CN103116707A - Heart disease intelligent diagnostic method based on case reasoning - Google Patents

Abstract

The invention discloses a heart disease intelligent diagnostic method based on case reasoning. The heart disease intelligent diagnostic method based on case reasoning is characterized in that: using various check indexes of a patient as a case to conduct searching in a case base, fining out a record which is most similar to the case as a diagnostic result; and meanwhile using water filling theory to obtain an allocation scheme of various characteristic attribute weights of the case, and rejecting redundant attributes according to the attribute weights, and thereby improving accuracy and speed of the heart disease diagnostic result. The heart disease intelligent diagnostic method comprises the following steps: defining a case expression form; constructing a history case set; using the water filling theory to conduct allocation of the characteristic attribute weights; extracting characteristic attributes according to the allocation result of the attribute weights; calculating similarity of cases; reusing the matched cases; and amending the cases. The heart disease intelligent diagnostic method based on case reasoning improves searching strategies of the case reasoning, conducts attribute reduction link through optimal allocation of the characteristic attribute weights of the cases, and ensures diagnosis accuracy and speed.

Description

A kind of heart disease intelligent diagnosing method based on reasoning by cases

Technical field

The present invention relates to a kind of disease intelligent diagnosing method, particularly for cardiopathic intelligent diagnosing method based on reasoning by cases.

Technical background

In medical field, the health that the contour danger disease of disease such as heart disease is threatening people constantly, the diagnostic procedure of this class disease has singularity and complicacy, the sick complexity of planting is various, otherness between individual patients has been brought difficulty to diagnosis in addition, and the high level medical expert is limited in China, and this just makes the cardiac can't obtain high-caliber diagnosis and treatment thereof.As emergence and the fast development of the artificial intelligence of one of the world's three large sophisticated technologies, its every intellectual technology has penetrated into each research field.This also makes the medical decision making support, comprises that the medical diagnosis field obtains paying close attention to fully and development, engineering in medicine and artificial intelligence study's fusion diagnosis method, simulated medical diagnosis, the thought process of prediction can provide diagnosis for the doctor, the secondary outcome for the treatment of and assessment.At present, the methods for the diagnosis of diseases of intelligence has obtained the attention of domestic and international medical domain.

The important indicator of estimating the medical diagnostic method quality mainly contains two of the rapidities of the accuracy of diagnosis and diagnosis, how fast and accurately disease to be made the emphasis that diagnosis is research.Both at home and abroad the research at medical diagnosis system is mainly concentrated on the method for building up of diagnostic model is studied, diagnostic model commonly used mainly contains at present:

1, artificial neural network: formed a highly interconnected and mutual processing unit system by the differentiation of Neurobiology model, artificial neural network similar to human brain can adapt to new input data by dynamic adjustment of training study, and artificial neural network has been successfully applied in the diagnosis of the diseases such as cancer, miocardial infarction, thyroid gland at present.

2, classification regression tree: a kind of analytical approach of tree construction, it is a kind of powerful diagnostic model instrument, especially for the processing of classification problem, very remarkable aspect the interaction that classification regression tree model implies between the excavation variable, and can while treatment classification data and successional data.This also makes classification regression tree model be widely used in the medical diagnosis field.

3, differential analysis and logistic regression: the differential analysis method is one of method that is applied to the earliest medical diagnosis, it is a kind of method of classification of determining by the degree of association of each base attribute of independent analysis, the logistic regression algorithm is to concentrate from large sample a kind of method of determining the optimum prediction subset, simultaneously also can with forward-backward algorithm progressively choosing method be combined to determine whether the patient suffers from disease, its advantage is that it does not need data is these hypothesis of normality, therefore makes the method be widely used.

4, RBR: by the thinking activities of simulative medicine expert diagnosis disease treatment, thereby carry out diagnosis and the auxiliary curing that reasoning and judging realizes disease, become at present the research emphasis of biomedical engineering.

Said method has been realized the function of medical assistance decision-making to a certain extent, has obtained at present significant progress and application, yet has separately some drawbacks such as neural network to need a large amount of training datas and mathematical model to be difficult for setting up, and easily is absorbed in local optimum; And the RBR formulation of rule at present depends on the expert, and the rule that different experts provides is different, the bottleneck problem that exists Rules Knowledge Acquisition to be difficult for.

drawback for the said method model, reasoning by cases becomes study hotspot in recent years, reasoning by cases is a kind of new inference technology emerging in artificial intelligence field, be applicable to not formalization fully, the prevailing field of INFORMATION OF INCOMPLETE and experimental knowledge, the appearance of reasoning by cases has overcome the acquirement of expert knowledge hard problem, its main thought is to retrieve the case the most close with present case its result is applied to present case in the case library of passing processing, its reasoning process mainly comprises: Case Retrieval, the case correction, case is reused, case storage Four processes, and in Four processes Case Retrieval be the key link that affects the reasoning by cases quality, also that reasoning by cases is applied in the committed step in medical diagnosis, for guaranteeing accuracy and the reliability of medical diagnosis on disease, the precision and the speed that improve Case Retrieval are core links.At first for improving the precision of retrieval, top priority is that the weight of case characteristic attribute is rationally determined.Secondly on the problem that improves retrieval rate, because characteristic attribute often has dividing of primary and secondary, some characteristic attribute there is no very important meaning to the diagnosis of disease just can delete to improve retrieval rate with such attribute this moment, the yojan of characteristic attribute mainly contains the method for extracting character subset and utilizing Attribute Significance to choose, and is to improve the key point of diagnosis efficiency to the reasonable distribution of case characteristic attribute weight from the above mentioned.The method of determining weight mainly contains subjective method and objective approach, and the power method of deciding that subjective analysis is commonly used has: expert consulting method, investigation statistics method, indifference eclectic method, relevant function method etc.Subjective method depends on certain domain expert's priori and subjective judgement, has randomness, impacts can for undoubtedly the accuracy of Similar case search.At present, the method that objective approach is optimized weight has: genetic algorithm, rough set, neural network etc., but these methods all exist self some intrinsic defective, such as neural network structure is difficult for determining and needing a large amount of training samples; Genetic algorithm easily is absorbed in local minimum; Discretize to attribute in rough set is comparatively difficult.

In sum, during the application case inference method, at first to solve the assignment problem of case attribute weight; And instruct case attribute reduction problem to guarantee the operation of the stability and high efficiency of medical diagnosis system with this according to the weight allocation result that obtains.

Summary of the invention

The object of the invention is to, a kind of heart disease intelligent diagnosing method based on reasoning by cases is provided, with every Index for examination of patient as searching in a case routine storehouse on record, find out with its most close record as diagnostic result, meanwhile utilize water-filling to obtain the allocative decision of each characteristic attribute weight of case, and come the attribute of eliminate redundancy according to the weight of attribute.Thereby improve precision and the speed of heart disease diagnosis result.

The present invention completes by following steps:

Step 1: definition case representation form;

Data target to be diagnosed is expressed as the treatable form of reasoning by cases system, and the attribute that specifically follow-up conclusion example is added diagnostic result with following attribute is to being described:

C=(x ₁,x ₂,...,x _n;D) (1)

Wherein, x ₁, x ₂..., x _nN the index that obtains of diagnosing in the expression case, D represents the diagnostic result of this case.

Step 2: build historical casebook;

Utilize above-mentioned representation that the idagnostic logout of history is described:

C _k=(x _k1,x _k2,...,x _kn;D _k),k=1,2,3...,m (2)

Wherein k represents the number of historical case.

Step 3: utilize water-filling characteristic attribute to be carried out the distribution of weight;

Standard deviation and the variance of calculating each characteristic attribute value are:

$\left\{\begin{array}{c}{\mathrm{\σ}}_i=\sqrt{1/m\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{(x_{\mathrm{ki}}-{\mathrm{\μ}}_i)}^2}\\ {\mathrm{\μ}}_i=1/m\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{x}_{\mathrm{ki}}\end{array}\right.---\left(3\right)$

Wherein, σ _iThe standard deviation of i attribute, μ _iIt is the average of i attribute; As follows as evaluation index structure attribute capacity function with this:

$C_W=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\log }_2(1+{\mathrm{\μ}}_i{\mathrm{\ω}}_i/{\mathrm{\σ}}_i)---\left(4\right)$

Wherein, C _WExpression attribute capacity is used for weighing the size of attribute institute inclusion information amount, ω _iRepresent that namely i characteristic attribute can distribute the weighted value that obtains, its value satisfies

Following formula is constructed Lagrange's equation and weight is carried out differentiate as follows for obtaining optimum weight allocation scheme:

$L(\mathrm{\ω},\mathrm{\λ})=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\log }_2(1+{\mathrm{\μ}}_i{\mathrm{\ω}}_i/{\mathrm{\σ}}_i)+\mathrm{\λ}(1-\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\ω}}_i)$

$\left\{\begin{array}{c}\∂L/{\∂\mathrm{\ω}}_1=(1/\ln 2)[{\mathrm{\μ}}_i/({\mathrm{\σ}}_i+{\mathrm{\ω}}_1{\mathrm{\μ}}_i)]=0\\ \∂L/{\∂\mathrm{\ω}}_2=(1/\ln 2)[{\mathrm{\μ}}_i/({\mathrm{\σ}}_i+{\mathrm{\ω}}_2{\mathrm{\μ}}_i)]=0\\ ................\\ \∂L/{\∂\mathrm{\ω}}_n=(1/\ln 2)[{\mathrm{\μ}}_i/({\mathrm{\σ}}_i+{\mathrm{\ω}}_n{\mathrm{\μ}}_i)]=0\end{array}\right.---\left(5\right)$

Finally, the computing formula by the definite case attribute weight of water-filling is:

ω _i=max(0,ε-σ _i/μ _i) (6)

Wherein, σ _iThe standard deviation of i attribute value set, μ _iIt is the average of i attribute value set; ε is the threshold value of case attribute weight.

Step 4: the allocation result by attribute weight is extracted characteristic attribute;

The characteristic attribute weight is arranged as ω according to from big to small order _i(i=1,2,3 ..., n; ω _iω _i-1), setting the Attribute Significance threshold value is ω _d, with the n of importance degree greater than this threshold value ₁(n ₁≤ n) individual attributes extraction out will remain importance degree and reject less than the attribute of threshold value and complete the link of choosing of attribute, and the weight of each characteristic attribute after yojan is done corresponding adjustment:

${\mathrm{\ω}}_j={\mathrm{\ω}}_i/\underset{i=0}{\overset{n_1}{\mathrm{\Σ}}}{\mathrm{\ω}}_i,(j=\mathrm{1,2,3},...,n_1;j=i)---\left(7\right)$

Step 5: calculate case similarity;

Each record in target case and case library is carried out the calculating of similarity, similarity utilizes following formula to be described:

S(T,X _k)=1-Ω(T,X _k),k=1,2,...,m (8)

Wherein, k is the number of historical case, and T represents target case, X _kRepresent k historical case history, Ω is the Euclidean distance between target case and historical case, and available following formula obtains:

$\mathrm{\Ω}(T,X_k)={\left(\underset{j=1}{\overset{n_1}{\mathrm{\Σ}}}{\mathrm{\ω}}_j{(t_j-x_{\mathrm{kj}})}^2\right)}^{1/2}---\left(9\right)$

Wherein, the weights omega of characteristic attribute variable _j, satisfy:

$\underset{j=1}{\overset{n_1}{\mathrm{\Σ}}}{\mathrm{\ω}}_j,=1---\left(10\right)$

Step 6: the case that coupling is obtained is reused;

Can find the some historical cases close with present case by Case Retrieval, the threshold value of setting similarity is S _v∈ (0,1], the coupling case number that obtains retrieving take this threshold value is as l, and similarities of these coupling cases are:

$\left\{\begin{array}{c}{S}_o,o=\mathrm{1,2},\·\·\·,l\\ s.t.S_o\≥S_v\end{array}\right.---\left(11\right)$

The diagnostic result of these cases just can be reused in new case and go so, and the diagnostic result of target case is so:

$D_T=\left(\underset{o=1}{\overset{l}{\mathrm{\Σ}}}{s}_o{D}_o\right)/\underset{k=o}{\overset{l}{\mathrm{\Σ}}}{s}_o---\left(12\right)$

D wherein _T, D _oThe diagnostic result of the close case that represents respectively new case and retrieve.

Step 7: the correction of case;

The diagnostic result that obtains is estimated, if diagnostic result is exported in satisfaction, otherwise this case is revised, by adjusting similarity threshold S _v∈ (0,1] obtain different coupling cases and go to step 6.

The present invention compared with prior art has following obvious advantage and beneficial effect:

The present invention is applied to can realize the high speed to case in real heart disease diagnosis, efficient diagnosis has not only improved the degree of accuracy of diagnosis, and reduced the diagnosis required time, reached the effect that the present invention is successfully applied to heart disease diagnosis.

For the high-risk disease of this class of heart disease, its diagnostic procedure has singularity and complicacy, sick plant complicated various of heart disease, add the otherness between individual patients, make cardiopathic pathological characters, the variation of clinical symptoms is intricate, for diagnosis has brought great difficulty, and the intelligent diagnosing method that the thought of utilizing reasoning by cases is set up can obtain corresponding diagnostic knowledge by the experience of obtaining and the data in study patient medical processing procedure, and then can make diagnosis accurately to disease.The index that is used for evaluation diagnostic method quality is mainly precision and the speed of diagnosis, this is also the Focal point and difficult point in heart disease diagnosis, the present invention has improved the search strategy in the reasoning by cases for these two indexs, distribute and carry out the attribute reduction link with this by the optimization to the case characteristic attribute weight, having guaranteed diagnostic accuracy and speed.

Description of drawings

Fig. 1 is structure principle chart of the present invention;

Fig. 2 is the schematic flow sheet that the present invention is based on the heart disease intelligent diagnosing method of reasoning by cases.

Embodiment

Below in conjunction with Figure of description, the specific embodiment of the present invention is described in detail.

Consult shown in Figure 1ly, be structure principle chart of the present invention.Case representation form according to definition is inputted the heart disease diagnosis index, retrieve the most close with it case in case library, and reuse the diagnostic result that provides present case by case, and this result is made evaluation, if dissatisfied carry out the correction of case, until find satisfied case result, simultaneously case is stored this diagnostic result output in case library finding the solution for new problem at last.

Shown in Figure 2, be the schematic flow sheet of heart disease intelligent diagnosing method of the present invention.As can be seen from the figure, concrete implementation step is as described in process flow diagram:

Step 1: definition case representation form;

Step 2: build historical casebook;

Step 3: utilize water-filling characteristic attribute to be carried out the distribution of weight;

Step 4: the choosing of characteristic attribute, set the importance degree threshold value, if the attribute weight value that is obtained by step 3 keeps greater than the threshold value of setting and upgrades case library; Otherwise, with this attribute deletion;

Step 5: carry out the retrieval of case;

Step 6: reusing of diagnosis scheme obtains the diagnostic result of target case;

Step 7: the result to diagnosis is estimated, if satisfied with this diagnostic result output, otherwise this diagnosis case is revised accordingly, and change step 6 over to.

Major programme is as described below:

Case history representation in Definition Model is:

C=(x ₁,x ₂,...,x _n;D) (1)

Wherein, x ₁, x ₂..., x _nN the index that obtains of diagnosing in the expression case, D represents the diagnostic result of this case.Utilize above-mentioned representation that the idagnostic logout of history is described:

C _k=(x _k1,x _k2,...,x _kn;D _k),k=1,2,3...,m (2)

Wherein k represents the number of historical case.

The key link of reasoning by cases is Case Retrieval, its objective is the case history the most close with the target case that search in case library, specifically each record in target case and case library is carried out the calculating of similarity, similarity utilizes following formula to be described:

S(T,X _k)=1-Ω(T,X _k),k=1,2,...,m (3)

Wherein, k is the number of historical case, and T represents target case, T _kRepresent k historical case history, Ω is the Euclidean distance between target case and historical case, and available following formula obtains:

$\mathrm{\Ω}(T,X_k)={\left(\underset{j=1}{\overset{n_1}{\mathrm{\Σ}}}{\mathrm{\ω}}_j{(t_j-x_{\mathrm{kj}})}^2\right)}^{1/2}---\left(4\right)$

Wherein, the weights omega of characteristic attribute variable _j, satisfy:

$\underset{j=1}{\overset{n_1}{\mathrm{\Σ}}}{\mathrm{\ω}}_j=1---\left(5\right)$

The weight allocation of characteristic attribute affects the precision of Case Retrieval and then has influence on cardiopathic diagnostic result, for guaranteeing precision and the speed of diagnosis, the thought that the present invention proposes water-filling is distributed the characteristic attribute weight, and instruct the attribute reduction link with the allocation result of weight, be implemented as follows:

Standard deviation and the variance of calculating each characteristic attribute value are:

$\left\{\begin{array}{c}{\mathrm{\σ}}_i=\sqrt{1/m\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{(x_{\mathrm{ki}}-{\mathrm{\μ}}_i)}^2}\\ {\mathrm{\μ}}_i=1/m\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{x}_{\mathrm{ki}}\end{array}\right.---\left(6\right)$

Wherein, σ _iThe standard deviation of i attribute, μ _iIt is the average of i attribute; As follows as evaluation index structure attribute capacity function with this:

$C_W=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\log }_2(1+{\mathrm{\μ}}_i{\mathrm{\ω}}_i/{\mathrm{\σ}}_i)---\left(7\right)$

Wherein, C _WExpression attribute capacity is used for weighing the size of attribute institute inclusion information amount, ω _iRepresent that namely i characteristic attribute can distribute the weighted value that obtains, its value satisfies

Following formula is constructed Lagrange's equation and weight is carried out differentiate as follows for obtaining optimum weight allocation scheme:

$L(\mathrm{\ω},\mathrm{\λ})=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\log }_2(1+{\mathrm{\μ}}_i{\mathrm{\ω}}_i/{\mathrm{\σ}}_i)+\mathrm{\λ}(1-\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\ω}}_i)$

$\left\{\begin{array}{c}\∂L/{\∂\mathrm{\ω}}_1=(1/\ln 2)[{\mathrm{\μ}}_i/({\mathrm{\σ}}_i+{\mathrm{\ω}}_1{\mathrm{\μ}}_i)]=0\\ \∂L/{\∂\mathrm{\ω}}_2=(1/\ln 2)[{\mathrm{\μ}}_i/({\mathrm{\σ}}_i+{\mathrm{\ω}}_2{\mathrm{\μ}}_i)]=0\\ ................\\ \∂L/{\∂\mathrm{\ω}}_n=(1/\ln 2)[{\mathrm{\μ}}_i/({\mathrm{\σ}}_i+{\mathrm{\ω}}_n{\mathrm{\μ}}_i)]=0\end{array}\right.---\left(8\right)$

Finally, the computing formula by the definite case attribute weight of water-filling is:

ω _i=max(0,ε-σ _i/μ _i) (9)

Wherein, σ _iThe standard deviation of i attribute value set, μ _iIt is the average of i attribute value set; ε is the threshold value of case attribute weight.

The characteristic attribute weight is arranged as ω according to from big to small order _i(i=1,2,3 ..., n; ω _iω _i-1), setting the Attribute Significance threshold value is ω _d, with the n of importance degree greater than this threshold value ₁(n ₁≤ n) individual attributes extraction out will remain importance degree and reject less than the attribute of threshold value and complete the link of choosing of attribute, and the weight of each characteristic attribute after yojan is done corresponding adjustment:

${\mathrm{\ω}}_j={\mathrm{\ω}}_i/\underset{i=0}{\overset{n_1}{\mathrm{\Σ}}}{\mathrm{\ω}}_i,(j=\mathrm{1,2,3},...,n_1;j=i)---\left(10\right)$

Can find the some historical cases close with present case by Case Retrieval, the threshold value of setting similarity is S _v∈ (0,1], the coupling case number that obtains retrieving take this threshold value is as l, and similarities of these coupling cases are:

$\left\{\begin{array}{c}{S}_o,o=\mathrm{1,2},\·\·\·,l\\ s.t.S_o\≥S_v\end{array}\right.---\left(11\right)$

The diagnostic result of these cases just can be reused in new case and go so, and the diagnostic result of target case is so:

$D_T=\left(\underset{o=1}{\overset{l}{\mathrm{\Σ}}}{s}_o{D}_o\right)/\underset{k=o}{\overset{l}{\mathrm{\Σ}}}{s}_o---\left(12\right)$

D wherein _T, D _oThe diagnostic result of the close case that represents respectively new case and retrieve.

In order to verify validity of the present invention, utilize the data of heart disease diagnosis to form case library.This data set has 297 cases, comprises 76 attributes (the present embodiment adopted wherein 13 attributes process) and 5 diagnostic results (representing whether be heart disease and the cardiopathic order of severity occurs with 0-4 respectively).Concrete data are as shown in table 1, utilize the weighted value of 13 attributes that water-filling method that the present invention proposes asks to be respectively 0.0272,0.0518,0.0350,0.0340,0.0350,0.2176,0.0772,0.0206,0.1253,0.0734,0.0879,0.1238,0.0913.Carry out on this basis following experiment.

Table 1 heart disease diagnosis data

The heart disease intelligent diagnosing method based on reasoning by cases that the present invention is proposed compares from the accuracy rate that the diagnostic method of routine obtains under different test sets, and it is 0.21 that the present invention sets the importance degree threshold value, therefore use 12 attributes and with carry out the contrast of working time without the conventional method of yojan, result is as shown in table 2.

The contrast of table 2 heart disease diagnosis result

By table 2, can find out, the present invention has improved approximately 2% on cardiopathic diagnostic accuracy, descended widely on speed.Application of the present invention has improved precision and the speed of heart disease diagnosis, has realized cardiopathic intelligent diagnostics.

Claims (2)

1. heart disease intelligent diagnosing method based on reasoning by cases, it is characterized in that, the heart disease diagnosis data are inputted as case, diagnostic method based on reasoning by cases can retrieve the most close with it case according to the index of current input in case library, and make the diagnostic result of present case by reusing of case, the result that obtains is compared evaluation, carry out the correction of case until find satisfied solution with unsatisfied, export diagnostic result and the storage of this case this moment; Comprise the following steps:

Step 1: definition case representation form;

Data target to be diagnosed is expressed as the treatable form of reasoning by cases system, adds the attribute of diagnostic result with following attribute to being described:

C=(x ₁,x ₂,...,x _n;D) (1)

Wherein, x ₁, x ₂..., x _nN the index that obtains of diagnosing in the expression case, D represents the diagnostic result of this case.

Step 2: build historical casebook;

Utilize above-mentioned representation that the idagnostic logout of history is described:

C _k=(x _k1,x _k2,...,x _kn;D _k),k=1,2,3...,m (2)

Wherein k represents the number of historical case;

Step 3: utilize water-filling characteristic attribute to be carried out the distribution of weight;

Standard deviation and the variance of calculating each characteristic attribute value are:

$\left\{\begin{array}{c}{\mathrm{\σ}}_i=\sqrt{1/m\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{(x_{\mathrm{ki}}-{\mathrm{\μ}}_i)}^2}\\ {\mathrm{\μ}}_i=1/m\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{x}_{\mathrm{ki}}\end{array}\right.---\left(3\right)$

Wherein, σ _iThe standard deviation of i attribute, μ _iIt is the average of i attribute; As follows as evaluation index structure attribute capacity function with this:

$C_W=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\log }_2(1+{\mathrm{\μ}}_i{\mathrm{\ω}}_i/{\mathrm{\σ}}_i)---\left(4\right)$

Wherein, C _WExpression attribute capacity is used for weighing the size of attribute institute inclusion information amount, ω _iRepresent that namely i characteristic attribute can distribute the weighted value that obtains, its value satisfies Following formula is constructed Lagrange's equation and weight is carried out differentiate as follows for obtaining optimum weight allocation scheme:

$L(\mathrm{\ω},\mathrm{\λ})=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\log }_2(1+{\mathrm{\μ}}_i{\mathrm{\ω}}_i/{\mathrm{\σ}}_i)+\mathrm{\λ}(1-\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\ω}}_i)$

$\left\{\begin{array}{c}\∂L/{\∂\mathrm{\ω}}_1=(1/\ln 2)[{\mathrm{\μ}}_i/({\mathrm{\σ}}_i+{\mathrm{\ω}}_1{\mathrm{\μ}}_i)]=0\\ \∂L/{\∂\mathrm{\ω}}_2=(1/\ln 2)[{\mathrm{\μ}}_i/({\mathrm{\σ}}_i+{\mathrm{\ω}}_2{\mathrm{\μ}}_i)]=0\\ ................\\ \∂L/{\∂\mathrm{\ω}}_n=(1/\ln 2)[{\mathrm{\μ}}_i/({\mathrm{\σ}}_i+{\mathrm{\ω}}_n{\mathrm{\μ}}_i)]=0\end{array}\right.---\left(5\right)$

Finally, the computing formula by the definite case attribute weight of water-filling is:

ω _i=max(0,ε-σ _i/μ _i) (6)

Wherein, σ _iThe standard deviation of i attribute value set, μ _iIt is the average of i attribute value set; ε is the threshold value of case attribute weight;

Step 4: the choosing of characteristic attribute;

The characteristic attribute weight is arranged as ω according to from big to small order _i(i=1,2,3 ..., n; ω _iω _i-1), setting the Attribute Significance threshold value is ω _d, with the n of importance degree greater than this threshold value ₁(n ₁≤ n) individual attributes extraction out will remain importance degree and reject less than the attribute of threshold value and complete the link of choosing of attribute, and the weight of each characteristic attribute after yojan is done corresponding adjustment:

${\mathrm{\ω}}_j={\mathrm{\ω}}_i/\underset{i=0}{\overset{n_1}{\mathrm{\Σ}}}{\mathrm{\ω}}_i,(j=\mathrm{1,2,3},...,n_1;j=i)---\left(7\right)$

Step 5: the retrieval of case;

Each record in target case and case library is carried out the calculating of similarity, similarity utilizes following formula to be described:

S(T,X _k)=1-Ω(T,X _k),k=1,2,...,m (8)

Wherein, k is the number of historical case, and T represents target case, X _kRepresent k historical case history, Ω is the Euclidean distance between target case and historical case, and available following formula obtains:

$\mathrm{\Ω}(T,X_k)={\left(\underset{j=1}{\overset{n_1}{\mathrm{\Σ}}}{\mathrm{\ω}}_j{(t_j-x_{\mathrm{kj}})}^2\right)}^{1/2}---\left(9\right)$

Wherein, the weights omega of characteristic attribute variable _j, satisfy:

$\underset{j=1}{\overset{n_1}{\mathrm{\Σ}}}{\mathrm{\ω}}_j,=1---\left(10\right)$

Step 6: the reusing of case;

Can find the some historical cases close with present case by Case Retrieval, the threshold value of setting similarity is S _v∈ (0,1], the coupling case number that obtains retrieving take this threshold value is as l, and similarities of these coupling cases are:

$\left\{\begin{array}{c}{S}_o,o=\mathrm{1,2},\·\·\·,l\\ s.t.S_o\≥S_v\end{array}\right.---\left(11\right)$

The diagnostic result of these cases just can be reused in new case and go so, and the diagnostic result of target case is so:

$D_T=\left(\underset{o=1}{\overset{l}{\mathrm{\Σ}}}{s}_o{D}_o\right)/\underset{k=o}{\overset{l}{\mathrm{\Σ}}}{s}_o---\left(12\right)$

D wherein _T, D _oThe diagnostic result of the close case that represents respectively new case and retrieve;

Step 7: the correction of case;

The diagnostic result that obtains is estimated, if diagnostic result is exported in satisfaction, otherwise this case is revised, by adjusting similarity threshold S _v∈ (0,1] obtain different coupling cases and go to step 6.

2. a kind of heart disease intelligent diagnosing method based on reasoning by cases according to claim 1, it is characterized in that: described heart disease diagnosis data comprise: age, sex, the type of having a chest pain, blood pressure, cholesterol, fasting blood-glucose, cardiogram result, maximum heart rate.

Images