JP6053167B2 - Numerical data analysis apparatus and program - Google Patents

Description

  The present invention is capable of appropriately discretizing somewhat special statistical numerical data such as medical data according to the configuration of the distribution, and continuous numerical data depending on the discretized result. The present invention also relates to a numerical data analysis apparatus and program that can apply the potential Dirichlet distribution method to the above.

  As represented by Patent Document 1 and Patent Document 2, the health management system and the like are expanding greatly. In such a health management system, health advice is often given to users. However, as shown in Patent Document 3, advice is given after classifying users based on actual health data. Are more likely to lead to behavioral changes.

JP 2013-085626 JP 2010-264088 JP 2010-170534 A

  There are scenes where detailed discretization is required for the health data or medical data. For example, there is a case where it is desired to analyze medical data for application to the health management system using a classification machine that can handle only discrete data. In some cases, the discretization may classify patients according to specific numerical values.

  An example of a classification machine that can handle only discrete data is LDA (latent dirichlet allocation). LDA is a document clustering method that performs topic classification of multiple documents with high accuracy by estimating latent variables. Document clustering is known to have better performance than clustering methods such as the k-means method.

  However, since LDA is originally a method used in document topic classification, it can only handle discretized data. Therefore, in order to use LDA using continuous value data such as biological data, discretization is required as described above. By performing such clustering, it is possible to extract a dangerous patient group.

  At this time, in the general discretization technique of continuous data, the classification is often performed by classification by frequency or by calculating the deviation of the standard deviation from the average value.

  However, in general, the frequency distribution of medical data varies greatly depending on the type of data to be handled, and it is difficult to handle it centrally. There are few cases where the inspection value itself increases linearly simply. Therefore, the calculation of the deviation from the average is often meaningless.

  In addition, a method of separating data by a method such as a tertile uniformly is also common. In this case, since the data is separated only by the frequency information, when there is a group with low frequency and very large data, the group is absorbed by the group with high frequency.

  As described above, when it is considered to discretize medical data for application to a health management system or the like, the general continuous data discretization method that is not limited to medical data can perform appropriate discretization. There was no problem.

  In view of the above problems, an object of the present invention is to provide a numerical data analysis apparatus and program that can appropriately discretize and classify even continuous data such as medical data.

  In addition, the present invention performs appropriate classification even for continuous data such as medical data, and, according to the classified result, each sample in the original continuous data is classified with high accuracy based on discrete data. It is an object of the present invention to provide a numerical data analysis apparatus and program that can be analyzed by the above.

  In order to achieve the above object, the present invention is a numerical data analysis apparatus that handles numerical data of a predetermined item obtained from a sample group, a frequency classification unit that classifies numerical data based on the frequency deviation, and a numerical value A numerical value classification unit that classifies data based on the numerical value bias, and for the input numerical data, the frequency classification unit includes normal data and non-normal data at the upper end and / or lower end; The numerical classification unit further classifies the classified non-normal data into individual non-normal data, and the frequency classifying unit further classifies the classified normal data with individual normal data, The upper and / or lower end non-ordinary data is classified, and the classification by the numerical classification unit and the frequency classification unit satisfies the predetermined convergence condition for each of the normal and non-normal data classified individually. Characterized in that it is continuous.

  Further, according to the present invention, the numerical data analysis device performs classification by the frequency classification unit and the numerical classification unit on the numerical data of one or more predetermined items, and performs the first process, the second process, and the third process. An estimation unit; in the first process, a numerical label is assigned to each of the classified normal and non-normal data in the numerical data of the predetermined item, and in the second process, the numerical data of the predetermined item is For each sample to be configured, a document based on the given numeric label is generated for the numeric data of each predetermined item in the sample, and in the third process, the generated document for all samples By applying the latent Dirichlet allocation method to the set, the topic set in the document set and the topic ratio in each sample are estimated.

  Furthermore, the present invention is a numerical data analysis program that causes a computer to function as the numerical data analysis apparatus.

  According to the present invention, even continuous data such as medical data is separated based on the numerical value and the frequency deviation, and therefore appropriate separation reflecting a tendency peculiar to the data becomes possible.

  Further, according to the present invention, by using the above separation result, it is possible to perform analysis by the latent Dirichlet distribution method, which is a high-precision classifier based on discrete data, with respect to initial continuous data. It becomes.

It is a functional block diagram of the numerical data analyzer which concerns on one Embodiment. It is a figure for demonstrating the outline | summary of the process of a 1st function part. It is an example of medical examination data. In the example of the same medical examination data as in FIG. 3, the distinction of the portion showing the non-linear increase is clearly shown on the graph, which is an example of the input data in the description of the specific example. It is a figure which shows the frequency distribution of the test value of triglyceride, and its occurrence frequency. It is a figure which shows the non-normal data obtained by isolation | separation of the input data of FIG. It is a figure which shows the normal data obtained by isolation | separation of the input data of FIG. It is a figure which shows the process in which the non-normal data of FIG. 6 are isolate | separated sequentially by the numerical classification part. It is a figure which shows the non-normal data isolate | separated in the process of FIG. It is a figure which shows the non-normal data in the middle of being isolate | separated in the process of FIG. It is a figure which shows the non-normal data isolate | separated in the process of FIG. It is a figure which shows the non-normal data in the middle of being isolate | separated in the process of FIG. It is a figure which shows the non-normal data isolate | separated in the process of FIG. It is a figure which shows the non-normal data isolate | separated in the process of FIG. It is a figure which shows the process in which the normal data of FIG. 7 are isolate | separated sequentially by the frequency classification part and the numerical value classification part. It is a figure which shows the normal data isolate | separated in the process of FIG. It is a figure which shows the non-normal data in the middle of being isolate | separated in the process of FIG. It is a figure which shows the non-normal data in the middle of being isolate | separated in the process of FIG. It is a figure which shows the non-normal data in the middle of being isolate | separated in the process of FIG. It is a figure which shows the non-normal data isolate | separated in the process of FIG. It is a figure which shows the non-normal data isolate | separated in the process of FIG. It is a figure which shows the non-normal data isolate | separated in the process of FIG. It is a figure which shows the non-normal data isolate | separated in the process of FIG. It is a figure which shows the non-normal data isolate | separated in the process of FIG. It is a figure which shows the flow of an example of an additional process in the case of the document production | generation in the 2nd process by an estimation part.

  FIG. 1 is a functional block diagram of a numerical data analysis apparatus according to an embodiment. The numerical data analysis device 1 includes a first function unit 10 and an estimation unit 5 as a second function unit (no reference sign is given). The first function unit 10 includes a frequency classification unit 2, a numerical classification unit 3, and a classification result update holding unit 4, receives numerical data as an input to the numerical data analysis device 1, and outputs the classification result.

  When outputting the classification result, each unit of the first function unit 10 functions as follows. The frequency classification unit 2 classifies the numerical data based on the frequency deviation. The numerical value classification unit 3 classifies the numerical data based on the deviation of the numerical values. The classification is performed by separating and grouping each numerical data according to the range. The classification result update holding unit 4 holds and updates the intermediate results when the classification is performed sequentially as follows.

  Here, the frequency classification unit 2 and the numerical classification unit 3 perform classification processing in cooperation with each other on the input numerical data, and the classification proceeds sequentially. That is, separation by the range of numerical data proceeds sequentially. At this time, additional separation processing is continued while maintaining the result of the separation. Holds the result that has been separated halfway, provides the intermediate result when the frequency classification unit 2 or numerical classification unit 3 performs further processing, and updates the classification result with the role of updating the separation result with the further processed result The holding part 4 bears.

  In the following description, the classification result update holding unit 4 is not explicitly referred to as the operating subject, but the classification result update is performed when the frequency classification unit 2 and the numerical classification unit 3 perform each process continuously. The holding unit 4 has an auxiliary role of holding and providing the classification result up to the middle.

  On the other hand, the estimation unit 5 receives the classification result output from the first function unit 10, and applies a LDA (potential Dirichlet distribution method) to obtain a topic set and the like. The estimation unit 5 will be described later, and will be described below from the first function unit 10.

  FIG. 2 is a diagram for explaining an outline of sequential separation processing by the frequency classification unit 2 and the numerical value classification unit 3 in the first function unit 100. Here, the processing will be described from a formal viewpoint, and the substantial meaning will be described later.

  In FIG. 2, numerical data on a predetermined item obtained from a sample group is shown as input numerical data D in (1). Here, as a specific example, a medical examination value of a patient group (value of a predetermined item in a medical examination) ) Is used. The input numerical data D is converted into a format as shown in (1) of FIG. 2 as preprocessing. That is, it is converted into the form of the cumulative frequency function based on the medical examination value.

  First, as shown in (1) of FIG. 2, the frequency classification unit 2 uses the frequency to calculate the numerical data D as the cumulative frequency function based on the frequency, that is, on the upper end and / or lower end side, that is, the frequency. The function graph is separated into non-normal data which is on the right end and / or left end side of the function graph and can be regarded as extremely low in frequency (and not normal as a medical examination value), and normal data in the middle.

  In (1) of FIG. 2, an example is shown in which the data is separated into three data, that is, non-normal data S1 on the lower end side, normal data N1 in the middle, and non-normal data S2 on the upper end side. In some cases, the data is separated into two data, that is, non-normal data only at one of the upper end and the lower end and normal data of the remaining portion.

  Next, as shown in (2) and (3) of FIG. 2, the numerical value classification unit 3 further separates the non-normal data obtained by the frequency classification unit 2 into two based on the numerical value. For example, as shown in (2), the non-ordinary data S1 is individually plotted against the cumulative frequency, and the sorted medical examination values are plotted (the inverse function of the cumulative frequency function, and the content is the cumulative frequency function Is separated into two pieces of non-normal data S1-S1 and S1-S2. Similarly, as shown in (3), the non-normal data S2 is separated into two, non-normal data S2-S1 and S2-S2.

  As used above, in the description of the present invention, the data to be separated is described according to the following rules. That is, data obtained by further separating certain data A is described by adding a hyphen and a sign to the data A from which the data A is separated, such as A-B, A-C, and A-D. Data obtained by further separating A-B is described as A-B-X, A-B-Y, A-B-Z, and so on. According to the notation, as will be described later with reference to FIGS. 8 and 15, the separated processes are represented in a tree format.

  Similarly, the numerical value classification unit 3 recursively separates each separated non-normal data. In the case of FIG. 2, the non-normal data S1-S1 and S1-S2 in (2) are further targeted for separation in the same manner as the non-normal data S1 is separated. Similarly, the non-normal data S2-S1 and S2-S2 in (3) are also further separated.

  On the other hand, the frequency classification unit 2 similarly performs recursive separation processing on normal data. That is, the normal data N1 obtained from the initial numerical data D shown in (4) is subject to separation into normal data and non-normal data, like the data D.

  Note that, as a result of the separation of the normal data N1, the non-normal data is subject to separation by the numerical value classification unit 3 as shown in (1), (2), and (3). Thereafter, the subsequent normal data obtained recursively is similarly separated by the numerical value classification unit 3 at the time when the non-normal data is obtained.

  As described above, the frequency classifying unit 2 recursively separates normal data and non-normal data into the normal data (except for the first time for the entire input data D), and the numerical classifier 3 recursively separates non-normal data into non-normal data. Here, in the separated normal data and non-normal data, when the condition described later is satisfied, no further separation is performed, and the data constituting one element in the final classification result for the initial input data D is Will be obtained.

  In the above, the separation process has been described with reference to FIG. A specific example of the sequential separation will be described later. On the other hand, the contents of the individual separation processing at the time of the sequential separation are as follows.

  The frequency classifying unit 2 performs fitting of a sigmoid function (logistic function) to the cumulative frequency function given by the data, and by determining a frequency non-normal value by threshold determination on each of the upper and lower sides, The data can be separated.

  The sigmoid function is as shown in the following formula (1), and c, b, and K are parameters. When fitting to the cumulative frequency function, x is the health check value and y is the cumulative frequency.

  The above formula (1) is transformed into the following formula (2), and the log of both sides is taken to obtain the following formula (3).

  In the above formula (3), when the following formula (4) is set, formula (5) is obtained.

  Here, for the value of K, the maximum value of the target cumulative frequency function, that is, the total number of samples in the target data is used, and P and Q are obtained by the least square method using the above equation (5). Thus, b and c can be obtained from Equation (4).

  As described above, in Formula (1) [y = y (x)] obtained by function fitting, data separation is performed by determining the range of the medical examination value x by threshold determination using K that determines the range of the cumulative frequency y. Just do it.

  For example, the lower end side {x | y (x) ≦ 0.01K} and the upper end side {x | y (x) ≧ 0.99K} are defined as non-normal data on the lower end side and non-normal data on the upper end side, respectively, and the remaining The portion {x | 0.01K <y (x) <0.99K} may be determined as normal data. Here, the threshold value on the lower end side for defining the range of the medical examination value x is set to 0.01K and the threshold value on the upper end side is set to 0.99K. However, other predetermined ratio values with respect to K may be used as the threshold value.

  On the other hand, the numerical classification unit 3 performs regression line or support vector regression, ridge regression, lasso on non-normal data given by cumulative frequency on the horizontal axis and sorted medical examination values on the vertical axis. Draw a straight line by robust regression such as (Lasso regression), and based on whether or not it is determined to be on the straight line by threshold judgment for deviation from the straight line, the target non-normal data is divided into two To separate. In other words, whether or not the separation criterion is a linear increase portion or whether it is not an outlier or an outlier.

  Note that when applying a regression line, only a predetermined number of points continuous from the end of the non-normal data (which may be either the left end / right end or the lower end / upper end) may be used. The predetermined number may be determined as a predetermined ratio with respect to the total number of samples of the non-normal data of interest.

  Further, in the frequency classifying unit 2 and the numerical value classifying unit 3, the condition determination (separation convergence determination) in which no further separation is performed on the separated individual data may be performed as follows.

  As a first determination method, if the correlation coefficient of data configured as a cumulative frequency function or its inverse function is calculated and the value can be determined to be close to 1 by the threshold determination, a convergence determination may be made . In this determination, it is checked whether the data increases linearly.

  As a second determination method, the convergence determination may be made when the total number of data falls below a threshold value. The threshold value may be provided in a form proportional to the total number of numerical data that is initially input, so that the convergence determination may be made when the threshold value is equal to or less than a predetermined ratio with respect to the entire data.

  In addition, what is necessary is just to make a convergence determination actually, when at least any one of the conditions in said 1st and 2nd determination method is satisfy | filled.

  Next, the significance of the first function unit 10 classifying numerical data as described above will be described.

  Fig. 3 shows an example of medical examination data. From the medical examination data, a person with a value measured as a triglyceride value is labeled in ascending order, and the label is plotted on the X axis and the value is plotted on the Y axis. FIG. That is, it is a diagram plotted in the form of an inverse function of the cumulative frequency. As can be seen from FIG. 3, as the numerical value increases, a simple correlation does not hold. The same applies when the numerical value is too low. (However, in FIG. 3, it is not apparent that the simple difference is not established in terms of the display range for the low numerical value side, but this fact can be read by enlarging the range.)

There are two reasons for the behavior of data as shown in FIG.
(1) The higher the numerical value, the more the biological function is destroyed, and the numerical value increases abnormally due to an unusual rule. (The same applies to the case where it falls according to the type of data.)
(2) The frequency is low for people with extremely high and low figures.

  FIG. 4 clearly shows a portion where a non-linear increase is observed in the same data and the same graph as FIG. Looking at this data, it is difficult to say that it is increasing linearly, and although it increases linearly up to neutral fat value 420, it is exponentially above 420 as surrounded by frame F. It can be seen that there is a rapid increase.

  The behavior of the data can be interpreted as follows. That is, up to 420 is an increase in the normal rule in the living body, and after 420 surrounded by a frame F, a rule different from this rule, for example, a function to suppress neutral fat is destroyed. This is thought to be due to an increase in It is considered that this behavior is largely due to the above reason (1).

  Therefore, if the reasons (1) and (2) are adopted as classification criteria and numerical data is classified, classification that reflects a qualitative difference is possible, unlike simply dividing continuous data. The classification in the first functional unit 10 realizes the classification from this viewpoint, with the frequency classification unit 2 mainly responsible for the classification (2) and the numerical classification unit 3 mainly responsible for the classification (1). Yes. In addition, since a difference such as (1) and (2) may exist in the group once separated, recursive separation is performed. The frequency classifying unit 2 performs the separation by a method according to the reason (2). On the other hand, the significance of separation of the numerical classification unit 3 is as follows.

  FIG. 5 shows the frequency distribution of each triglyceride test value and the frequency of occurrence thereof (the person who took the test value). With reference to FIG. 5, the significance of separation of the numerical value classification unit 3 will be described.

  The graph of FIG. 5 is very similar to the income distribution in capitalist countries, and it is known that the graph often works well when approximated by a lognormal distribution on the left side and a Pareto distribution on the right side. The reason for the two distributions is that each follows a different rule. (In the case of Japanese income, the distribution is said to be different with an annual income of 20 million yen as the boundary.)

  As shown in Fig. 5, people with extremely high and extremely low test values have very low frequency of occurrence, and the robust regression (or linear regression) with the numerical classification unit 3 method has little effect. Absent. Therefore, it is possible to separate the extreme numbers of human parts.

  Next, a specific example of numerical data classification by the first function unit 10 will be described.

  In this specific example, the data in FIG. 4 (the same data as in FIG. 3 as described above) is initially input numerical data. First, the result of classification by the frequency classification unit 2 is the non-normal data in the frame F shown in FIG. 4 and the other normal data. This is an example in which the non-normal data is obtained only on the upper end side and not on the lower end side. The respective data separated into two pieces with the value 420 as a boundary are shown as non-normal data S1 and normal data N1, respectively, as shown in FIGS.

  FIG. 8 is a diagram showing a process in which the non-normal data S1 in FIG. 6 is sequentially separated by the numerical classification unit 3, and the procedure (1) shown on the left side is the start, and the separation is performed by taking a tree structure. It is done sequentially. Moreover, FIGS. 9-14 is a figure which shows each data at the time of the said sequential isolation | separation. Hereafter, each procedure (1)-(7) of FIG. 8 is demonstrated.

  In the procedure (1), the non-normal data S1 is separated into non-normal data S1-S1 and S1-S2 with the value 660 as a boundary. In the procedure (2), it is determined that it is unnecessary to separate the non-normal data S1 to S1 shown in FIG. 9 (separation convergence determination). In procedure (3), the non-normal data S1-S2 shown in FIG. 10 is further separated into non-normal data S1-S2-S1 and S1-S2-S2 with the value 900 as a boundary.

  In procedure (4), separation / convergence is determined for the non-normal data S1-S2-S1 shown in FIG. In step (5), the non-normal data S1-S2-S2 shown in FIG. 12 is further separated into non-normal data S1-S2-S2-S1 and S1-S2-S2-S2 with the value 1400 as a boundary. . In step (6), the non-normal data S1-S2-S2-S1 shown in FIG. In procedure (7), the non-normal data S1-S2-S2-S2 shown in FIG.

  As described above, according to the procedure of FIG. 8, the non-normal data S1 is shown by the numerical classification unit 3 as the non-normal data S1-S1, S1-S2-S1, S1-S2-S2 shown on the end side of the tree structure. -S1 and S1-S2-S2-S2 are separated into 4 non-normal data.

  On the other hand, FIG. 15 is a diagram showing a process in which the normal data N1 of FIG. 7 is sequentially separated by the frequency classifying unit 2 and the numerical value classifying unit 3, and the procedure (1) shown on the left side is the start, Separation is done sequentially taking the structure. Moreover, FIGS. 16-24 is a figure which shows each data at the time of the said sequential isolation | separation. Hereinafter, each procedure starting from procedure (1) in FIG. 15 will be described.

  In the procedure (1), the normal data N1 in FIG. 7 has the values 50 and 150 as boundaries, and the frequency classification unit 2 performs normal data N1-N1 (value range 50 to 150) shown in FIG. 18, two pieces of non-normal data N 1 -S 1 (value range is 50 or less) and N 1 -S 2 (value range is 150 or more). In the procedure (2), the normal data N1-N1 shown in FIG.

  In the procedure (2) of FIG. 15, an example is shown in which separation convergence is determined by the normal data N1-N1, but when the separation convergence is not determined, the normal data N1-N1 is further determined. The separation process is recursively continued such as -N1 and non-normal data N1-N1-S1 and N1-N1-S2.

  In the procedure (3), the numerical classification unit 3 uses the non-normal data N1-S1 as the boundary and the non-normal data N1-S1-S1 shown in FIG. 19 and the non-normal data N1-S1- shown in FIG. Separate to S2. In the procedure (3a), the numerical classification unit 3 uses the non-normal data N1-S1-S1 as the boundary and the value 28 as the non-normal data N1-S1-S1-S1 shown in FIG. 21 and the non-normal data shown in FIG. Separated into N1-S1-S1-S2.

  In the procedure (3aa), the numerical value classification unit 3 determines separation convergence for the non-normal data N1-S1-S1-S1 shown in FIG. In step (3ab), the numerical value classification unit 3 determines separation and convergence for the non-normal data N1-S1-S1-S2 shown in FIG. On the other hand, in the procedure (3b), the numerical value classification unit 3 determines the separation convergence for the non-normal data N1-S1-S2 shown in FIG.

  Also, in the procedure (4), the numerical value classification unit 3 uses the non-normal data N1-S2 as the boundary with the value 250 as the non-normal data N1-S2-S1 shown in FIG. 23 and the non-normal data N1-S1 shown in FIG. Separate into S2-S2. In the procedure (4a), the numerical classification unit 3 determines separation convergence for the non-normal data N1-S2-S1. In the procedure (4b), the numerical classification unit 3 determines separation convergence for the non-normal data N1-S2-S2.

  As described above, according to the procedure shown in FIG. 15, the normal data N1 is converted into the non-normal data N1-S1-S1-S1, N1-S1 indicated by the frequency classifying unit 2 and the numerical value classifying unit 3 at the end of the tree structure. -Separated into 6 data, 5 normal data of S1-S2, N1-S1-S2, N1-S2-S1 and N1-S2-S2 and 1 normal data N1-N1 .

  The specific examples described above are summarized as follows. After the frequency classification unit 2 separates the numerical data of FIG. 4 as input into the normal data N1 and the non-normal data S1, the frequency classification unit 2 and the numerical classification unit 3 of FIG. 8 and FIG. According to the procedure, the final classification result is classified into a total of 10 data, one normal data and 4 + 5 = 9 non-normal data.

  As described above, the first functional unit 10 classifies the input numerical data into sections that can be regarded as having the same tendency. However, each classified section is further classified by a normal method. May be.

  For example, the average and standard deviation σ are calculated for the data in each classified section, and the average ± ασ range (with the average at the center and the width of σ The section may be further subdivided by a series of sections that are a constant multiple).

  In addition, regarding the section for which the convergence determination was made when the total number of data became less than or equal to the threshold value by the second determination method described above, the application of the statistical method as described above is inappropriate, and the subdivision is It may not be performed.

  Next, details of the estimation unit 5 will be described. First, in order to explain the input to the estimation unit 5, the premise items will be described.

  As described above, the first function unit 10 classifies the numerical data of the predetermined items obtained from the sample group by dividing them into sections of normal data and non-normal data. Here, the said classification | category is possible for every predetermined item which defines numerical data. For example, in the case of medical examination data, the predetermined items can be set from among the various items of the test items. In addition to the triglyceride value used in the example described above, height, γ-GTP (1 of blood test) Item) Others can be classified by obtaining numerical data from each patient group.

  Thus, the input to the estimation unit 5 becomes the respective classification results by the first function unit 10 of the numerical data in the one or more predetermined items. The estimation unit 5 assigns the numerical labels described below to the classification results of the numerical data of the various items in the sample population, and generates a “document” for each sample (each patient if it is medical examination data). Then, LDA is applied to the document set of all samples, and the topic set and the topic ratio in each sample are estimated.

  Therefore, according to the estimation unit 5, for the medical examination data (continuous value data) of various items in the patient group, the first functional unit 10 receives the classification based on the health tendency as an input, A topic set in the patient group and a topic ratio of each patient are estimated. Here, the topic set corresponds to a group of health trends of the patient group, and the topic ratio is a group ratio of the health trend of each patient, and corresponds to a feature vector related to the health trend of each patient. Therefore, the output from the estimation unit 5 can be applied to a health management system or the like.

  Hereinafter, the specific process of the estimation unit 5 will be described by dividing it into a first process, a second process, and a third process.

  In the first process, a natural number label is assigned to each section (normal data or non-normal data) classified for numerical data of a predetermined item according to a predetermined rule. In one example, a rule may be used in which a small value is assigned to a normal data interval, and the value to be assigned is increased as the value deviates from the normal interval. The assignment rule may be determined for each predetermined item for which numerical data is defined.

  An example of the numerical value labeling will be described with reference to an example in which the numerical data D of the medical examination value (neutral fat value) in FIG. 4 is separated as shown in FIGS. 8 and 15. The separation results are listed in order from the data section with the smaller value as follows. Since all data D is first separated into N1 and S1, and N1 is a section having a value smaller than S1, each section in FIG. 15 separated from N1 is each section in FIG. 8 separated from S1. The value is smaller than.

  N1-S1-S1-S1, N1-S1-S1-S2, N1-S1-S2, N1-N1, N1-S2-S1, N1-S2-S2 (each section in Fig. 15), S1-S1 , S1-S2-S1, S1-S2-S2-S1, S1-S2-S2-S2 (Each section in Fig. 8)

  Since it is normal data “N1-N1” that can be regarded as “most normal” in each of the above-mentioned sections, the smallest number “1” is assigned thereto. Thereafter, 2, 3, 4, 5,. When there is no room to be given to either the left or right, it may be given sequentially to the remaining one side. The results given by this rule are listed below. Note that “N1-N1” to which “1” is assigned is expressed as “[1] N1-N1”, and the others are also expressed in the same manner.

  [7] N1-S1-S1-S1, [5] N1-S1-S1-S2, [3] N1-S1-S2, [1] N1-N1, [2] N1-S2-S1, [4] N1-S2-S2 (each section in Fig. 15), [6] S1-S1, [8] S1-S2-S1, [9] S1-S2-S2-S1, [10] S1-S2-S2 -S2 (each section in Fig. 8)

  In the second process, first, a vector having the numerical label given in the first process as an element is generated for the section to which the value of each predetermined item data of each patient belongs. For example, it is assumed that all medical checkup data are prepared for three predetermined items of height, neutral fat value, and γ-GTP value. In this case, the height of a patient A corresponds to the numeric label L1A (that is, the height belongs to the interval GA, and the interval GA is given the numeric label L1A), and the triglyceride value corresponds to the numeric label L2A. If the γ-GTP value corresponds to the numerical label L3A, the vector of the patient A has these elements (L1A, L2A, L3A).

  In the second process, the “document” of the patient is generated by further arranging the words corresponding to the elements of the vector by the label values of the elements. Here, as is well known in LDA, the “document” is a document in the meaning of Bag Of Words, and is not a meaningful document having human readability.

  For example, in the example where all the medical examination data is given by the above three types of predetermined items, if the words corresponding to each element are set as “height”, “neutral fat”, and “alcohol”, respectively. , The document generated for patients with vector (1, 1, 1) is “height, neutral fat, alcohol” and the document generated for patients with vector (3, 2, 1) is “height, "Height, height, neutral fat, neutral fat, alcohol".

  In the above description, a simple vector having the number n of elements corresponding to the number n of types of predetermined items has been described. However, one dummy element is added to the top position and the number of elements is increased to n + 1. It is also preferable to generate a vector. The vector value in the dummy element is a predetermined value common to all patients. The significance is as follows.

  For example, a patient with a bad specific numerical value (a numerical value of an examination item) often has a bad other numerical value. However, the vector of labels = (10, 10, 10) has a vector, generally unhealthy people, and the vector of labels (1, 1, 1) and generally healthy people Although the numerical values are different in size, they are likely to be classified into the same cluster because their directions are the same. So, if you add one dummy element to the simple three-element vector and expand it to four elements, and (a, 10, 10, 10) and (a, 1, 1, 1), the direction is apparent. It is possible to avoid matching and being classified into the same cluster.

  When generating a document from dummy elements, a predetermined dummy word, for example, “dummy” may be used. In the above example, if the simple vector is (1, 1, 1) and the document produces `` height, neutral fat, alcohol '', add a dummy element with its value a = 3 In the case of an extension vector of (3, 1, 1, 1), the generated document will have “a = 3 dummy words at the beginning and“ dummy, dummy, dummy, height, neutral fat, alcohol ” It becomes.

  Hereinafter, in the document generation by the second process, an example of an additional process for performing classification with higher accuracy will be described in the same manner as the addition of the dummy element. The additional processing can be performed in combination with the use of the dummy element.

  FIG. 25 is a diagram showing a flow of additional processing when generating a document by the second processing. In FIG. 25, steps S101 to S105 are shown, and an example EX101 to explain each step is shown in a column C100. Hereinafter, each step will be described with reference to the example.

  In step S101, the total number of words used for document generation is determined in advance for the numerical data of each predetermined item (specifically, numerical data of each inspection item). For example, as shown in Example EX101, the numerical data of the predetermined item is three types related to “neutral fat”, “BMI”, and “blood glucose level” (hereinafter described in FIG. 25 using the example). Then, for each of these, the total number of words is determined as “30 words”. Note that the total number of words may be set differently for each numerical data of the predetermined item.

  In step S102, in addition to the predetermined word for generating the document as described above, a predetermined counter word is determined in advance as corresponding to the predetermined word for the numerical data of each predetermined item. . For example, as shown in Example EX102, a predetermined word for numerical data of a predetermined item is defined as “neutral fat”, “BMI”, and “blood glucose level” with the same name as the numerical data of the predetermined item. In this case, each word is defined as “mathematics”, “English”, and “science”.

  The word to be determined in advance in step S102 may be determined as some word corresponding to the predetermined word. For example, it is necessary to consider the meaning such as using a synonym (strictly interpreted as a meaning interpreted by humans). There is no need to consider.

  In step S103, for each examination item data of each patient (numerical data of a predetermined item of each sample), a numerical label is assigned by the first process in the above-described normal method, so that the second in the above-described normal method. The number of predetermined words in the process is determined. In Example EX103A and Example EX103B, the words “neutral fat”, “BMI”, and “blood glucose level” for patients A and B are “18 words, 20 words, 10 words” and “12 words, 15 An example where it is decided to use "16 words" is shown.

  In step S104, for each examination item data of each patient (numerical data of predetermined items of each sample), the number of words for the paired words determined in step S102 is calculated from the total number of words set in step S101, step S103. This is determined by subtracting the number of the predetermined words determined in (1).

  For example, in the case of Mr. A shown in the example EX103A, the number of words as opposed to the word as shown in the example EX104A is determined. The number of words of “neutral fat” versus the word “mathematics” is subtracted from “18 words” determined for “neutral fat” from the total number of words “30 words” set in the “neutral fat”. Therefore, it is determined as “12 words”. Similarly, regarding the words “English” and “Science” for “BMI” and “blood glucose”, the number of words is determined as “30-20 = 10 words” and “30-10 = 20 words”, respectively. The

  Similarly, in the case of Mr. B shown in the example EX103B, the number of words as shown in the example EX104B is determined. That is, the number of words in each of “math”, “English”, and “science” is “30-12 = 18 words”, “30-15 = 15 words”, and “30-16 = 14 words”, respectively. As determined.

  In step S105, a document for each patient is generated by using the number of words determined in steps S103 and S104 for each predetermined word and its corresponding word in each examination item data. For example, if Mr. A shown in Example EX104A, using “18 words” for “neutral fat”, “20 words” for “BMI”, and “10 words” for “blood glucose level” as predetermined words, A document is generated using “12 words” for “mathematics”, “10 words” for “English”, and “20 words” for “science” as counter words.

  As described above, according to the flow of the additional process in FIG. 25, a predetermined number of counter words can be prepared in addition to the predetermined word, and the document of each patient can be generated. In addition, when applying LDA to the document collection of all patients, it is the same as when not using the said additional process. Note that the number of dummy words described above is set independently of the total number determined in step S101.

  In the flow of FIG. 25, as is clear from the description, steps S101 and S102 need to be executed in advance by the administrator of the numerical data analyzer 1 before reading input data. The process automatically performed by the numerical data analysis apparatus 1 after reading the input data is after step S103.

  In the third process, the LDA is applied to the document set in which the documents prepared for each patient in the second process are collected over all patients, and the topic set and the topic ratio of each patient are estimated. Get the final result by part 5. Since application of the LDA itself is well known, description thereof is omitted.

  As described above, according to the present invention, it is possible to perform classification, that is, discretization by appropriately reflecting the numerical value and frequency deviation in the medical examination data, and it is possible to perform LDA analysis using the classification result.

  Hereinafter, supplementary matters in the present invention will be described.

  (1) The present invention is provided as a numerical data analysis program that is read into a computer and causes the computer to function as the numerical data analysis apparatus 1 in FIG. 1 or cause the computer to execute procedures corresponding to the respective parts in FIG. May be.

  (2) In the present invention, the numerical data to be input is described as being a medical examination value, that is, related to medical care, but data having similar properties, for example, biological data not limited to uses such as medical care and health, The present invention can also be applied to general biological data and the like.

  DESCRIPTION OF SYMBOLS 1 ... Numerical data analysis apparatus, 10 ... First function part, 2 ... Frequency classification part, 3 ... Numerical value classification part, 4 ... Classification result update holding part, 5 ... Estimation part

Claims (11)

A numerical data analysis apparatus that handles numerical data of predetermined items obtained from a sample group,
A frequency classification unit that classifies numerical data based on frequency deviations;
A numerical value classification unit for classifying numerical data based on the numerical value bias,
For the input numerical data, the frequency classification unit classifies normal data and upper and / or lower end non-normal data,
The numerical classification unit further classifies the classified non-normal data into individual non-normal data, and the frequency classifying unit further classifies the classified normal data into individual normal data, upper end and / or Classify it into non-regular data at the bottom,
The numerical data analysis apparatus characterized in that the classification by the numerical classification unit and the frequency classification unit is continued until a predetermined convergence condition is satisfied for each of normal and non-normal data classified individually.   The frequency classifying unit performs sigmoid function fitting on numerical data as a cumulative frequency function, and each of the ranges belonging to the upper end side and the lower end side according to a threshold condition in the fitted sigmoid function The numerical data analysis apparatus according to claim 1, wherein the numerical data analysis apparatus classifies the data other than the classified abnormal data as normal data.   The said numerical classification part performs a regression line fitting with respect to the numerical data as an inverse function of a cumulative frequency function, and classify | categorizes by the threshold value determination with respect to the deviation from the said regression line. Numerical data analysis device described in 1.   4. The numerical data analysis apparatus according to claim 3, wherein the regression line to be applied is made by using a predetermined number of data continuous from the end of the numerical data. In each of the normal and non-normal data classified separately, the predetermined convergence condition is:
The correlation coefficient is determined to be close to 1 by the threshold determination, or
5. The numerical data analysis apparatus according to claim 1, wherein the total number of data is equal to or less than a threshold value. The numerical data analysis device performs classification by the frequency classification unit and the numerical classification unit for numerical data of one or more predetermined items,
An estimation unit that performs the first process, the second process, and the third process;
In the first process, a numerical label is assigned to each of the classified normal and non-normal data in the numerical data of the predetermined item,
In the second process, for each sample constituting the numerical data of the predetermined item, a document based on the numerical label assigned to the numerical data of each predetermined item in the sample is generated,
The third process is characterized by estimating a topic set in the document set and a topic ratio in each sample by applying a latent Dirichlet allocation method to the generated document set for all samples. The numerical data analysis apparatus according to any one of claims 1 to 5.   7. The numerical value according to claim 6, wherein in the first process, the estimation unit assigns a minimum numerical label to the normal data, and assigns a numerical value label having a larger value as it deviates from the normal data. Data analysis device.   When the estimation unit generates the document in the second process, a predetermined word corresponding to the predetermined item corresponding to the number of numerical labels assigned to the numerical data of the predetermined item in the sample The numerical data analysis apparatus according to claim 6, wherein the document is generated as a combination of predetermined dummy words. When the estimation unit generates the document in the second process,
A predetermined word corresponding to the predetermined item, for the number of numerical labels assigned to the numerical data of each predetermined item in the sample;
A predetermined number of words corresponding to the predetermined word corresponding to the predetermined item, by the number obtained by subtracting the number of the given numerical labels from the predetermined total number of words corresponding to the predetermined item;
9. The numerical data analysis apparatus according to claim 6, wherein a document is generated as a combination of the two.   The numerical data analysis device according to claim 1, wherein the numerical data of the predetermined item relates to medical treatment.   A numerical data analysis program for causing a computer to function as the numerical data analysis apparatus according to any one of claims 1 to 10.