CN103488857A - Spontaneous abortion risk prediction system and method for establishing system - Google Patents

Abstract

The invention discloses a spontaneous abortion risk prediction system and a method for establishing the system. The spontaneous abortion risk prediction system comprises an input unit, a prediction unit and an output unit, wherein the input unit is used for inputting an influencing parameter of influencing spontaneous abortion; the prediction unit is used for performing prediction on spontaneous abortion risk on the basis of the input influencing parameter; the output unit is used for outputting a spontaneous abortion risk prediction result. According to the spontaneous abortion risk prediction system, as long as a user simply inputs the related influencing parameter, the prediction result can be rapidly and conveniently obtained. On one hand, the time and the energy of the user are saved, and on the other hand, the workload of technical service personnel is reduced. Further, when the prediction result is unfavorable, the user can go to the hospital or ask an expert for further diagnosis.

Description

A kind of spontaneous abortion Risk Forecast System and build the method for this system

Technical field

The present invention relates to a kind of spontaneous abortion Risk Forecast System and build the method for this system.

Background technology

Spontaneous abortion refers to the not enough 1000g in pregnant less than 28 weeks, fetal weight and spontaneous termination gestation.Spontaneous abortion, according to the asynchronism(-nization) occurred, can be divided into again early stage spontaneous abortion, midtrimester abortion and premature labor several.The early pregnancy spontaneous abortion generally referred to before fetus can survive, and flowed out outside parent.Spontaneous abortion accounts for ten Percent five to 30 left and right of all gestation.Massive haemorrhage occurs and causes hemorrhagic shock in part spontaneous abortion women's federation, or accompanying infection, causes pelvic infecton, abdominal distension inflammation, general infection and infectious shock.Spontaneous abortion harm women's is healthy, and the while is also caused harmful effect to couple's psychology, and severe patient may cause depression.

At present, what most of family planning service organization carried out that reproductive health service still takes is traditional mode, by family planning specially the dry propaganda and education of visiting, register one's residence and follow up a case by regular visits to, provide contraceptive medicines and devices, or organize regularly Child-bearing Aged Women in Countryside to local service station to look into ring and look into pregnant, the instructive information amount that the service of this passive type provides is science seldom and not, inefficiency, the masses' acceptability is also poor.For example, reproductive population is when running into relevant puzzled of reproductive health service, urgent hope can obtain expert's consultation guidance service, yet resource-constrained, and the expert is difficult to accomplish 24 hours persistent services, this traditional consultation way has significant limitation popularizing aspect knowledge on reproductive health.For example, some of the staff are because the covering of family planning service network is not enough or not free/money goes to hospital to carry out the prediction of spontaneous abortion risk.

Therefore, wishing a kind of novel spontaneous abortion Risk Forecast System comes more effectively, more easily the spontaneous abortion risk is predicted.

Summary of the invention

The object of the present invention is to provide a kind of novel spontaneous abortion Risk Forecast System to come more effectively, more easily the spontaneous abortion risk is predicted.

The invention provides a kind of spontaneous abortion Risk Forecast System, described spontaneous abortion Risk Forecast System comprises: input media, and it is for inputting the parameter that affects that affects spontaneous abortion; Prediction unit, its parameter that affects based on inputted is predicted the spontaneous abortion risk; And output unit, for exporting spontaneous abortion risk profile result.

By described spontaneous abortion Risk Forecast System, the user inputs the relative influence parameter simply, can be fast, obtain and predict the outcome easily.This has saved user's time and energy on the one hand; Alleviated on the other hand the workload of service technician.Further, the user, predicting the outcome when unfavorable, can go hospital or consultant expert further to diagnose.

Preferably, described input media is connected with prediction unit by internet with output unit.

Preferably, described input media is connected with prediction unit by wired mode with output unit.

Preferably, described input media is connected with prediction unit by fixed telephone network with output unit.

Preferably, described input media is connected with prediction unit by wireless mode with output unit.

Preferably, described input media is connected with prediction unit by cell phone network with output unit.

Preferably, described output unit is audio output device.

By adopting above-mentioned various connected modes, can greatly facilitate the use of user to prediction unit.Thus, by growing infotech and day by day universal Internet resources, build healthy reproduction education and digital home's consultation service platform, carry out reproductive health service.For the user provides intelligentized consulting service, alleviated medical worker's working strength by the collaborative mode exchanged.

Preferably, described prediction unit is the artificial neural network prediction unit.By adopting the neural network prediction device, can, so that prediction unit has self-learning function, improve constantly the accuracy rate of prediction.

The present invention also provides a kind of method that builds the spontaneous abortion Risk Forecast System, comprises the steps:

1) data of input are carried out to normalized;

2) data after normalized are weighted to summation, the weighted sum formula is:

$A_j(\stackrel{\‾}{x},\stackrel{\‾}{w})={\mathrm{\Σ}}_{i=0}^n{x}_i{w}_{\mathrm{ji}}$

Wherein xi means input, and wji means corresponding weights, and Aj means the output of j node;

3) set up excitation function, the excitation function formula is:

$O_j(\stackrel{\‾}{x},\stackrel{\‾}{w})=\frac{1}{1+e^{A_j(\stackrel{\‾}{x},\stackrel{\‾}{w})}}$

Wherein, Oj means the output valve of j node after excitation function;

4) training of neural network, the neural metwork training error calculation formula:

$E(\stackrel{\‾}{x},\stackrel{\‾}{w},\stackrel{\‾}{d})=\underset{j}{\mathrm{\Σ}}{(O_j(\stackrel{\‾}{x},\stackrel{\‾}{w})-d_j)}^2$

Wherein, E means the error of whole neural network;

5) foundation of optimal network model, according to the variation of output valve, carry out differentiate to error to weights, thereby can be finely tuned weights, and after making each fine setting, error amount can reduce.

$\mathrm{\Δ}{w}_{\mathrm{ji}}=-\mathrm{\η}\frac{\∂E}{\∂w_{\mathrm{ji}}}$

Wherein η is the learning rate of network, and learning rate is higher, and the convergence of neural network is faster, but may miss optimum solution.

Preferably, in step 1) in choose following parameter as affecting parameter: male/female side's schooling, male/female side's occupation type, male/female side's age, the wife's side previously whether conceived, whether whether using contraceptives, the wife's side to take food, whether the whether whether smoking of apocleisis vegetables, the wife's side of meat egg, the wife's side, the wife's side contact whether second-hand-cigarette, the wife's side drink, wife's side blood group and wife's side haemoglobin check the value.

The accompanying drawing explanation

Fig. 1 is the structural representation of the spontaneous abortion Risk Forecast System invented of the present invention.

Fig. 2 is the spontaneous abortion risk forecast model construction step of pregnant front factor.

Embodiment

The invention provides a kind of spontaneous abortion Risk Forecast System, described spontaneous abortion Risk Forecast System comprises: input media 1, and it is for inputting the parameter that affects that affects spontaneous abortion; Prediction unit 2, its parameter that affects based on inputted is predicted the spontaneous abortion risk; And output unit 3, for exporting spontaneous abortion risk profile result.

By described spontaneous abortion Risk Forecast System, the user inputs the relative influence parameter simply, can be fast, obtain and predict the outcome easily.This has saved user's time and energy on the one hand; Alleviated on the other hand the workload of service technician.Further, the user, predicting the outcome when unfavorable, can go hospital or consultant expert further to diagnose.

Utilize data mining technology, information in conjunction with pregnant front eugenic health examination gestation queuing data storehouse, pregnant front risk to spontaneous abortion is predicted, through comparing, select artificial nerve network model to carry out the structure of risk warning model, introduce following a few class risk factors (affecting parameter): male/female side's schooling, male/female side's occupation type, male/female side's age, whether the wife's side is previously conceived, whether used contraceptives, the wife's side meat egg of whether taking food, whether the wife's side apocleisis vegetables, whether smoking of the wife's side, whether the wife's side contacts second-hand-cigarette, whether drink on the wife's side, wife's side blood group, wife's side haemoglobin check the value etc.In one embodiment, be to comprise above-mentioned whole variablees.Can also introduce other input variable.

Described input media 1 can be for example keyboard (computor-keyboard, cell phone keyboard), mouse, touch-screen, voice input device etc.

Described output unit 3 is such as being display, printer, facsimile recorder, audio output device etc.

Described prediction unit 2 is such as being the main frame that carries expert database (or expert system), central processing unit, webserver etc.

Described input media 1 can also comprise the data normalization device, so that all input variables are carried out to normalized.The normalization formula of taking is:

V _new＝(V _old-V _min)/(V _max-V _min)。

Wherein, V _newthe variate-value after normalization, V _oldthe original variable value, V _maxto obtain maximal value, V in the original variable group _minto obtain minimum value in the original variable group.

In the present invention's invention, all input variables are carried out to normalized.

Preferably, described input media 1 is connected with prediction unit by internet with output unit 3.

Preferably, described input media 1 for example, is connected with prediction unit by wired mode (fixed telephone network) with output unit 3.

Preferably, described input media 1 for example, is connected with prediction unit 2 by wireless mode (cell phone network) with output unit 3.

Preferably, described output unit 3 is audio output device.

By adopting above-mentioned various connected modes, can greatly facilitate the use of user to prediction unit.Thus, by growing infotech and day by day universal Internet resources, build healthy reproduction education and digital home's consultation service platform, carry out reproductive health service.For the user provides intelligentized consulting service, alleviated medical worker's working strength by the collaborative mode exchanged.

Preferably, described prediction unit 2 is the artificial neural network prediction unit.By adopting the neural network prediction device, can, so that prediction unit has self-learning function, improve constantly the accuracy rate of prediction.

Advantage based on artificial neural network, in the present invention's invention, be applied to artificial neural network spontaneous abortion youngster's prediction.Introduce altogether 4182 routine samples and carry out models fitting, wherein 2889 examples (70%) are for model creation, and 1293 examples (30%) are for model testing.

The concrete steps that described risk warning model builds are as follows:

1. the normalized of data.

2. normalization data is weighted to summation.

3. set up excitation function.

4. the training of neural network.

5. build the optimal network model

In described artificial nerve network model, artificial neural network is divided into three layers, and one is input layer (Input layer), and numerous neurons (Neuron) are accepted a large amount of non-linear input messages.The information of input is called input vector.It two is output layer (Output layer), information transmission in the neuron link, analyze, balance, form Output rusults.The information of output is called output vector.It three is hidden layer (Hidden layer), is called for short " hidden layer ", is the every aspect that between input layer and output layer, numerous neurons and link form.Hidden layer also can have multilayer, but selects one deck in the present invention.

Fig. 2 is the spontaneous abortion risk forecast model construction step of pregnant front factor.

Below with regard to each step, describe in detail

Determine the parameter that affects that affects spontaneous abortion, and gathered as input variable affecting parameter.

According to clinical experience in conjunction with the results of univariate logistic analysis, introduce 18 class risk factors and enter model as independent variable, comprise male/female side's schooling, male/female side's occupation type, male/female side's age, the wife's side previously whether conceived, whether whether using contraceptives, the wife's side to take food, whether the whether whether smoking of apocleisis vegetables, the wife's side of meat egg, the wife's side, the wife's side contact whether second-hand-cigarette, the wife's side drink, wife's side blood group, wife's side haemoglobin check the value etc.Introduce altogether 4182 routine samples and carry out models fitting, wherein 2889 examples (70%) are for model creation, and 1293 examples (30%) are for model testing.

The normalized of data

Utilize following formula to carry out normalized to input variable:

Normalization formula: V _new=(V _old-V _min)/(V _max-V _min).

V wherein _newthe variate-value after normalization, V _oldthe original variable value, V _maxto obtain maximal value, V in the original variable group _minto obtain minimum value in the original variable group.

In the present invention, be about to all input variables and carry out normalized.

To normalized data weighting summation

Each node is sued for peace after the input variable of this node and respective weights need to being multiplied each other.

The weighted sum formula:

$A_j(\stackrel{\‾}{x},\stackrel{\‾}{w})={\mathrm{\Σ}}_{i=0}^n{x}_i{w}_{\mathrm{ji}}$

X _imean input, w _jimean corresponding weights, A _jmean the output of j node.

Set up excitation function

After node is weighted summation, need to be to being processed with value, even pass through excitation function with value.The excitation function of selecting in the present invention is the SIGMOID function, has guaranteed the property led of function.

Excitation function formula (sigmoid function):

$O_j(\stackrel{\‾}{x},\stackrel{\‾}{w})=\frac{1}{1+e^{A_j(\stackrel{\‾}{x},\stackrel{\‾}{w})}}$

Oj means the output valve of j node after excitation function

The training of artificial neural network

Just need to be trained artificial neural network afterwards obtaining final output, be enabled to there is simple judgement as the people.

The neural metwork training error calculation formula:

$E(\stackrel{\‾}{x},\stackrel{\‾}{w},\stackrel{\‾}{d})=\underset{j}{\mathrm{\Σ}}{(O_j(\stackrel{\‾}{x},\stackrel{\‾}{w})-d_j)}^2$

E means the error of whole neural network.

The foundation of optimum artificial neural network

According to the variation of output valve, error is carried out to differentiate to weights, thereby can be finely tuned weights, after making each fine setting, error amount can reduce.

$\mathrm{\Δ}{w}_{\mathrm{ji}}=-\mathrm{\η}\frac{\∂E}{\∂w_{\mathrm{ji}}}$

Wherein η is the learning rate of network, and learning rate is higher, and the convergence of neural network is faster, but may miss optimum solution.

In the models fitting process, the contribution margin of input variable is sorted, result shows previously conceived history, adopts the measure that avoids conception and control birth, men and women side's age, wife's side schooling, wife's side occupation, whether the high-risk contribution degree for models fitting is larger.

Utilize experimenter's performance curve (receiver operating characteristic curve, be called for short the ROC curve) fitting effect is checked, the ROC area under curve is 65%, in test data, the correct Prediction ratio is about 61.3%, for the correct Prediction rate that the spontaneous abortion crowd occurs, is wherein 65.4%.

The mode of whole spontaneous abortion Risk Forecast System comprises:

1 original optimization model creates

By the pregnant front eugenic health examination data of input, layering, the neuron of system intelligence decision modified artificial neural network are chosen and training process, set up optimal neural network model, realize the intelligent predicting to the pregnant front risk of spontaneous abortion.

2 new samples inputs

The pregnant front eugenic health examination related data that the input new samples is concentrated.

3 risk models calculate

Utilize the optimal neural network model created automatically calculate and differentiate the risk of these data.

4 assessment results show

Utilize information to represent the value-at-risk that module will calculate automatically and differentiate result and offer the service object.

The foregoing is only preferred embodiment of the present invention, not be used for limiting the scope of the invention; The claim of protection scope of the present invention in claims limits, and every according to inventing the equivalence variation of doing and revising, all within the protection domain of patent of the present invention.

Claims (8)

1. a spontaneous abortion Risk Forecast System comprises:

Input media, for inputting the parameter that affects that affects spontaneous abortion;

Prediction unit, its parameter that affects based on inputted is predicted the spontaneous abortion risk;

Output unit, for exporting spontaneous abortion risk profile result.

2. spontaneous abortion Risk Forecast System as claimed in claim 1, is characterized in that, described input media is connected with prediction unit by internet with output unit.

3. spontaneous abortion Risk Forecast System as claimed in claim 1, is characterized in that, described input media is connected with prediction unit by wired mode and/or wireless mode with output unit.

4. spontaneous abortion Risk Forecast System as claimed in claim 1, is characterized in that, described input media is connected with prediction unit by fixed telephone network and/or cell phone network with output unit.

5. spontaneous abortion Risk Forecast System as described as any one in claim 1-4, is characterized in that, described output unit is audio output device.

6. spontaneous abortion Risk Forecast System as described as any one in claim 1-4, is characterized in that, described prediction unit is the artificial neural network prediction unit.

7. a method that builds the spontaneous abortion Risk Forecast System, comprise the steps:

1) data of input are carried out to normalized;

2) data after normalized are weighted to summation, the weighted sum formula is:

$A_j(\stackrel{\‾}{x},\stackrel{\‾}{w})={\mathrm{\Σ}}_{i=0}^n{x}_i{w}_{\mathrm{ji}}$

X wherein _imean input, w _jimean corresponding weights, A _jmean the output of j node;

3) set up excitation function, the excitation function formula is:

$O_j(\stackrel{\‾}{x},\stackrel{\‾}{w})=\frac{1}{1+e^{A_j(\stackrel{\‾}{x},\stackrel{\‾}{w})}}$

Wherein, O _jmean the output valve of j node after excitation function;

4) training of neural network, the neural metwork training error calculation formula:

$E(\stackrel{\‾}{x},\stackrel{\‾}{w},\stackrel{\‾}{d})=\underset{j}{\mathrm{\Σ}}{(O_j(\stackrel{\‾}{x},\stackrel{\‾}{w})-d_j)}^2$

Wherein, E means the error of whole neural network;

5) foundation of optimal network model, according to the variation of output valve, carry out differentiate to error to weights, thereby can be finely tuned weights, and after making each fine setting, error amount can reduce.

$\mathrm{\Δ}{w}_{\mathrm{ji}}=-\mathrm{\η}\frac{\∂E}{\∂w_{\mathrm{ji}}}$

Wherein η is the learning rate of network, and learning rate is higher, and the convergence of neural network is faster, but may miss optimum solution.

8. the method for structure spontaneous abortion Risk Forecast System as claimed in claim 7, it is characterized in that, in step 1) in choose following parameter as affecting parameter: male/female side's schooling, male/female side's occupation type, male/female side's age, the wife's side previously whether conceived, whether whether using contraceptives, the wife's side to take food, whether the whether whether smoking of apocleisis vegetables, the wife's side of meat egg, the wife's side, the wife's side contact whether second-hand-cigarette, the wife's side drink, wife's side blood group and wife's side haemoglobin check the value.

Images