KR102359652B1 - System and Method for Classifying Disease Using Class Association Rules - Google Patents

Abstract

A disease classification system and method using class association rules establishes a classification model of Tibetan medical syndromes by selecting high-priority sets to apply a strong association rule between symptoms and syndromes to diagnose and treat Tibetan medical syndromes, and general It has the effect of improving the accuracy of clinical diagnosis.

Description

System and Method for Classifying Disease Using Class Association Rules

The present invention relates to a disease classification system, and in particular, class association for diagnosing and treating Tibetan medical syndromes by selecting a high-priority set to apply a strong association rule between symptoms and syndromes to establish a classification model for Tibetan medical syndromes It relates to a disease classification system and method using rules.

Tibetan medicine has traditional domestic medicine with a long history. An important part of Chinese medicine has a unique theoretical system and clinical efficacy.

Tibetan Medicine provides medicinal advice for farmers in northwestern China, but the clinical research base is weak due to the complexity of the cultural background and local environment, and clinical characteristics are not standardized.

Because Tibet is located in the highlands, the incidence of altitude sickness is high, and the risk of stomach cancer is increasing due to the repetition of altitude sickness.

Since Tibet does not have clinical principles based on medical theory, there is a problem in that there is a large amount of subjective factors by doctors in the process of diagnosing and treating diseases.

Korean Patent No. 10-1950112

In order to solve this problem, the present invention establishes a classification model of Tibetan medical syndrome by selecting a high-priority set in order to apply a strong association rule between symptoms and syndromes to diagnose and treat Tibetan medical syndromes. An object of the present invention is to provide a disease classification system and method using rules.

A disease classification system using a class association rule according to a feature of the present invention for achieving the above object,

an input unit for receiving a Tibetan medical record document from the medical information database unit;

By removing stopwords irrelevant to the generation of the association rule from the medical record document, a plurality of medical words that are the target of generating the association rule from among a plurality of words included in the medical record document are extracted, and the extracted medical word is used for data mining. a preprocessor that generates a suitable data set;

In the data set, a symptom of a disease is set as a condition attribute, a medical syndrome is set as a decision attribute, and a class association rule between the symptom of the disease and the medical syndrome is generated using an association rule mining algorithm. wealth; and

A medical disease that determines the priority of each class association rule by using the method of sorting class association rules by reliability and support or by sorting class association rules by Lift (bottom-up method), and sets the rule set with the highest priority as a classification model It is characterized in that it includes a classification unit.

A disease classification method using a class association rule according to a feature of the present invention,

receiving a Tibetan medical record document from a medical information database unit;

By removing stopwords irrelevant to the generation of the association rule from the medical record document, a plurality of medical words that are the target of generating the association rule from among a plurality of words included in the medical record document are extracted, and the extracted medical word is used for data mining. generating a suitable data set;

setting a symptom of a disease as a condition attribute in the data set, setting a medical syndrome as a decision attribute, and generating a class association rule between the symptom of the disease and the medical syndrome using an association rule mining algorithm;

The method of sorting class association rules by reliability and support or the step of determining the priority of class association rules by Lift (bottom-up method) using sorting, and setting the rule set with the highest priority as a classification model characterized by including.

According to the above configuration, the present invention establishes a classification model of Tibetan medical syndrome by selecting a high-priority set in order to apply a strong association rule between symptoms and syndromes to diagnose and treat Tibetan medical syndromes and general clinical diagnosis. It has the effect of improving accuracy.

1 is a diagram schematically illustrating the configuration of a disease classification system using a class association rule according to an embodiment of the present invention.
2 is a diagram illustrating the concept of a connection relationship between symptoms and syndromes according to an embodiment of the present invention.
3 is a diagram illustrating a disease classification method using a class association rule according to an embodiment of the present invention.
Figure 4 is a view comparing the classification model based on the Conf1Sup2 and Lift alignment method according to an embodiment of the present invention.

Throughout the specification, when a part "includes" a certain element, it means that other elements may be further included, rather than excluding other elements, unless otherwise stated.

1 is a diagram schematically illustrating the configuration of a disease classification system using class association rules according to an embodiment of the present invention, and FIG. 2 is a diagram illustrating the concept of a connection relationship between symptoms and syndromes according to an embodiment of the present invention, 3 is a diagram illustrating a disease classification method using a class association rule according to an embodiment of the present invention.

The disease classification system 100 using a class association rule according to an embodiment of the present invention includes an input unit 110 , a preprocessor 120 , a data mining unit 130 , and a medical disease classification unit 140 .

The input unit 110 receives Tibetan medical record documents (clinical treatment documents, prescription record documents, medical observation documents, and medical test data) from the medical information database unit 101 .

The preprocessor 120 removes stopwords irrelevant to the generation of the association rule from the medical record document, extracts a plurality of medical words that are the target of generating the association rule from among a plurality of words included in the medical record document, and extracts the extracted medical word. A plurality of data sets suitable for data mining are generated by using (S100, S101).

The data mining unit 130 simply uses the frequency information of keywords appearing in the document to separate the documents based on the medical word extracted using the Apriori algorithm, which is an association rule analysis technique, unlike a statistical method of designating a classification category. By extracting a set of related keywords, a classification rule can be created with keywords having semantically representativeness for each category.

The data mining unit 130 sets the symptom of altitude sickness as a condition attribute in the data set, and sets the Tibetan medical syndrome as a decision attribute. This results in Symptoms => Syndrome.

The data mining unit 130 generates a class association rule between the symptoms of altitude sickness and Tibetan medical syndrome by using an Apriori association rule mining algorithm (S102). In this case, the thresholds of support and confidence of classification are predetermined.

Association rules refer to useful patterns expressed in terms of conditions and results (Condition-Conclusion) in the form of “X => Y(support, confidence)” extracted from transactions such as medical words.

Here, X is an item constituting a transaction, and "X => Y" means that when an item X corresponding to a condition occurs, an item Y corresponding to the result also occurs.

The purpose of the association rule is to data mining rules that satisfy preset support thresholds and reliability thresholds.

Support refers to the ratio of transactions involving item X and item Y simultaneously in all transactions, and confidence refers to the ratio of transactions including item Y in transactions including item X.

A class association rule is a part of an association rule, and the subsequent items of the rules are fixed as a classification attribute, which can be used to distinguish categories of data objects.

The class association rule reflects the prediction of the rule as well as the internal relevance, and data mining the rule that satisfies the minimum support and minimum reliability in order to predict the data object.

The data mining unit 130 defines a class association rule (CAR) in the data set as a transaction data set (D) as shown in Equation 1 below.

The form of Rule r is X => Y, I is the set of all items (symptoms) in the data set (D), and C is the set of all class (syndrome) levels in the data set (D). For example, elements of I include cough, sputum, and runny nose, and elements of C include pneumonia and colds. D is the full set of all cases.

The data mining unit 130 calculates the degree of support of the class association rule by using Equation 2 below.

The degree of support (Sup(r)) of the class association rule represents the ratio of the number of items matching the rule of the prerequisite (X) and the classification attribute (Y) and the total number of all items in the data set (D).

The degree of support of the rule r is defined as in Equation 2, and is 1 when the symptoms and syndromes of the rule intersect (connect). Count all of these and divide by the whole set D.

For example, if cough -> cold, runny nose -> cold, phlegm -> undecided, 2/3 is Sup.

The data mining unit 130 calculates the reliability of the class association rule using Equation 3 below.

The confidence level (Conf(r)) of the class association rule is the ratio of the number of items that satisfy the rule of the antecedent condition (X) and the classification attribute (Y) and the total number of items of the antecedent condition (X) in the data set (D). indicates.

Equation 3 is replaced with count(r.x) instead of D in Equation 2 described above.

For example, if cough -> cold, runny nose -> cold, phlegm -> undecided, 2/3 is Conf.

The class association rule is a rule that satisfies the conditions of Equation 1, Equation 2, and Equation 3, and minSup and minConf are the minimum support threshold and the minimum reliability boundary in which support and reliability are preset (Equation 4).

The data mining unit 130 uses a breadth first iterative search method to calculate the support for each class, compares it with the minimum support boundary, and selects the frequency 1- of the class connection rule. Search for itemset (the set of items that appear most frequently).

Next, the data mining unit 130 determines the candidate 2-itemset and the frequent 2-itemset of the class association rule by using the result of the class association rule of 1-itemset (the item set with the highest frequency). . This process continues until the k-itemset of the class association rule can no longer occur. As a result, based on the diagnosis and treatment of Tibetan medical care, a rule is determined that the candidate set of class association rules is “any symptom, any syndrome”.

In other words, the data mining unit 130 searches for a type of event that occurs frequently in a unit transaction of a large data set, searches for a frequent item set that satisfies a minimum support threshold or more for a transaction, and searches for a frequent item Creates a candidate set with , repeats until no new frequent item set is created, and creates association rules between items that satisfy the minimum reliability threshold or higher among items included in a large number of searched item sets. .

Since the rules of the candidate CAR set are redundant and have priority, it is necessary to remove the redundant rules to improve the accuracy of the classification model, and the CAR with higher priority is selected as the classification model.

The medical disease classification unit 140 removes the class association rule based on the function of correctly classifying the sample. For each rule r of the candidate CAR set, whether there is an item including the rule r in the training set is detected.

The medical disease classifier 140 indicates that if rule r correctly classifies one or more samples, then rule r is marked as a potentially useful class association rule.

When all of the rules of the candidate CAR set or the items of the training set are traversed, the medical disease classification unit 140 blocks a rule that cannot correctly classify the items.

The medical disease classification unit 140 prioritizes CAR reliability as a first keyword, ConflSup2 and Lift method as a second keyword, and selects a high-priority rule set to select a classification model is set (S103). The description of the above two sorting methods is as follows.

(1) The method of arranging class association rules by reliability and support is called ConflSup2.

When the reliability of r1 (rule 1) is greater than the reliability of r2 (rule 2),

When the reliability of r1 (rule 1) is equal to that of r2 (rule 2), and the support of r1 (rule 1) is greater than that of r2 (rule 2),

When the reliability of r1 (rule 1) is equal to that of r2 (rule 2), the support of r1 (rule 1) is equal to that of r2 (rule 2), and the category of r1 is smaller than that of r2 (r2 is r1) including)

Here, Sup represents the support of the rule and Conf represents the reliability of the rule. If r1 and r2 satisfy any one of the above rules, priority is given to r1, expressed as r1 > r2.

(2) Sort class association rules by Lift (upward method).

It represents the lift of r1 (rule 1) > the lift of r2 (rule 2). If r1 and r2 satisfy the above condition, the priority of r1 is higher than that of r2, and it is expressed as r1 > r2.

The medical disease classification unit 140 performs a method of aligning class association rules by reliability and support (ConflSup2) or a method of aligning class association rules by Lift (bottom-up method), and an alignment method with high accuracy of classification results is selected as the classification model.

As shown in Table 1 and Figure 4 below, when the class association rules are sorted based on Conf1Sup2, the accuracy increases as the number of rules increases when the number of rules is in the range of 40 to 400, and the selected If the number of rules is 80, the maximum is 80.78%.

However, if the number of rules is between 400 and 800, the accuracy does not change even if the number of rules increases.

In the case of sorting class association rules based on lift, when the number of rules is in the range of 40 to 400, the accuracy increases as the number of rules increases, and when the number of rules is 400, the accuracy is up to 82.72%. However, when the number of rules is between 400 and 800, the accuracy decreases as the number of rules increases.

Regardless of the method used to sort the class association rules, the accuracy of the classification model constructed based on Lift is higher than that based on Conf1Sup2, except for the number of rules with 200.

When comparing the two sorting methods, the medical disease classification unit 140 sorts the class association rules based on the lift sorting method, and when the number of rules is 400, the accuracy is the highest at 82.72%, so ConflSup2 or Lift depending on the number of rules You can set a classification model that selects a high-accuracy sorting method.

That is, as shown in FIG. 4 , when the number of rules is 200, the medical disease classification unit 140 sorts the priorities of class association rules using the ConflSup2 sorting method, In this case, the priority of class association rules is sorted by Lift sorting method.

In addition, the medical disease classification unit 140 sorts the priority of the class association rules based on the ConflSup2 or Lift sorting method, and selects 40, 80, 200, 400, 800, and 1000 different sized rules in a high order The classification model can be set according to the order of priority from the lowest to the lowest.

The medical disease classification unit 140 uses a classification model based on class association rules to execute a process of verifying that the symptoms for the disease are correctly classified according to the rule priority of the Tibet medical syndrome classification model. (S104).

In the above, the embodiment of the present invention is not implemented only through the apparatus and/or method, and may be implemented through a program for realizing a function corresponding to the configuration of the embodiment of the present invention, a recording medium in which the program is recorded, etc. And, such an implementation can be easily implemented by an expert in the technical field to which the present invention belongs from the description of the above-described embodiment.

Although the embodiments of the present invention have been described in detail above, the scope of the present invention is not limited thereto. is within the scope of the right.

100: disease classification system
101: medical information database unit
110: input unit
120: preprocessor
130: data mining unit
140: medical disease classification unit

Claims (7)

an input unit for receiving a Tibetan medical record document from the medical information database unit;
By removing stopwords irrelevant to the generation of the association rule from the medical record document, a plurality of medical words that are the target of generating the association rule from among a plurality of words included in the medical record document are extracted, and the extracted medical word is used for data mining. a preprocessor that generates a suitable data set;
In the data set, a symptom of a disease is set as a condition attribute, a medical syndrome is set as a decision attribute, and a class association rule between the symptom of the disease and the medical syndrome is generated using an association rule mining algorithm. wealth; and
A medical disease that determines the priority of each class association rule by using the method of sorting class association rules by reliability and support or by sorting class association rules by Lift (bottom-up method), and sets the rule set with the highest priority as a classification model contains a classification,
The medical disease classification unit determines the priority of the class association rule by Lift (a bottom-up method) by sorting by Equation 2 below, and when r1 and r2 satisfy the condition of Equation 2 below, r1 A disease classification system using a class association rule, characterized in that the priority is higher than r2, and r1 > r2.
[Equation 2]

The lift of r1 (rule 1) > denotes the lift of r2 (rule 2). According to claim 1,
The medical disease classification unit determines the priority by arranging the class association rule based on reliability and support according to Equation 1 below, and if r1 and r2 satisfy any one of the rules of Equation 1 below, priority is given to r1 A disease classification system using a class association rule, characterized in that given and expressed as r1 > r2.
[Equation 1]

When the reliability of r1 (rule 1) is greater than the reliability of r2 (rule 2),
When the reliability of r1 (rule 1) is equal to that of r2 (rule 2), and the support of r1 (rule 1) is greater than that of r2 (rule 2),
When the reliability of r1 (rule 1) is equal to that of r2 (rule 2), the support of r1 (rule 1) is equal to that of r2 (rule 2), and the category of r1 is smaller than that of r2 (r2 is r1) including)
Sup indicates the degree of support of the rule and Conf indicates the reliability of the rule. delete According to claim 1,
The data mining unit defines the class association rule in the data set as shown in Equation 3 below, and the degree of support of rule r is 1 when the symptoms and syndromes of the rule intersect (connection) as shown in Equation 4 below, and the intersection A disease classification system using a class association rule, characterized in that counting all the things that are, dividing the whole set D, and defining the reliability of the rule r as in Equation 5 below.
[Equation 3]

X is the prerequisite of the rule, Y is the classification attribute, I is the set of all items (symptoms) in the data set (D), C is the set of all class (syndrome) levels in the data set (D), D is The full set of all cases.
[Equation 4]

[Equation 5]

receiving a Tibetan medical record document from a medical information database unit;
By removing stopwords irrelevant to the generation of the association rule from the medical record document, a plurality of medical words that are the target of generating the association rule from among a plurality of words included in the medical record document are extracted, and the extracted medical word is used for data mining. generating a suitable data set;
setting a symptom of a disease as a condition attribute in the data set, setting a medical syndrome as a decision attribute, and generating a class association rule between the symptom of the disease and the medical syndrome using an association rule mining algorithm;
The method of sorting class association rules by reliability and support or the step of determining the priority of class association rules by Lift (bottom-up method) using sorting, and setting the rule set with the highest priority as a classification model includes,
The step of setting the classification model is,
The priority of the class association rule by Lift (a bottom-up method) is determined using sorting by Equation 2 below, and when r1 and r2 satisfy the condition of Equation 2 below, the priority of r1 is r2 A disease classification method using a class association rule, characterized in that higher, r1 > r2.
[Equation 2]

The lift of r1 (rule 1) > denotes the lift of r2 (rule 2). 6. The method of claim 5,
The step of setting the classification model is,
Priority is determined by arranging the class association rule based on the reliability and support by Equation 1 below, and if r1 and r2 satisfy any one of the rules of Equation 1 below, priority is given to r1, r1 > A disease classification method using a class association rule, characterized in that it is denoted by r2.
[Equation 1]

When the reliability of r1 (rule 1) is greater than the reliability of r2 (rule 2),
When the reliability of r1 (rule 1) is equal to that of r2 (rule 2), and the support of r1 (rule 1) is greater than that of r2 (rule 2),
When the reliability of r1 (rule 1) is equal to that of r2 (rule 2), the support of r1 (rule 1) is equal to that of r2 (rule 2), and the category of r1 is smaller than that of r2 (r2 is r1) including)
Sup indicates the degree of support of the rule and Conf indicates the reliability of the rule. delete

Images