CN114822875B - Disease and medicine matching method based on naive Bayesian network - Google Patents

Abstract

The invention discloses a matching method for disease and medicine based on naive Bayesian network, wherein disease symptoms ( ^x1 , ^x2 , ..., ^x1 , ..., ^x1 ) are used as input nodes, each of which represents a disease symptom; medicines ( ^y1 , ^y2 , ..., ^yj , ..., ^yJ ) are used as output nodes, each of which represents a medicine; the connection between the input node and the output node represents the mapping relationship between the disease symptoms and the medicine. The disease symptom node and the medicine node are represented by conditional probability, and the conditional probability is used to reflect the probability strength between each input node and the output node. The greater the conditional probability, the stronger the correlation between the disease symptoms and the medicine at both ends of the connection, that is, the more suitable the medicine is for treating the disease symptoms. Through the invention, the doctor only needs to determine the disease symptoms of the patient, which can provide a reference basis for determining the treatment plan, recommend reasonable, effective and efficient medicines to the doctor, and reduce the situation of irrational use of medicines.

Description

Matching method between disease and medicine based on naive Bayesian network

Technical field:

The present invention relates to the field of biomedicine technology, and in particular to a method for matching symptoms and drugs based on a naive Bayesian network.

Background technology:

In the traditional Chinese and Mongolian medicine treatment process, doctors need to prescribe medicine for patients based on what they have learned. However, due to the complexity and diversity of symptoms and the efficacy of medicinal materials, the doctor is highly dependent on his or her experience when prescribing drugs. This leads to irrational use of drugs, which will directly affect the treatment effect and even the health of the patient.

At present, there are many methods to study the relationship between the drugs in the prescription and the diseases they treat, and most of them use the method of decomposing the prescription, such as the drug pair research method, the single drug research method, etc. This kind of method can obviously only change one or two drugs while keeping other drugs unchanged, explore the relationship between the drugs in the prescription and the disease, and study one or two drugs for the diseases treated in the prescription. Obviously, it is not possible to accurately conduct a comprehensive study on the nonlinear mapping relationship between the diseases treated in the prescription and the drugs from an overall perspective, and it is impossible to provide a reference for doctors to prescribe medicines, and it is impossible to efficiently and effectively reduce the situation of irrational drug use and the degree of dependence on doctors' experience.

Summary of the invention:

The purpose of the present invention is to provide a method for matching symptoms and drugs based on a naive Bayesian network, which can provide a reference for determining a treatment plan, recommend reasonable, effective and efficient drugs to doctors, reduce the situation of irrational use of drugs, reduce the degree of dependence on doctors' experience, and improve doctors' work efficiency.

The present invention is implemented by the following technical solutions:

The matching method of disease and medicine based on naive Bayesian network includes the following steps:

S1. Collect several traditional prescriptions, match the symptoms X corresponding to the prescriptions with the drugs Y used, and form a prescription sample set Z;

S2, using part of the recipes in the recipe sample set Z established in S1 as a training recipe set, binarizing the training recipe set to obtain a training set, using the training set to train the naive Bayesian network model, and calculating the probability P( ^xi ) of the disease symptoms, the probability P( ^yj ) of the drug, and the conditional probability P( ^xi =t| ^yj =l) between the disease symptoms and the drug in the training set, where t=0, 1 represents the two situations of the disease symptom ^xi not existing and existing, and l=0, 1 represents the two situations of the drug ^yj not being used and being used;

S3, taking the symptom X' of the case to be prescribed as an input parameter, and calculating the posterior probability P(y j =1| ^X ) of each drug being used and the posterior probability P(y ^j =0|X) of each drug not being used according ^to the probability P(xi) of the symptom, the probability P(y ^j ) of the drug, and the conditional probability P( ^xi =t|y ^j =l) between the symptom and the drug calculated in S2, wherein X=( ^x1 , ^x2 , ..., ^xi , ..., xI) represents each symptom of the patient, and the value of ^xi (i=1, 2, ..., ^I ) is 1 or 0;

S4. Compare P(y ^j =1|X) and P(y ^j =0|X). If P(y ^j =1|X)≥P(y ^j =0|X), then determine that the medicine is a medicine that matches the symptoms X' of the case. All medicines that match the symptoms X' of the case are combined to obtain a matching medicine result Y=(y ¹ , y ² , ..., y ^J ).

Preferably, in said S1,

Recipe sample set Z = {(X ₁ , Y ₁ ), (X ₂ , Y ₂ ), … , (X _n , Y _n ), … , (X _N , Y _N )},

Where N is the total number of recipes in the recipe sample set;

The corresponding symptom characteristics in the nth recipe are recorded as In vector It is the symptom of the i-th disease corresponding to the n-th recipe;

The drug combination in the nth prescription is recorded as In vector It is the jth medicine in the nth recipe.

Preferably, S2 includes the following steps: the method of binarizing the training prescription set is that in a prescription, if the disease symptom x ⁱ exists, the disease symptom x ⁱ is assigned a value of 1; if the disease symptom x ⁱ does not exist, the disease symptom x ⁱ is assigned a value of 0; similarly, if the prescribed prescription contains a drug y ^j , the drug y ^j is assigned a value of 1; if the prescribed prescription does not contain a drug y ^j , the drug y ^j is assigned a value of 0.

Preferably, S2 comprises the following steps:

(1) Calculate the probability of symptoms P( ^xi ) and the probability of drugs P( ^yj ) in the training set. Specifically,

In formula (1), represents the cumulative frequency of symptom x ⁱ = t in the training set, t = 0, 1 represents the two situations where symptom x ⁱ does not exist or exists, that is, t = 0 represents the absence of symptom x ⁱ in the nth recipe, t = 1 represents the presence of symptom x ⁱ in the nth recipe, N is the total number of recipes in the recipe sample set;

In formula (2), represents the cumulative frequency of drug y ^j = l in the training set; l = 0, 1 represents the two situations of drug y ^j not being used and being used, that is, l = 0 represents that drug y ^j is not used in the nth formula, l = 1 represents that drug y ^j is used in the nth formula, and N is the total number of formulas in the formula sample set;

(2) Calculate the probability P( ^xi = t, ^yj = l) when the disease symptom x ⁱ = t and the drug y ^j = l in the training set,

In formula (3), represents the cumulative frequency of prescriptions when the disease symptom x ⁱ = t and the drug y ^j = l in the training set;

(3) Calculate the conditional probability P( ^xi = t| ^yj = l) between the disease symptoms and the drug.

Substituting equations (2) and (3) into equation (4) we can find the conditional probability.

Preferably, in S3: the calculation method of the posterior probability P(y ^j =l|X) of each drug being used and not used is as follows:

In formula (5), X = ( ^x1 , ^x2 , ..., ^xi , ..., xI) represents the symptoms of various diseases of the patient, the value of ^xi (i = 1, 2, ..., ^I ) is 1 or 0; l = 0, 1 represents the two situations of drug ^yj not being used and using, that is, P( ^yj = 0|X) represents the posterior probability that drug ^yj is not used, that is, P( ^yj = 1|X) represents the posterior probability that drug ^yj is used.

Advantages of the present invention:

The present invention uses disease symptoms ( ^x1 , ^x2 , ..., ^x1 , ..., ^xI ) as input nodes, each of which represents a disease symptom; uses drugs ( ^y1 , ^y2 , ..., ^yj , ..., ^yJ ) as output nodes, each of which represents a drug; the connection between the input node and the output node represents the mapping relationship between the disease symptoms and the drug. The disease symptom nodes and the drug nodes are represented by conditional probabilities, and the conditional probabilities are used to reflect the probability strength between each input node and the output node. The greater the conditional probability, the stronger the correlation between the disease symptoms and the drug at both ends of the connection, that is, the more suitable the drug is for treating the disease symptom.

Through the present invention, doctors only need to determine the patient's symptoms and input the symptoms to output matching drug results, which can provide a reference for determining treatment plans, recommend reasonable, effective and efficient drugs to doctors, reduce irrational use of drugs, reduce dependence on doctors' experience, and improve doctors' work efficiency.

Description of the drawings:

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

FIG1 is a flow chart of a method for matching symptoms and drugs based on a naive Bayesian network according to the present invention;

FIG2 is a structural model diagram of the method for matching symptoms and drugs based on a naive Bayesian network according to the present invention.

Specific implementation method:

The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

Embodiment 1:

The matching method of disease and medicine based on naive Bayesian network as shown in FIG1 and FIG2 includes the following steps:

S1. Collect 24 traditional Mongolian medicine prescriptions, match the symptoms X corresponding to the prescriptions with the medicine Y used, and form a prescription sample set

Z={(X ₁ , Y ₁ ), (X ₂ , Y ₂ ), ..., (X _n , Yn ₎ , ..., (X ₂₄ , Y ₂₄ )}, as shown in Table 1.

Table 1 Recipe sample set

Table 1 involves 46 disease symptoms and 78 medicines. Table 2 lists the disease symptoms ^xi (i=1, 2, ..., 46) and medicines ^yj (j=1, 2, ..., 78).

Table 2 Symptoms and medications

S2. The first 22 recipes in the recipe sample set Z established in S1 are used as the training recipe set, as shown in Table 3; the last 2 are used as the test recipe set.

Table 3 Training recipe set

The symptoms ^xi and drugs ^yj in the training formula set in Table 3 are binarized. That is, in a formula, if the symptom ^xi exists, the symptom ^xi is assigned a value of 1; if the symptom ^xi does not exist, the symptom xi ^is assigned a value of 0; When , it means that the symptom of the i-th disease exists in the n-th prescription, that is, the drug combination in the n-th prescription can treat the symptom of the disease; when When , it means that the symptom of the ith disease does not exist in the nth prescription, that is, the drug combination in the nth prescription does not treat the symptom of the disease. Similarly, if the prescription contains drug y ^j , the drug y ^j is assigned a value of 1; if the prescription does not contain drug y ^j , the drug y ^j is assigned a value of 0; when When the jth drug is used in the nth formula, When , it means that the jth drug is not used in the nth formula. Finally, the training set shown in Table 4 is formed.

Table 4 Training set

The naive Bayesian network model is trained using the training set to calculate the probability P( ^xi ) of the symptom, the probability P( ^yj ) of the drug, and the conditional probability P( ^xi =t| ^yj =l) between the symptom and the drug in the training set, where t=0, 1 represents the two situations of the symptom ^xi not existing and existing, and l=0, 1 represents the two situations of the drug ^yj not being used and being used. Specifically, the following steps are included:

(1) Calculate the probability of symptoms P( ^xi ) and the probability of drugs P( ^yj ) in the training set. Specifically,

In formula (1), represents the cumulative frequency of symptom x ⁱ = t in the training set, t = 0, 1 represents the two situations that symptom x ⁱ does not exist and does not exist, that is, t = 0 represents the absence of symptom x ⁱ in the nth recipe, t = 1 represents the existence of symptom x ⁱ in the nth recipe, N is the total number of recipes in the recipe sample set; using formula (1), the probability of non-existence P (x ⁱ = 0) and the probability of existence P (x ⁱ = 1) of each symptom can be calculated.

In formula (2), represents the cumulative frequency of drug ^yj = l in the training set; l = 0, 1 represents the two situations of drug ^yj not being used and being used, that is, l = 0 represents that drug ^yj is not used in the nth formula, l = 1 represents that drug ^yj is used in the nth formula, and N is the total number of formulas in the formula sample set; Formula (2) can be used to calculate the probability of non-use P( ^yj = 0) and the probability of drug use P( ^yj = 1) of each drug.

(2) Calculate the probability P( ^xi = t, ^yj = l) when the disease symptom x ⁱ = t and the drug y ^j = l in the training set,

In formula (3), represents the cumulative frequency of prescriptions when the disease symptom x ⁱ = t and the drug y ^j = l in the training set;

Formula (3) can be used to calculate the probabilities of drugs and symptoms in the prescription training set being in different situations, which are the four probabilities of P( ^xi = 1, ^yj = 1), P( ^xi = 1, ^yj = 0), P( ^xi = 0, ^yj = 1) and P( ^xi = 0, ^yj = 0).

(3) Calculate the conditional probability P( ^xi = t| ^yj = l) between the disease symptoms and the drug.

Substitute the calculated data P( ^yj =1) and P( ^xi =1, yj=1) into formula (4) to calculate the conditional probability P( ^xi =1| ^yj =1) of drug use and the presence of disease symptoms in the training set; substitute the calculated data P( ^yj =1) and P( ^xi =0, ^yj =1) into formula (4) to calculate the conditional probability P( ^xi =0| ^{yj =} 1) of drug use and the absence of disease symptoms in the training set; substitute the calculated data P( ^yj =0) and P( ^xi =1, ^yj =0) into formula (4) to calculate the conditional probability P( ^xi =1| ^yj =0) of drug non-use and the presence of disease symptoms in the training set; substitute the calculated data P( ^yj =0) and P( ^xi =0, ^yj =0) into formula (4) to calculate the conditional probability P( ^xi =0|yj=1) of drug non-use and the absence of disease symptoms in the training set. ^j = 0).

S3, taking the symptom X' of the case to be prescribed as an input parameter, and calculating the posterior probability P(y j =1| ^X ) of each drug being used and the posterior probability P(y ^j =0|X) of each drug not being used according ^to the probability P(xi) of the symptom, the probability P(y ^j ) of the drug, and the conditional probability P( ^xi =t|y ^j =l) between the symptom and the drug calculated in S2, wherein X=( ^x1 , ^x2 , ..., ^xi , ..., xI) represents each symptom of the patient, and the value of ^xi (i=1, 2, ..., ^I ) is 1 or 0;

The calculation method of the posterior probability P(y ^j = l|X) of each drug being used or not used is as follows:

In formula (5),

X＝(x ¹ , x ² , …, x ⁱ , …, x ^I ) represents the symptoms of the patient, and the value of x ⁱ (i＝1, 2, …, I) is 1 or 0. l＝0, 1 represents the two situations of drug y ^j not being used and using, that is, P(y ^j ＝0|X) represents the posterior probability of drug y ^j not being used, and P(y ^j ＝1|X) represents the posterior probability of drug y ^j being used.

S4. Compare P(y ^j =1|X) and P(y ^j =0|X). If P(y ^j =1|X)≥P(y ^j =0|X), then determine that the medicine is a medicine that matches the symptoms X' of the case. All medicines that match the symptoms X' of the case are combined to obtain a matching medicine result Y=(y ¹ , y ² , ..., y ^J ).

Experimental Example 1:

The symptoms corresponding to the two prescriptions in the test prescription set in Example 1 were used to obtain drugs matching them using the method of the present invention, and then the consistency of the drug compatibility in the traditional prescription and the drug compatibility given by the method of the present invention was verified. The symptoms and drug compatibility are listed in Table 5.

Table 5 Test recipe set

The disease symptoms ^xi and drugs ^yj in the test prescription set in Table 5 are binarized to form a test set.

Taking the symptoms in the test set as input parameters, according to the method of Example 1, first, the probability of the symptom P( ^xi ), the probability of the drug P( ^yj ), and the conditional probability between the symptom and the drug P( ^xi =t| ^yj =l) are calculated, and then the posterior probability P( ^yj =1|X) of using a certain drug and the posterior probability P( ^yj =0|X) of not using a certain drug are calculated. P( ^yj =0|X) and P( ^yj =1|X) corresponding to the same drug are compared. If P( ^yj =1|X)≥P( ^yj =0|X), it is judged that the drug is used under the symptom, otherwise it is judged that the drug is not used under the symptom; finally, all the drugs judged to be used under the symptom are matched, that is, the drug result Y=( ^y1 , ^y2 , ..., ^yJ ) matching the symptom is obtained by using the method of the present invention, and the results are listed in Table 6.

Table 6 Drug results obtained by using the method of the present invention to match

Comparing the traditional drug combinations in Table 5 with the drug results obtained by using the method of the present invention in Table 6, it can be seen that the two groups of drug combinations in sequence number 2 are exactly the same. In sequence number 1, there are 9 drugs in the traditional formula of Table 5, while there are 10 drugs in Table 6 using the method of the present invention to obtain the drug results matching the traditional drug combinations, wherein Table 6 replaces rabbit heart and clove in Table 5 with Gardenia jasminoides, Toosendan fructus and Carthamus tinctorius.

For the symptoms in sequence number 2, the matching drug results obtained by the traditional formula and the present invention are different, so the following analysis is performed:

According to formula (5) in Example 1, the posterior probability P( ^xi = 1| ^yj = 1) of using a single drug under a single symptom can be obtained, which can reflect the therapeutic effect of the drug on the symptom. Substituting the data of P( ^xi = 1), P( ^yj = 1) and P( ^xi = 1| ^yj = 1) into formula (5), the posterior probability P( ^yj = 1| ^xi = 1) between a single symptom and a single drug is obtained, that is, the use and non-use of a single drug are obtained, and the results are shown in Table 7.

Table 7 Use and non-use of single drugs

The symptoms listed in Table 7 are the symptoms that appear in sequence number 2, and it can be seen that the therapeutic effects of various medicines on various symptoms. From Table 7, it can be seen that in the treatment of palpitations, in the traditional formula of drug compatibility, cloves and rabbit hearts are better than gardenia, Chuanlianzi and safflower; and in terms of chest pain, gardenia and Chuanlianzi have better therapeutic effects than cloves and rabbit hearts. Safflower has better therapeutic effects than cloves and rabbit hearts in terms of cough and white sputum. Therefore, for patients with severe chest pain, cough and white sputum, the drug compatibility in Table 5 can be used; for patients with severe palpitations, the drug compatibility in Table 4 can be used.

In summary, it can be seen that the method for matching medicines to disease symptoms of the present invention is reliable.

The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (5)

1. A method for matching a condition to a drug based on a naive bayes network, comprising the steps of:

S1, collecting a plurality of traditional formulas, and corresponding symptoms X of the formulas to the used medicines Y to form a formula sample set Z;

S2, taking part of formulas in the formula sample set Z established in the S1 as a training formula set, carrying out binarization treatment on the training formula set to obtain a training set, training a naive Bayesian network model by utilizing the training set, and calculating the probability P (x ⁱ) of symptoms in the training set, the probability P (y ^j) of medicines and the conditional probability P (x ⁱ＝t|y^j =l) between symptoms and medicines, wherein t=0, 1 represents that symptoms x ⁱ are not present and are present, and l=0, and 1 represents that the medicines y ^j are not used and are used;

S3, taking a symptom X' of a to-be-prescribed case as an input parameter, and calculating a posterior probability P (y ^j = 1|X) of each medicine and an posterior probability P (y ^j = 0|X) of each medicine which are not used according to the probability P (X ⁱ) of the symptom, the probability P (y ^j) of the medicine and the conditional probability P (X ⁱ＝t|y^j =l) between the symptom and the medicine calculated in the S2, wherein X= (X ¹,x²,…,xⁱ,…,x^I) represents each symptom of a patient, and the value of X ⁱ (i=1, 2, …, I) is 1 or 0;

and S4, comparing the sizes of P (Y ^j = 1|X) and P (Y ^j = 0|X), if P (Y ^j＝1|X)≥P(y^j = 0|X), judging that the medicine is a medicine matched with the symptom X 'of the case, and matching all medicines matched with the symptom X' of the case to obtain a matched medicine result Y= (Y ¹,y²,…,y^J).

2. The method for naive bayes network based condition to drug matching according to claim 1, wherein in S1,

Recipe sample set Z＝{(X₁,Y₁),(X₂,Y₂),…,(X_n,Y_n),…,(X_N,Y_N)},

Wherein N is the total number of formulas in the formula sample set;

the corresponding symptoms in the nth recipe are noted as In vectorsIs the corresponding ith symptom of the disease in the nth formula;

The compatibility of medicines in the nth formula is recorded as In vectorsIs the jth medicine in the nth formula.

3. A method of matching a naive bayes network based condition to a drug according to claim 2, wherein said S2 comprises the steps of: the method for binarizing the training formula set is that in a first formula, if the symptom x ⁱ exists, 1 is assigned to the symptom x ⁱ; if symptom x ⁱ is absent, assigning 0 to symptom x ⁱ; similarly, if the prescribed prescription contains a drug y ^j, then the drug y ^j is assigned 1; if the prescribed prescription does not contain the drug y ^j, then the drug y ^j is assigned a value of 0.

4. A method of matching a naive bayes network based condition to a drug according to claim 2, wherein said S2 comprises the steps of:

(1) The probability of symptoms of the disorder in the training set P (x ⁱ) and the probability of the drug P (y ^j) are calculated, in particular,

In the formula (1), the amino acid sequence of the formula (1),The cumulative frequency of symptoms x ⁱ =t in the training set, t=0, 1 indicates both the absence and presence of symptoms x ⁱ, i.e., t=0 indicates the absence of symptoms x ⁱ in the nth formulation, t=1 indicates the presence of symptoms x ⁱ in the nth formulation, and N is the total number of formulations in the formulation sample set;

in the formula (2), the amino acid sequence of the formula (2), The cumulative frequency of drugs y ^j = l in the training set; l=0, 1 indicates both the non-use and use of drug y ^j, i.e., l=0 indicates that drug y ^j is not used in the nth formulation, l=1 indicates that drug y ^j is used in the nth formulation, and N is the total number of formulations in the formulation sample set;

(2) The probability P (x ⁱ＝t,y^j =l) is calculated for the symptom x ⁱ =t and the drug y ^j =l in the training set,

In the formula (3), the amino acid sequence of the compound,Formulation accumulation frequency, representing symptoms x ⁱ =t and drug y ^j =l in the training set;

(3) The conditional probability P between symptoms of the disorder and the drug is calculated (x ⁱ＝t|y^j = l),

The conditional probability can be obtained by substituting the expression (2) and the expression (3) into the expression (4).

5. The method of naive bayes network based condition to drug matching of claim 4 wherein in S3: the calculation method of the posterior probability P (y ^j =l|x) of each drug used and unused is as follows:

in formula (5), x= (X ¹,x²,…,xⁱ,…,x^I) represents each symptom of the disease in the patient, and X ⁱ (i=1, 2, …, I) has a value of 1 or 0; l=0, 1 indicates both the unused and used cases of drug y ^j, i.e., P (y ^j = 0|X) indicates the posterior probability of drug y ^j being unused, and P (y ^j = 1|X) indicates the posterior probability of drug y ^j being used.