-10-18 CN CNA2005100611523A patent/CN1757375A/en active Pending
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110688421A (en) * | 2018-06-20 | 2020-01-14 | 南京网感至察信息科技有限公司 | Intelligent customizable data management and analysis method |
CN109935327A (en) * | 2019-03-15 | 2019-06-25 | 南方医科大学顺德医院(佛山市顺德区第一人民医院) | Hypertensive patient's cardiovascular risk grading appraisal procedure based on intelligence decision support system |
CN109935327B (en) * | 2019-03-15 | 2023-08-08 | 南方医科大学顺德医院(佛山市顺德区第一人民医院) | Cardiovascular risk layering evaluation method for hypertension patient based on intelligent decision support |
CN111248879A (en) * | 2020-02-20 | 2020-06-09 | 电子科技大学 | Hypertension old people activity analysis method based on multi-mode attention fusion |
CN111248879B (en) * | 2020-02-20 | 2021-12-07 | 电子科技大学 | Hypertension old people activity analysis method based on multi-mode attention fusion |
CN113712524A (en) * | 2021-09-14 | 2021-11-30 | 北京大学人民医院 | Data processing device, system and kit for auxiliary evaluation of cardiovascular risk |
CN113712524B (en) * | 2021-09-14 | 2023-08-01 | 北京大学人民医院 | Data processing device, system and kit for auxiliary assessment of cardiovascular risk |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108714026A (en) | The fine granularity electrocardiosignal sorting technique merged based on depth convolutional neural networks and on-line decision | |
Shankarapillai et al. | Periodontitis risk assessment using two artificial neural networks-a pilot study | |
CN1757375A (en) | Lamination estimate method for cardiovascular danger of hyperpietic based artificial nervous network | |
CN102670190A (en) | Heart rate variability nonlinear characteristic-based automatic diagnosis method for congestive heart failure | |
Dick et al. | Geometric models to explore mechanisms of dynamic shape change in skeletal muscle | |
CN113362954A (en) | Postoperative infection complication risk early warning model for old patients and establishment method thereof | |
CN115295114A (en) | Analgesic administration method based on objective pain assessment system | |
Das et al. | Relation between New York Heart Association functional class and objective measures of cardiopulmonary exercise in adults with congenital heart disease | |
CN1275179C (en) | Computer evaluating method for human body sub-health status | |
Dymecka et al. | Validity and reliability of the Polish adaptation of the Health-Related Hardiness Scale–the first confirmatory factor analysis results for a commonly used scale | |
CN114724717A (en) | Stomach early cancer high-risk screening system | |
CN113903450A (en) | Construction system of type 2 diabetes risk prediction model | |
CN106845140A (en) | A kind of kidney failure method for early warning monitored based on specific gravity of urine and urine volume and system | |
CN109460942A (en) | Method and Related product based on data assay hospital | |
CN1559341A (en) | Method and apparatus for detecting, and analysing heart rate variation predication degree index | |
CN110739072A (en) | Bleeding event occurrence evaluation method and system | |
CN109192315B (en) | Comprehensive age detection system based on weighted kernel regression and packaged deviation search | |
CN112216391B (en) | Method and device for evaluating cerebral apoplexy incidence risk based on carotid atherosclerosis | |
CN106570346A (en) | Physiological status assessment factor determination method and physiological status assessment factor determination system | |
CN114864086A (en) | Disease prediction method based on lung function report template | |
CN114550896A (en) | Dizziness patient emergency pre-examination triage decision-making method, device and model based on artificial neural network | |
CN112863681A (en) | Method for predicting maximum oxygen uptake by multi-parameter screening | |
Reybrouck et al. | Long-term assessment and serial evaluation of cardiorespiratory exercise performance and cardiac function in patients with atrial switch operation for complete transposition | |
CN110544535B (en) | Turner syndrome bone health evaluation system and establishment method thereof | |
CN116434954B (en) | Evaluation method of clinical treatment effect |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
WD01 | Invention patent application deemed withdrawn after publication |