KR20190069687A - Multi questionnaire analysis system for depression diagnosis and the method thereof - Google Patents

Abstract

The present invention relates to a multi-questionnaire analysis system for analyzing and providing depression-related medical information based on multiple questionnaires collected through a health examination and a method thereof. By integrating and analyzing different questionnaires while verifying a complex nonlinear relationship between questionnaires related to a depression by using a latent variable model through a deep learning technique, a medical expert can be provided with complex and efficient analysis information.

Description

[0001] MULTI QUESTIONNAIRE ANALYSIS SYSTEM FOR DEPRESSION DIAGNOSIS AND THE METHOD THEREOF [0002]

The present invention relates to a system and method for analyzing multiple questionnaires for diagnosing depression, and more particularly, to a system for analyzing and providing depression-related medical information based on multiple questionnaires collected through health screening.

If depression is left untreated, it can seriously hurt social life and cause serious problems such as suicide. In this regard, depression is often referred to as the coldness of the heart, which is likened to being pneumonia that can lead to death if left untreated even with mild symptoms such as colds.

The suicide rate in Korea is 27.3 per 100,000 population as of 2014, which is the overwhelming number one in the OECD member countries, about 2.5 times the average suicide rate. However, unlike the world's highest level of physical health screening system, which achieves the goal of health improvement and medical cost reduction by early discovery of disease in spite of reality in Korea, mental health examination related to depression is not available due to lack of objective diagnostic technology It is not systematized. It is also supported by the statistics that only 37.7% of patients suffering from mood disorders such as domestic depression use mental health services more than once in their lifetime.

The depression diagnosis process consists of three stages, similar to general medical activities. First, the preliminary diagnosis is made through a questionnaire including questions on demographic statistics, physical health and mental health before the examination, and based on this, based on the analysis, analysis of vital signs such as an examination, blood test, And the screening method is secondarily selected. Then, finally, the complex brain image through fine medical imaging equipment such as fMRI (functional Magnetic Resonance Imaging), MRS (Magnetic Resonance Imaging) and PET (Positron Emission Tomography) Measurement is carried out to perform third-degree precise diagnosis.

In recent years, advances in machine learning technology based on in-depth learning have led to the automated diagnosis and analysis of diseases. Most of the studies related to the diagnosis of depression include diagnosis of depression by analyzing blood signals, brain waves, heartbeats, and other vital signs and brain images. In other words, the research on depression related questionnaires focuses on designing or verifying the validity of the depression related questionnaires (BDI, PHQ, etc.). Even if the questionnaire is extended to other disease related questionnaires, the analysis of the questionnaires by the machine learning methodology is at the level of the feature analysis of the grouping through the most basic Kmeans clustering algorithm.

In other words, the technology for diagnosing and predicting depression as an objective index is still in the early stage in the world, and there is no technology using large-scale group-based machine learning based on health screening.

The purpose of the present invention is to solve the limitations of the existing depression diagnostic questionnaire, fear of stigma and stigma, self-report scale based on sincere answers, artificial cutoff by scale, difficulty of reflecting interactions between scales .

It is also an object of the present invention to integrate multiple different questionnaires using a multimodal latent variable model based on in-depth learning and to provide medical and machine learning analysis related to depression.

It is also an object of the present invention to provide a solution for learning an in-depth learning model by integrating the results of tens of thousands of depression-related questionnaires which are difficult to be unified in one form by using an integrated latent variable model for multiple questionnaires.

The multiple questionnaire analyzing system for diagnosing depression according to an embodiment of the present invention includes a questionnaire history data related to depression of a patient, a collection unit for collecting medical data of the patient and databaseing the questionnaire database, A model learning unit that learns a latent variable model for each of a plurality of survey items from medical data and learns an integrated latent variable model for multiple questionnaires using the learned latent variable model, A diagnostic predictor for converting target questionnaire data related to depression received from a target patient into a latent variable and comparing the converted latent variable with the questionnaire database to predict a diagnosis of depression for the target patient; To provide a diagnosis of depression Includes a transfer unit.

The collecting unit may database each of the question history history data, the medical data, and the personal data related to the depression of the patient with the questionnaire database.

The collecting unit may encrypt the personal data for the patient included in the questionnaire history data, and construct the questionnaire database using the encrypted personal data and the check date as a primary key.

The collecting unit may store one or more tables for each questionnaire type corresponding to the questionnaire type used in the mental health examination in the questionnaire database to store and maintain the questionnaire history data, which is a survey result.

The model learning unit classifies the collected questionnaire history data and medical data into a common question item or a unique questionnaire item, learns the latent variable model in each classified item, and uses each of the learned latent variable models The integrated potential model for the multiple questionnaires can be learned.

Wherein the diagnostic predicting unit converts the survey result of the questionnaire survey data into the latent variable using the integrated latent variable model and compares the converted latent variable with existing latent variables stored and maintained in the survey database The taxon can be judged.

Wherein the diagnostic predicting unit analyzes the importance of the subject questionnaire data in the determined taxon, and compares and analyzes the questionnaire history data stored and maintained in the questionnaire database with the subject questionnaire data to diagnose the subject's depression Can be predicted.

In an operation method of a multiple questionnaire analysis system for diagnosing depression according to an exemplary embodiment of the present invention, a collecting unit collects survey history data related to depression of a patient and medical data of the patient and database them into a questionnaire database A model learning unit learns a latent variable model for each of a plurality of survey items from the collected survey history data and medical data, and learns an integrated latent variable model for multiple questionnaires using the learned latent variable model Wherein the diagnostic predictor converts the subject questionnaire data related to depression received from the subject patient to the latent variable using the integrated latent variable model and compares the converted latent variable with the questionnaire database, To predict the depression diagnosis In the transfer unit, and includes the step of providing a depression diagnosis for patients.

The step of creating a database into the questionnaire database may database each of the questionnaire history data, the medical data, and the personal data related to the depression of the patient with the questionnaire database.

The step of creating a database into the questionnaire database may encrypt the personal data for the patient included in the questionnaire history data and construct the questionnaire database using the encrypted personal data and the check date as a primary key .

The step of forming the database into the questionnaire database may include storing a questionnaire history data as a result of the questionnaire data, constituting one table for each questionnaire type corresponding to the questionnaire type used in the mental health examination in the questionnaire database have.

The step of learning the integrated latent variable model for the multiple questionnaires includes a first step of classifying the questionnaire history data into questionnaire types in the questionnaire database, a common questionnaire item or a unique questionnaire item in each of the classified questionnaires A third step of learning the latent variable model based on a ladder network for the common question item, a third step of learning the latent variable model based on a ladder network based on the unique questionnaire item, A fourth step of learning a latent variable model and a fifth step of learning the integrated latent variable model for the multiple questionnaires using the learned latent variable models.

Wherein the step of predicting the diagnosis of depression for the subject patient comprises the steps of: converting the questionnaire result of the questionnaire survey data to the latent variable using the integrated latent variable model; storing the converted latent variable in the questionnaire database; The taxa can be judged by comparing the existing potential variables to be maintained.

Wherein the step of predicting depression diagnosis of the subject patient comprises analyzing the importance of the subject questionnaire data in the determined taxon group and comparing and analyzing the questionnaire history data stored in the questionnaire database and the subject questionnaire data The diagnosis of depression for the subject can be predicted.

According to the embodiment of the present invention, the complex nonlinear relationship of the questionnaire items related to depression can be grasped by using the latent variable model through the in-depth learning technique, and by analyzing the different questionnaires in an integrated manner, .

1 is a block diagram illustrating a detailed configuration of a multiple questionnaire analysis system for diagnosing depression according to an embodiment of the present invention.
FIG. 2 is a flowchart illustrating a method for analyzing multiple questionnaires for diagnosing depression according to an embodiment of the present invention.
FIG. 3 shows an example of a schematic diagram of an integrated latent variable model according to an embodiment of the present invention.
4A and 4B show examples of latent variable models and integrated latent variable models, respectively, according to an embodiment of the present invention.
FIG. 5 illustrates an example of a graph representing latent variables using an integrated latent variable model according to an embodiment of the present invention.
FIG. 6 shows a result of analyzing the importance of a question item according to an embodiment of the present invention.
FIG. 7 illustrates an example of a graph that provides medical information according to an embodiment of the present invention.

Hereinafter, embodiments according to the present invention will be described in detail with reference to the accompanying drawings. However, the present invention is not limited to or limited by the embodiments. In addition, the same reference numerals shown in the drawings denote the same members.

Also, terminologies used herein are terms used to properly represent preferred embodiments of the present invention, which may vary depending on the viewer, the intention of the operator, or the custom in the field to which the present invention belongs. Therefore, the definitions of these terms should be based on the contents throughout this specification.

The present invention relates to a machine learning-based survey analysis system for rapid diagnosis and analysis of depression, which is based on an in-depth learning method, and includes machine learning which integrates different depression-related questions on multiple questionnaires collected through health screening Algorithm and a system for analyzing and providing depression-related medical information using the same.

As a part of a machine learning based survey algorithm that can replace the existing self - report questionnaire - based assessment for depression diagnosis, integrate the results of various questionnaires, and reflect the multivariate effects of symptom factors, Is a necessary skill to reduce the pain and socioeconomic costs of depression. As a result, the present invention, which is a machine learning based survey analysis system for rapid diagnosis and analysis of depression, is expected to facilitate securing objective diagnostic techniques with high sensitivity and specificity and revitalizing the mental health examination industry.

In general, mental health checkups are subject to evaluation by a psychiatrist only if they are judged to be meaningful in the results of the survey. In the case of questionnaires obtained for large-scale personnel, there is a limitation in that it is difficult to conduct a face-to-face interview with a clinical doctor, and thus it is not possible to create a golden standard for high-risk groups applying in-depth learning based on instructional learning. Accordingly, the present invention can be a solution to such a problem because it uses an in-depth learning model based on non-background learning.

Hereinafter, a multiple questionnaire analysis system and method for diagnosing depression according to embodiments of the present invention will be described in detail with reference to FIG. 1 to FIG.

FIG. 1 is a block diagram illustrating a detailed configuration of a multiple questionnaire analysis system for diagnosing depression according to an embodiment of the present invention. FIG. 2 is a flowchart of a method for analyzing multiple questionnaires for diagnosing depression according to an embodiment of the present invention. FIG.

The multiple questionnaire analysis system 100 for diagnosing depression according to an embodiment of the present invention predicts the diagnosis of depression according to subject questionnaire data related to depression collected from a subject using an integrated latent variable model.

The multiple questionnaire analysis system 100 for diagnosing depression according to an embodiment of the present invention includes a collecting unit 110, a model learning unit 120, a diagnostic predicting unit 130, and a transmitting unit 140 .

In this case, the collecting unit 110, the model learning unit 120, the diagnostic predicting unit 130, and the transmitting unit 140 may be configured to perform the steps 210 to 240 of FIG.

In step 210, the collecting unit 110 collects survey data and patient medical data related to the depression of the patient and stores the collected data in the survey database.

The collecting unit 110 may store and store survey history data, medical data, and personal data related to the depression of the patient in a database. More specifically, the collecting unit 110 collects survey history data including the depression-related information from the patient, encrypts the personal data included in the question history data, and uses the encrypted personal data and the check date as a primary key, An investigation database can be constructed.

First, the collection unit 110 collects questionnaire history data, which is a result of a questionnaire survey made by the patient, on the questionnaire item. For example, the questionnaire provided to the patient may include the following information. The questionnaire may include personal data that can identify a patient, such as name, resident registration number, date of birth, occupation number for an employee, and student number for a student. The questionnaire can also include demographic characteristics of the patient such as age, sex, family composition, marital status, number of friends, number of counselors, job title for employees, grade for students, And physical or mental health-related questionnaires that are indirectly related to health and mental health. More specifically, the questionnaire may include, but is not limited to, smoking, drinking, sleeping, regularity of sleep, attention deficit hyperactivity disorder, and a smartphone addiction scale, and may be directly related to depression and mental illness For example, questions such as Beck Depression Inventory (BDI), Patient Health Questionnaire (PHQ), Generalized Anxiety Disorder (GAD), Resilience Appraisals Scale (RAS), Liebowitz Social Anxiety (LSAS), and Scale of Suicide Ideation .

In this case, the medical data may include personal medical records of patients, for example, duration of illness, abuse of medication, duration of abuse of alcohol, scores according to psychiatric condition measurement scale, PHQ, GAD, BPRS, SANS, SAPS, BDII, BDIII , BHS, SCST, RLI, SSI, and the like.

Then, the collecting unit 110 can encrypt the sensitive personal data of the patient in the question history data. This is a procedure to exclude the personal data from the questionnaire history data, which is the result of the patient survey, and to make it widely available for comparative analysis. The encryption can encrypt the patient's personal data with an encryption key using an encryption algorithm, for example, AES (Advanced Encryption Standard), and hash the personal data using a hash algorithm, for example, SHA (Secure Hash Algorithm) . It is then possible for only a medical professional to examine the patient himself or the patient through the encrypted patient's personal data and provide a medical judgment to index the patient's question history data in the questionnaire database with the patient's password value.

Thereafter, the collecting unit 110 may store a questionnaire history data, which is a survey result, by constructing one table for each questionnaire type corresponding to the questionnaire type used in the mental health examination in the questionnaire database. This is because when the survey history data is stored in the table, the patient's personal data and the date of the examination are used as the primary key to compare and analyze the depression-related medical information, and the patient's questionnaire We want to specify the result of the questionnaire history data.

In step 220, the model learning unit 120 learns a latent variable model for each of a plurality of question items from collected survey history data and medical data, and calculates a combined latent variable model for multiple questionnaires using the learned latent variable model .

More specifically, the model learning unit 120 includes a first step of classifying survey history data into questionnaire types in a questionnaire database, a second step of separating questionnaire items into common questionnaires or unique questionnaires in each of the classified questionnaires, The third step is to learn the latent variable model based on the ladder network, the fourth step to learn latent variable model based on the ladder network for the original questionnaire item, And a fifth step of learning an integrated latent variable model for multiple questionnaires using learned latent variable models.

In the first step, the survey history data is classified into the questionnaire database by the questionnaire type and stored in the table, so that the model learning unit 120 can acquire the stored values corresponding to the plurality of tables. At this time, since the personal data of the encrypted patient forming the primary key of the questionnaire history data by the questionnaire type in the first step and the examination date are not included in the process of learning the integrated latent variable model for the multiple questionnaires, Can be excluded.

In the second step, the model learning unit 120 acquires the intersection of the questionnaire items of the questionnaire types and sets them as a common questionnaire item among the questionnaire items. The unique questionnaire items are the questionnaire items of each questionnaire type and the common questionnaire items It can be obtained by a difference set. In the following steps, the value of each question item can be expressed as a floating point number. In the case of questionnaires that include category values as an exception, category values with linear significance in the order are represented by floating point, and category values without meaning in the order are represented by onehot expression, for example, Can be expressed as 0 for women and 1 for women.

In the third step, the model learning unit 120 improves performance through a lateral connection based on an autoencoder which is a machine learning methodology for learning a latent variable model among a representative non-mapping deep learning method Using a ladder network, you can learn latent variable models for common questionnaires among questionnaires.

In this case, the ladder network may be a non-geodesic artificial neural network model including a purpose of finding a weight that minimizes the cost function L _R as in Equation (1) below.

[Equation 1]

는 인공신경망에 의해 재구성된(reconstructed) 공통적인 설문 문항을 의미하며,

는 인코딩된(encoding)

번째 인공신경망의 은닉층(hidden layer)을 의미한다. 또한,

는 디코딩된(decoding)

번째 인공신경망의 은닉층(hidden layer)을 의미하고, L은 인공신경망의 은닉층 개수를 의미한다.In this case, x represents a common question item among the questionnaires as input values,

Means a common question item reconstructed by an artificial neural network,

Lt; RTI ID = 0.0 >

And the hidden layer of the second artificial neural network. Also,

Decode < / RTI >

(Hidden layer) of the second artificial neural network, and L represents the number of hidden layers of the artificial neural network.

는 (a)공통적인 설문 문항과 인공신경망에 의해 재구성된 공통적인 설문 문항 간의 제곱 편차(squared error), 또는 (b)인코딩된 은닉층과 디코딩된 은닉층 간의 제곱 편차를 각각 얼마만큼의 가중치를 두어 비용함수를 최소화하는지를 정해주는 계수를 의미한다. 이 때, (a)는 [수식 1]에서의

를 나타내며, (b)는 [수식 1]에서의

를 나타내는 것으로, 즉,

는 (a) 및 (b)가 비용함수에 부가될 때의 가중치를 의미한다.together,

(A) a squared error between a common survey item and a common survey item reconstructed by an artificial neural network, or (b) a square deviation between the encoded hidden layer and the decoded hidden layer, It means a coefficient that determines whether the function is minimized. At this time, (a) represents the value of

(B) shows the relationship between

That is,

Means a weight when (a) and (b) are added to the cost function.

을 최종적으로, 설문지들 간의 공통적인 설문 문항에 대한 잠재 변수인

를 학습할 수 있다. Through this optimization problem, the model learning unit 120 uses the ladder network to construct the best hidden layer

, Which is a potential variable for common questionnaires among the questionnaires

Can be learned.

로 사용할 수 있다. In the fourth step, the model learning unit 120 uses a ladder network in the same manner as the third step of learning a latent variable model based on a ladder network for a common question item It is possible to learn the potential variables of the questionnaire items unique to each questionnaire. At this time, the model learning unit 120 learns the same number of ladder networks as the number of questionnaire types, and determines the best hidden layer of each ladder network as a unique questionnaire item for the i-th questionnaire Potential variable

.

In the fifth step, the integrated latent variable model for multiple questionnaires considers each questionnaire as one modal, and it can be assumed that it is a specific example of the latent variable problem for conventional multi-modal. The in-depth learning solution of the latent variable for the ordinary multi-modal uses a method of learning a common feature point by using an autoencoder. In this case, the learning data representing the one-to-one correspondence with the same meaning in different modalities There was a limit that there should be more than a certain amount. In addition, the method can not be applied to the multiple questionnaire analysis system 100 for diagnosing depression according to the embodiment of the present invention. In the case of mental health examination, since the state of the patient changes with each examination, It is not possible to collect learning data that can correspond to one-on-one questionnaires.

Accordingly, in the fifth step, the model learning unit 120 of the multiple questionnaire analysis system 100 for diagnosing depression according to an embodiment of the present invention generates a weakly shared feature point capable of relatively one-to- representation is used to classify common questionnaires and unique questionnaires among the questionnaires, and a one-to-one correspondence can be made to learn the integrated latent variable model for multiple questionnaires.

In this case, weak shared feature point learning by the ladder network can be formed in the process of minimizing the cost function L as expressed by [Equation 2] below.

[Equation 2]

및

는 각각 공통적인 설문 문항과, i번째 설문지에 대한 고유의 설문 문항에 대한 [수식 1]로부터 획득되는 사다리형 회로망(Ladder network)의 비용 함수를 의미한다. 또한,

는 공통적인 설문 문항의 잠재 변수와, i번째 설문지에 대한 고유의 설문 문항의 잠재 변수의 제곱 편차를 의미하며,

는 비용 함수들의 비율을 정하는 가중치를 의미한다. At this time,

And

Is a cost function of a ladder network obtained from [Equation 1] for a common questionnaire item and a unique questionnaire item for the i-th questionnaire, respectively. Also,

, Which means that the latent variable of the common question item and the square deviation of the latent variable of the original question item for the i-th questionnaire,

Is a weight that determines the ratio of cost functions.

를 추가적으로 최소화함으로써, 공통적인 설문 문항, 및 i번째 설문지에 대한 고유의 설문 문항은 제곱 편차를 최소화, 즉 서로 비슷한 분포를 포함하는 약한 공유 특징점(weakly shared representation)으로서 학습될 수 있다. 본 발명의 실시예에 따른 우울증 진단을 위한 다중 설문 분석 시스템(100)은 약한 공유 특징점(weakly shared representation) 학습을 각 설문지에 대해 조정 주기(epoch)마다 변경되는 임의의 순서로 학습함으로써, 최종적으로 공통적인 설문 문항의 잠재 변수 및 모든 설문지에 대한 고유의 설문 문항의 잠재 변수 간의 약간 공유 특징점, 즉 하기의 [수식 3]을 만족할 수 있다. At this time,

, The common questionnaire items and the unique questionnaire items for the i-th questionnaire can be learned as a weakly shared representation that minimizes the squared deviations, i. E., Contains similar distributions. The multiple questionnaire analysis system 100 for diagnosing depression according to an embodiment of the present invention learns weakly shared representation learning in an arbitrary order changed for each questionnaire per adjustment period (epoch) There are some shared feature points between the potential variables of the common questionnaire items and the latent variables of the original questionnaire items for all questionnaires, that is, [Equation 3] below.

[Equation 3]

In step 230, the diagnostic predicting unit 130 converts the subject questionnaire data related to the depression received from the subject patient to the latent variable using the integrated latent variable model, and compares the converted latent variable with the database, Predict the diagnosis.

More specifically, the diagnostic predicting unit 130 receives the questionnaire data on the depression from the patient who is a new health checkup target, receives the questionnaire data, and uses the integrated latent variable model for the learned multiple questionnaire data The results of the survey can be converted into latent variables, and the taxa can be determined by comparing the latent variables converted with the existing latent variables stored in the survey database. Thereafter, the diagnostic predicting unit 130 analyzes the importance of the questionnaire data in the determined taxon, and compares and analyzes the questionnaire historical data with the questionnaire data to predict the depression diagnosis of the subject.

For example, the process of collecting subject questionnaire data from a subject who is a new health checkup subject may be received through the collection unit 110 of the multiple questionnaire analysis system 100 for diagnosing depression according to an embodiment of the present invention .

The diagnosis predicting unit 130 divides the questionnaire data collected from the subject who is the new health checkup target into a questionnaire item common to the questionnaires and a unique questionnaire item for the questionnaires, , It is possible to output the latent variable of the questionnaire items common to each questionnaire and the latent variable of the unique questionnaire item corresponding to the questionnaire.

Thereafter, the diagnostic predicting unit 130 compares the latent variables of the questionnaire items common to each questionnaire with the latent variables of the unique questionnaire items corresponding to the questionnaire and the existing latent variables stored and maintained in the questionnaire database, It is possible to judge whether or not it corresponds. At this time, potential variables can be assigned to a GMM (Gaussian Mixture Model) model to compare which survey data is similar to which taxa in the survey database. The result of the comparison can be calculated as the probability belonging to each taxon by the GMM model.

The diagnostic predicting unit 130 may include a mathematical model, for example, LIME (Local Interpretable Model Analytic Explanations) and a PDA (Prediction Difference Analysis) method for analyzing the importance of feature points that affect the result values of the model regardless of the machine learning model And the importance of the influential questionnaire items can be analyzed.

Thereafter, the diagnosis predicting unit 130 can compare the questionnaire history data of the subject patient stored in the questionnaire database with the subject questionnaire data to predict the depression diagnosis of the subject. For example, the diagnostic predicting unit 130 acquires survey history data of a target patient stored and maintained in the survey database using the encrypted personal data of the target patient as a primary key, and transmits the survey history data to the integration It can be expressed as latent variable using latent variable model, and it can be assigned to GMM model to acquire which taxa are classified respectively. Accordingly, the diagnosis predicting unit 130 can predict the diagnosis of depression in the target patient.

In step 240, the transfer unit 140 provides a depression diagnosis result for the target patient.

For example, in order to more efficiently provide the diagnosis result of the depression to the subject predicted through the diagnostic predicting unit 130 to the medical professional, the transfer unit 140 may include a questionnaire Can be provided as color and numerical values.

That is, the multiple questionnaire analysis system 100 for diagnosing depression according to an embodiment of the present invention uses a machine learning algorithm that integrates different depressive questionnaire items, so that even if the questionnaire items of the questionnaire are not the same, And to provide more efficient and accurate depression-related medical information.

FIG. 3 shows an example of a schematic diagram of an integrated latent variable model according to an embodiment of the present invention.

Common questionnaires among multiple different questionnaires and unique questionnaires for each questionnaires output the latent variable z through each ladder network in the same form as shown in FIG. 3 do. At this time, each latent variable represents a weakly shared representation, so it can show a similar distribution.

4A and 4B show examples of latent variable models and integrated latent variable models, respectively, according to an embodiment of the present invention.

More specifically, FIG. 4A shows distribution results of latent variables extracted using a latent variable model using only a common survey item, except for a unique questionnaire item for each questionnaire to integrate multiple questionnaires. 4b shows the distribution of latent variables extracted using the integrated latent variable model for the multiple questionnaires proposed in the present invention.

4A and 4B illustrate the results obtained by using the results of three different types of depression-related questionnaires, which are actually different from each other, in two dimensions using tSNE (tDistributed Stochastic Neighbor Embedding) . Each model is composed of three layers with 500 hidden nodes, and finally can output 4-dimensional latent variables.

The left graph of each of FIGS. 4A and 4B is a result of clustering the learned latent variable into two clusters using a GMM (Gaussian Mixture Model), and the red and blue points are latent variables included in the respective clusters , And cyan points indicate potential variables for the survey of patients who were found to have depression as a result of the actual diagnosis.

4A and FIG. 4B are graphs showing the results of substituting the PHQ (Patient Health Questionnaire) value, which is a question directly related to depression, which is confirmed primarily in diagnosing depression of medical professionals, into latent variables, The closer to the red value, the greater the degree of depression. On the other hand, the closer to the blue value, the more normal it is.

That is, referring to FIGS. 4A and 4B, the integrated latent variable model for multiple questionnaires proposed in the multiple questionnaire analysis system for diagnosing depression according to an embodiment of the present invention relies more on data of actual depressed patients can confirm.

4A and 4B, the integrated latent variable model for multiple questionnaires proposed by the multiple questionnaire analysis system for diagnosing depression according to an embodiment of the present invention consistently shows PHQ as a latent variable, The cluster is also classified into the group with relatively high PHQ and the group with no PHQ.

FIG. 5 illustrates an example of a graph representing latent variables using an integrated latent variable model according to an embodiment of the present invention.

More specifically, FIG. 5 is a graph showing latent variables for questionnaires directly or indirectly related to depression among questionnaires unique to each questionnaires using an integrated latent variable model for multiple questionnaires according to an embodiment of the present invention Fig.

5, the graphs from the left show the BDI (Beck Depression Inventory) indicating the degree of depression directly, the GAD (Generalized Anxiety Disorder) indicating the degree of having generalized anxiety disorder, and the RSES Rosenberg's SelfEsteem Scale). In this case, the higher the index, the closer to the red value, and the lower the index, the closer to the blue value.

In addition to the BDI, which is directly related to the degree of depression, it can be seen that both GAD, which is higher in depression, and RSES, which is lower in level, show consistent results. This fact confirms that the integrated latent variable model proposed in the multiple questionnaire analysis system for diagnosing depression according to the embodiment of the present invention works smoothly.

FIG. 6 shows a result of analyzing the importance of a question item according to an embodiment of the present invention.

More specifically, FIG. 6 shows the results of analyzing the importance of the questionnaire items that are influential in determining the target patient as the taxa. In this case, the analysis result is the result of analyzing the importance of the questionnaire items influencing the actual depressive patients (or patients) to judge depressed high risk group using LIME (Local Interpretable Modelagnostic Explanations) model.

The prediction probabilities shown on the left side of FIG. 6 indicate that the probability of belonging to the high-risk group of depression is 100% as a result of clustering of the subject patients by the GMM. In addition, the middle graph in FIG. 6 shows the influence of each question item on the diagnosis of depression. As the corresponding value is larger, the influence of the questionnaire results on the determination of the subject group .

In addition, referring to the right table of FIG. 6, the values of the survey results and the influence of the values on the taxon groups are shown. In other words, referring to FIG. 6, in a depressed patient, although a part of the PSQI (Pittsburgh Sleep Quality Index) value indicating the quality of sleep is normal, the effect of the LSAS1 and LSAS2 indicating the sum of the PSQI values, the degree of social anxiety, . ≪ / RTI >

FIG. 7 illustrates an example of a graph that provides medical information according to an embodiment of the present invention.

More specifically, FIG. 7 illustrates an example of a graph that provides comparative analysis of questionnaire history data and subject questionnaire data for a subject patient stored in a questionnaire database according to an embodiment of the present invention, and provides medical information according to the analysis result It is.

FIG. 7 shows the results of a questionnaire survey of patients who were diagnosed with depression in January 2016 and who were treated with related drugs, and the potential variables of the patients were changed from a high risk group to a low risk group . In addition, it can be confirmed that the related question item PHQ value is lowered.

Accordingly, the multiple questionnaire analysis system for diagnosing depression according to the embodiment of the present invention uses a machine learning algorithm that integrates the different depression questionnaire items, so that even if the questionnaire items of the questionnaire are not the same, To provide more efficient and accurate depression-related medical information.

The apparatus described above may be implemented as a hardware component, a software component, and / or a combination of hardware components and software components. For example, the apparatus and components described in the embodiments may be implemented within a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable array (FPA) A programmable logic unit (PLU), a microprocessor, or any other device capable of executing and responding to instructions. The processing device may execute an operating system (OS) and one or more software applications running on the operating system. The processing device may also access, store, manipulate, process, and generate data in response to execution of the software. For ease of understanding, the processing apparatus may be described as being used singly, but those skilled in the art will recognize that the processing apparatus may have a plurality of processing elements and / As shown in FIG. For example, the processing unit may comprise a plurality of processors or one processor and one controller. Other processing configurations are also possible, such as a parallel processor.

The software may include a computer program, code, instructions, or a combination of one or more of the foregoing, and may be configured to configure the processing device to operate as desired or to process it collectively or collectively Device can be commanded. The software and / or data may be in the form of any type of machine, component, physical device, virtual equipment, computer storage media, or device , Or may be permanently or temporarily embodied in a transmitted signal wave. The software may be distributed over a networked computer system and stored or executed in a distributed manner. The software and data may be stored on one or more computer readable recording media.

The method according to an embodiment may be implemented in the form of a program command that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, and the like, alone or in combination. The program instructions to be recorded on the medium may be those specially designed and configured for the embodiments or may be available to those skilled in the art of computer software. Examples of computer-readable media include magnetic media such as hard disks, floppy disks and magnetic tape, optical media such as CDROMs and DVDs, magnetic optical media such as floppy disks, magnetooptical media, and hardware devices specifically configured to store and execute program instructions such as ROM, RAM, flash memory, and the like. Examples of program instructions include machine language code such as those produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter or the like. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed exemplary embodiments. For example, it is to be understood that the techniques described may be performed in a different order than the described methods, and / or that components of the described systems, structures, devices, circuits, Lt; / RTI > or equivalents, even if it is replaced or replaced.

Therefore, other implementations, other embodiments, and equivalents to the claims are also within the scope of the following claims.

Claims (15)

A collection section for collecting survey data related to the depression of the patient and medical data of the patient and database it into a questionnaire database;
A model learning unit that learns a latent variable model for each of a plurality of survey items from the collected survey history data and medical data and learns an integrated latent variable model for multiple questionnaires using the learned latent variable model;
The integrated latent variable model is used to convert subject questionnaire data related to depression received from the subject patient into latent variables and to compare the converted latent variable with the questionnaire database to diagnose depression diagnosis A prediction unit; And
A transfer unit for providing the result of the depression diagnosis on the subject patient,
Multiple questionnaire analysis system for the diagnosis of depression. The method according to claim 1,
The collecting unit
Wherein the questionnaire history data, the medical data, and the personal data related to the depression of the patient are each databaseed into the questionnaire database. 3. The method of claim 2,
The collecting unit
Encrypting the personal data for the patient included in the questionnaire history data, and building the questionnaire database using the encrypted personal data and the check date as a primary key. 3. The method of claim 2,
The collecting unit
A plurality of questionnaire analysis systems for diagnosing depression in which one table for each questionnaire type corresponding to the questionnaire type used in the mental health examination is stored in the questionnaire database and the questionnaire history data as a survey result is stored and maintained. The method according to claim 1,
The model learning unit
The method of claim 1, further comprising the steps of: classifying the collected questionnaire history data and medical data into a common question item or a unique questionnaire item, learning the latent variable model in each classified question, Multiple questionnaire analysis system for depression diagnosis that learns the integrated latent variable model. The method according to claim 1,
The diagnostic predicting unit
The survey result of the questionnaire survey data is converted into the latent variable using the integrated latent variable model and the existing latent variable stored in the survey database is compared with the converted latent variable to determine the taxa Multiple questionnaire analysis system for depression diagnosis. The method according to claim 6,
The diagnostic predicting unit
A diagnosis of depression is made by analyzing the importance of the subject questionnaire data in the determined taxa group and comparing the questionnaire history data stored in the questionnaire database and the subject questionnaire data to analyze the subject, Multiple survey analysis system for. A method for operating a multiple questionnaire analysis system for diagnosing depression,
Collecting survey history data related to depression of a patient and medical data of the patient and collecting the data into a questionnaire database;
Learning a latent variable model for each of a plurality of survey items from the collected survey history data and medical data and learning an integrated latent variable model for multiple questionnaires using the learned latent variable model, ;
The diagnostic predictor may convert the subject questionnaire data related to the depression received from the subject patient to the latent variable using the integrated latent variable model and compare the converted latent variable with the questionnaire database to determine the depression Predicting a diagnosis; And
In the transfer section, a step of providing depression diagnosis results to the target patient
A method for analyzing multiple questionnaires for the diagnosis of depression. 9. The method of claim 8,
The step of databaseing the questionnaire database
Wherein the questionnaire history data, the medical data, and the personal data related to depression for the patient are each databaseed in the questionnaire database. 10. The method of claim 9,
The step of databaseing the questionnaire database
Encrypting the personal data for the patient included in the questionnaire history data, and constructing the questionnaire database using the encrypted personal data and the check date as a primary key. 10. The method of claim 9,
The step of databaseing the questionnaire database
A plurality of questionnaire analysis systems for diagnosing depression in which one table for each questionnaire type corresponding to the questionnaire type used in the mental health examination is stored in the questionnaire database and the questionnaire history data as a survey result is stored and maintained. 9. The method of claim 8,
The step of learning the integrated latent variable model for the multiple questionnaires
A first step of classifying the questionnaire history data in the questionnaire database according to the type of the questionnaire;
A second step of separating each of the classified questionnaires into a common question item or a unique questionnaire item;
A third step of learning the latent variable model based on a ladder network for the common question item;
A fourth step of learning the latent variable model based on the ladder network for the unique question item; And
A fifth step of learning the integrated latent variable model for the multiple questionnaires using the learned latent variable models,
A method for analyzing multiple questionnaires for the diagnosis of depression. 9. The method of claim 8,
The step of predicting the depression diagnosis of the subject patient
The survey result of the questionnaire survey data is converted into the latent variable using the integrated latent variable model and the existing latent variable stored in the survey database is compared with the converted latent variable to determine the taxa Multiple questionnaire method for the diagnosis of depression. 14. The method of claim 13,
The step of predicting the depression diagnosis of the subject patient
A diagnosis of depression is made by analyzing the importance of the subject questionnaire data in the determined taxa group and comparing the questionnaire history data stored in the questionnaire database and the subject questionnaire data to analyze the subject, Multiple survey analysis system for. 15. A computer program stored in a computer-readable medium for performing the method of any one of claims 8 to 14.

Images