KR20230168416A - System for predicting clinical outcome of depression and method thereof - Google Patents

Abstract

The clinical outcome prediction system according to one embodiment includes a BDI questionnaire result receiving unit that receives the user's BDI (Beck Depression Inventory) questionnaire result input from the user terminal, and HAMD (Hamilton Depression Rating Scale) result data corresponding to the input BDI result data. An artificial intelligence model provider that provides a machine-learned decision tree-based artificial intelligence model to output an artificial intelligence model, and an artificial intelligence model that generates the user's HAMD result data based on the user's BDI questionnaire results using the provided artificial intelligence model. It includes a HAMD result data generator, wherein the HAMD result data includes a plurality of items, and the artificial intelligence model provides a plurality of decision tree models based on correlations with a plurality of BDI items for each HAMD item. .

Description

Depression clinical outcome prediction system and clinical outcome prediction method {SYSTEM FOR PREDICTING CLINICAL OUTCOME OF DEPRESSION AND METHOD THEREOF}

The present invention relates to a depression clinical outcome prediction system and a clinical outcome prediction method. More specifically, the present invention relates to a depression clinical outcome prediction system and a clinical outcome prediction method. More specifically, the present invention relates to depression that generates HAMD (Hamilton Depression Rating Scale) outcome data through an artificial intelligence model based on the BDI (Beck Depression Inventory) questionnaire results. It relates to clinical outcome prediction systems and clinical outcome prediction methods.

Digital therapeutics is a technology that uses advanced software (mobile applications, games, virtual reality, chatbots, etc.) to prevent, manage, and treat diseases or disorders. In other words, digital therapeutics are used for the purpose of diagnosing, treating, alleviating, and preventing diseases. Digital therapeutics include various non-face-to-face treatment methods.

Digital therapeutics do not require manufacturing, transportation, and storage like general medicines, and are easy to supply in large quantities at low cost. Therefore, the use of digital therapeutics allows a small number of doctors to manage a large number of patients, which has the effect of reducing medical costs.

Digital therapeutics are relatively free from quarantine measures that must be considered during face-to-face treatment. Therefore, digital therapeutics can be continuously used in medical treatment, and related technology development is active. Some countries have begun to cover digital treatments related to mental illness with national health insurance, and there are cases where some face-to-face treatment related to mental illness has been replaced with digital treatments.

Due to the advantages of digital therapeutics technology, conventional technologies have been developed to diagnose and treat patients using software.

Doctors can determine the degree of a patient's mental illness through patient information (language characteristics, speaking habits, etc.). Depression treatment is based on the patient's responses to the doctor's specific questions (past mental illness, lifestyle habits, environment, etc.). Outpatient treatment for many depressed patients who are not hospitalized is provided at intervals of more than one month, and the doctor in charge checks the mental health status of the depressed patient based on questions for depression diagnosis and the patient's answers at each repeated outpatient treatment. .

Meanwhile, unlike hospitalized depression patients, outpatient depression patients must wait for a doctor's diagnosis until their next treatment. Therefore, it is more difficult for doctors to accurately diagnose and treat depression in outpatient depression patients than in hospitalized depression patients. If the mental health status of a depressed patient during his or her daily life can be communicated to the doctor in charge between outpatient treatments, the doctor in charge can provide more effective treatment than existing treatment methods.

Therefore, for patients with depression, technology is needed to collect the mental health status of the patient's daily life through various methods after treatment and before the patient's visit to the hospital, and to provide the collected results to the patient's doctor.

Meanwhile, many patients diagnosed with mental illness find it difficult to disclose their medical history to those around them. Additionally, some patients diagnosed with mental illness refuse to seek medical help. On the other hand, if mental illness diagnosis and treatment can be done non-face-to-face, even patients who do not receive ongoing treatment can participate in the diagnosis and treatment, allowing more patients to receive treatment than when using existing diagnosis and treatment methods. Therefore, technology that provides various non-face-to-face treatment methods is needed.

Meanwhile, digital therapeutics are also being applied to the treatment of mental illnesses such as depression.

The Beck Depression Inventory (BDI) is a self-report questionnaire consisting of 21 multiple-choice questions used to measure depression in children and adults. However, these self-report scales have several limitations. This is because these scales rely on subjective feelings of depression, so scores may vary depending on patient characteristics such as age, education level, personality traits, and gender. .

On the other hand, since the observer rating scale has relatively little change in score due to patient characteristics, the observer rating scale is preferentially used to measure the severity of depression and treatment response in most studies. The Hamilton Depression Rating Scale (HAMD or HDRS) was developed in 1960 for the purpose of measuring the severity of the disease when studying patients diagnosed with major depressive disorder, but was later used in studies to evaluate the effectiveness of treatment. Its use has expanded and is now considered the standard observer rating scale for depression. This scale consisted of 21 items when it was first developed, but since some items are not common in patients with depression and actually reduce internal consistency, a revised 17-item version with these items deleted is currently available. It is the most widely used.

As mentioned above, the problems with self-report scales such as the BDI questionnaire can be supplemented by clinical results using observer rating scales such as HAMD. However, due to the temporal and spatial burden and psychological resistance for direct questionnaires, it is difficult to provide patients with a smooth clinical trial. There are difficulties in providing results.

The technical problem to be achieved by the present invention is to provide a clinical outcome prediction system and a clinical outcome prediction method that can predict the results of the HAMD questionnaire, an observer evaluation scale, using the results of the BDI questionnaire, a self-report scale, without receiving a direct clinical examination from a doctor who is an observer. It is provided.

One embodiment of the present invention is a clinical outcome prediction system and clinical outcome prediction that can predict HAMD results based on the user's BDI questionnaire results through a decision tree-based artificial intelligence model generated by machine learning BDI data and HAMD data. The purpose is to provide a method.

An embodiment of the present invention provides a clinical outcome prediction system and a clinical outcome prediction method that can generate HAMD outcome data through a plurality of decision tree models provided based on the correlation between BDI questionnaire items and HAMD questionnaire items. The purpose is to

The technical problem to be achieved by the present invention is not limited to the technical problem mentioned above, and other technical problems not mentioned can be clearly understood by those skilled in the art from the description below. There will be.

In order to achieve the above technical task, the clinical result prediction system according to an embodiment includes a BDI questionnaire result receiver that receives the user's BDI (Beck Depression Inventory) questionnaire result input from the user terminal, and a HAMD corresponding to the input BDI result data. (Hamilton Depression Rating Scale) An artificial intelligence model provider that provides a machine-learned decision tree-based artificial intelligence model to output result data, and using the provided artificial intelligence model, based on the results of the user's BDI questionnaire A HAMD result data generator that generates user HAMD result data, wherein the HAMD result data includes a plurality of items, and the artificial intelligence model includes a plurality of items based on correlations between the plurality of BDI items for each of the HAMD items. Provides decision tree models.

The clinical outcome prediction method according to one embodiment is based on BDI (Beck Depression Inventory) data and HAMD (Hamilton Depression Rating Scale) data, and uses a machine-learned decision tree to output corresponding HAMD outcome data from the input BDI outcome data. A step of generating a based artificial intelligence model, receiving the user's BDI questionnaire results from a user terminal, inputting the user's BDI questionnaire results into the artificial intelligence model to generate HAMD result data for the user, and It includes the step of storing the generated HAMD result data for the user in a database and transmitting it to the user terminal according to the user's request.

According to an embodiment of the present invention, the clinical outcome prediction system and clinical outcome prediction method can provide the user with HAMD outcome data predicted through an artificial intelligence model without consulting a doctor based on the BDI questionnaire results entered by the user.

The clinical outcome prediction system and clinical outcome prediction method according to an embodiment of the present invention predicts HAMD results based on the user's BDI questionnaire results through a decision tree-based artificial intelligence model generated by machine learning BDI data and HAMD data. It is predictable.

The clinical outcome prediction system and clinical outcome prediction method according to an embodiment of the present invention generate the user's HAMD outcome data through a plurality of decision tree models provided based on the correlation between BDI questionnaire items and HAMD questionnaire items. You can.

The effects of the present invention are not limited to the effects described above, and should be understood to include all effects that can be inferred from the configuration of the invention described in the description or claims of the present invention.

Figure 1 shows an overview of a clinical outcome prediction system according to an embodiment of the present invention.
Figure 2 is a block diagram showing the configuration of a clinical outcome prediction system according to an embodiment of the present invention.
Figure 3 is a block diagram showing the configuration of an artificial intelligence model providing unit according to an embodiment of the present invention.
Figure 4 is a flowchart showing a method for predicting clinical results according to an embodiment of the present invention.
Figure 5 is a flowchart specifically showing the steps for generating the artificial intelligence model of Figure 4.
Figure 6 is a diagram showing the correlation between BDI data and HAMD data according to an embodiment of the present invention.
Figure 7 is a diagram illustrating a decision tree for calculating HAMD result data for a specific item according to an embodiment of the present invention.
FIG. 8 is a diagram illustrating a decision tree that calculates HAMD result data for items different from those of FIG. 7.

Hereinafter, the present invention will be described with reference to the attached drawings. However, the present invention may be implemented in various different forms and, therefore, is not limited to the embodiments described herein. In order to clearly explain the present invention in the drawings, parts that are not related to the description are omitted, and similar parts are given similar reference numerals throughout the specification.

Throughout the specification, when a part is said to be "connected (connected, contacted, combined)" with another part, this means not only "directly connected" but also "indirectly connected" with another member in between. "Includes cases where it is. Additionally, when a part is said to “include” a certain component, this does not mean that other components are excluded, but that other components can be added, unless specifically stated to the contrary.

The terms used herein are only used to describe specific embodiments and are not intended to limit the invention. Singular expressions include plural expressions unless the context clearly dictates otherwise. In this specification, terms such as “comprise” or “have” are intended to indicate the presence of features, numbers, steps, operations, components, parts, or combinations thereof described in the specification, but are not intended to indicate the presence of one or more other features. It should be understood that this does not exclude in advance the possibility of the existence or addition of elements, numbers, steps, operations, components, parts, or combinations thereof.

In this specification, “module” includes a unit comprised of hardware, software, or firmware, and may be used interchangeably with terms such as logic, logic block, component, or circuit, for example. A module may be an integrated part, a minimum unit that performs one or more functions, or a part thereof. For example, the module may be composed of an application-specific integrated circuit (ASIC).

Hereinafter, embodiments of the present invention will be described in detail with reference to the attached drawings.

Figure 1 shows an overview of the operation of a clinical outcome prediction system according to an embodiment of the present invention.

According to the network environment 1000 in which the clinical result prediction system can operate, the user terminal 200 can generate the user's BDI questionnaire result data through an app or application that provides a BDI questionnaire interface.

The user terminal 200 may be a smart device equipped with a processor, memory, and communication functions, such as a smartphone, tablet PC, note PC, or desktop. Alternatively, it can be a dedicated device, kiosk, or point of care (POC) equipment with display and input capabilities. The user terminal 200 may provide BDI questionnaire result data to the clinical outcome prediction system 100 through a wired or wireless network.

The clinical result prediction system 100 may receive BDI questionnaire result data from the user terminal 200. The clinical result prediction system 100 may generate the user's HAMD result data using an artificial intelligence model based on the user's BDI questionnaire result data received from the user terminal 200. The clinical outcome prediction system 100 may provide the user terminal 200 with an interface for an app, API, or application that generates HAMD outcome data.

Here, the clinical outcome prediction system 100 may be implemented as a server accessible to multiple users, a stand-alone device providing dedicated access, or a cloud system.

The database 300 may store the generated HAMD result data. In one embodiment, the database 300 may provide the stored HAMD result data to the user through an app or application of the user terminal 200 according to the user's request. Alternatively, if there is a user's request, the clinical result prediction system 100 may retrieve the user's HAMD result data stored in the database 300 and transmit it to the user terminal 200.

Through the above-described configuration, the user can receive the results of the HAMD questionnaire, which is an observer evaluation scale, online, based on the results of the BDI questionnaire performed by the user through the user terminal 200.

Figure 2 is a block diagram showing the configuration of a clinical outcome prediction system according to an embodiment of the present invention.

Referring to FIG. 2, the clinical outcome prediction system 100 may include a BDI result data receiving unit 110, an artificial intelligence model providing unit 120, a HAMD result data generating unit 130, and an interface providing unit 140. there is.

The BDI result data receiving unit 110 may receive the user's BDI questionnaire results from the user terminal 200 (see FIG. 1). The BDI result data receiving unit 110 may receive the BDI questionnaire results entered by the user through an application (hereinafter referred to as app) and provide them to the artificial intelligence model providing unit 120.

The artificial intelligence model providing unit 120 may generate an artificial intelligence model and provide it to the HAMD result data generating unit 130. The artificial intelligence model providing unit 120 may generate a decision tree-based artificial intelligence model. More specifically, the artificial intelligence model provider 120 may generate and provide a machine-learned decision tree-based artificial intelligence model to output corresponding HAMD result data from BDI result data. For example, a decision tree-based artificial intelligence model may include a regression model and a classification model.

In one embodiment, the artificial intelligence model provider 120 may generate a machine-learned artificial intelligence model using a plurality of BDI data and a plurality of HAMD data to output specific HAMD data corresponding to specific BDI data. For example, the artificial intelligence model provider 120 may generate a plurality of decision tree models (or decision trees). Here, the plurality of BDI data may correspond to data including a plurality of BDI questionnaire results input through a plurality of users. The plurality of HAMD data may correspond to data including HAMD questionnaire results collected by the same plurality of users. The artificial intelligence model may include correlation data between a plurality of BDI data and a plurality of HAMD data.

The HAMD questionnaire may contain multiple items (HAMD items). The artificial intelligence model may include a plurality of decision tree models that each generate HAMD result data for each of a plurality of items. For example, the plurality of decision tree models may include a first decision tree model for predicting and generating result data for the first item of the HAMD questionnaire, and a first decision tree model for generating result data for the 17th item. 17. It may include a decision tree model. Each decision tree model can be machine learned to generate HAMD result data for each of different HAMD items based on a plurality of BDI questionnaire result data.

The HAMD result data generation unit 130 may generate corresponding HAMD result data based on the BDI questionnaire results entered by the user using the generated artificial intelligence model. The HAMD result data generator 130 may generate HAMD result data for a plurality of items based on the user's BDI results. That is, the HAMD result data generator 130 may input the user's BDI questionnaire results into each of a plurality of decision tree models for a plurality of items to generate HAMD result data for each corresponding user. HAMD result data may include multiple items (or multiple questionnaire items). For example, HAMD result data may be generated for a plurality of items HAMD01, HAMD02 to HAMD10, respectively. Multiple items of HAMD result data may be referred to as HAMD items.

A plurality of decision tree models may be provided based on the correlation between a plurality of BDI questionnaire items for each of a plurality of HAMD items. For example, a decision tree model for the HAMD01 item may be created based on the correlation of the HAMD01 item with each of a plurality of BDI questionnaire items.

The interface provider 140 may provide the user terminal 200 with an interface for inputting BDI questionnaire results, generating HAMD result data, and requesting provision. Users can input BDI questionnaire results through the provided interface and receive corresponding HAMD result data. For example, the interface may include a Clinical Decision Support System (CDSS).

The API provider 400 may provide an API (Application Programming Interface). The API provider 400 can connect the database 300 and the application of the user terminal 200. In one embodiment, when a user requests HAMD result data through an application of the user terminal 200, the API provider 400 receives the HAMD result data from the database 300 and retrieves the HAMD result data from the user terminal 200. It can be delivered to users through an application.

Figure 3 is a block diagram showing the configuration of an artificial intelligence model providing unit according to an embodiment of the present invention.

Referring to Figure 3, the artificial intelligence model providing unit 120 includes a data learning unit 121, an oversampling decision unit 122, an accuracy calculation unit 123, a depth determination unit 124, and a decision tree model creation unit. It may include (125).

In Figure 3, the decision tree model generator 125 makes a plurality of decisions finally determined through the data learning unit 121, the oversampling decision unit 122, the accuracy calculation unit 123, and the depth decision unit 124. Tree models can be created.

The data learning unit 121 can learn BDI data, HAMD data, and correlation data between BDI and HAMD. The data learning unit 121 may learn a plurality of BDI questionnaire result data and a plurality of HAMD result data for a plurality of users. The data learning unit 121 can learn the correspondence relationship between BDI data and HAMD data based on the correlation between BDI and HAMD.

The oversampling decision unit 122 may determine whether oversampling is necessary for items lacking classes among the HAMD data among the learned data. The oversampling decision unit 122 may determine whether to perform oversampling to adjust class imbalance between BDI data and HAMD data. That is, the oversampling decision unit 122 can determine oversampling for necessary items of HAMD through ROS (Random Over Sampling) or SMOTE (Synthetic Minority Over-Sampling Technique) techniques. Here, the class may correspond to a score option in each item of HAMD. For example, the first item may have four classes: 0 points, 1 point, 2 points, and 3 points.

The oversampling decision unit 122 may determine to perform oversampling only on at least some necessary decision tree models among the plurality of decision tree models.

The accuracy calculation unit 123 can calculate how accurate the data learned by the data learning unit 121 of the artificial intelligence model is as an actual index. The accuracy calculation unit 123 may use the Gini index and/or entropy as indicators for calculating the accuracy of the learning data. High accuracy may mean low impurity (or high purity). Gini coefficient and entropy are proportional to impurity. In other words, the Gini coefficient and entropy may be closer to 0 as the accuracy increases and the impurity decreases. In this embodiment, the decision tree-based artificial intelligence model can proceed with separation until the Gini coefficient and entropy converge to 0. That is, if the impurity of the last node is 0, that node can be determined as a terminal node.

The depth determination unit 124 may determine the depth or depth of each of the plurality of decision tree models. The number of nodes, branches, or divisions in a decision tree model may be determined by the size of the depth. In other words, the larger the number of nodes, branches, or separations, the larger the depth can be. In order to prevent overfitting from learning data, the size of the depth must be determined at an appropriate level.

In one embodiment, the size of the depth may be determined based on the accuracy calculated through the accuracy calculation unit 123. The Gini coefficient or entropy, which is an indicator of data accuracy, converges to 0 at a certain depth and can remain at 0 thereafter. In this case, the depth determination unit 124 may reduce the depth through pruning for nodes after a certain depth that first converges to 0. For example, the entropy and/or Gini coefficient calculated through the accuracy calculation unit 123 may equally appear as 0 at a depth of 5 or more. In this case, nodes with a depth of 5 or more can be deleted through pruning, and nodes with a depth of 5 can be determined as terminal nodes.

The decision tree model generator 125 corresponds to each of a plurality of HAMD items through the data learning unit 121, the oversampling decision unit 122, the accuracy calculation unit 123, and the depth determination unit 124. Multiple decision tree models can be created. Each of the plurality of decision tree models may have a different oversampling number and depth size. In one embodiment, the decision tree model generator 125 selects BDI data with the greatest correlation with an item of a specific HAMD among a plurality of learned BDI data to the highest depth of the decision tree model for the item of the HAMD. It can be placed on the root node.

Figure 4 is a flowchart showing a method for predicting clinical results according to an embodiment of the present invention. It will be described below with reference to FIGS. 1 to 3.

According to the clinical outcome prediction method according to one embodiment, the artificial intelligence model providing unit 120 generates a decision tree-based artificial intelligence model through machine learning based on BDI data and HAMD data including the results of each questionnaire. It can be done (step S100). In one embodiment, the artificial intelligence model provider 120 may generate an artificial intelligence model including a plurality of decision tree models that generate HAMD data based on BDI data.

The data receiving unit 110 may receive the BDI questionnaire result data input by the user from the user terminal 200 (step S200). The data receiving unit 110 receives the BDI questionnaire result data entered through an application including the BDI questionnaire interface provided by the user from the interface providing unit 140, and provides the data to the artificial intelligence model providing unit 120 according to the user's request. can do.

The HAMD result data generation unit 130 applies the received user's BDI questionnaire result data to a decision tree-based artificial intelligence model to generate HAMD result data (or clinical result prediction data, hereinafter referred to as HAMD result data) for the user. can be generated (step S300). In one embodiment, the HAMD result data generator 130 may generate HAMD result data for each of the plurality of items with respect to the HAMD result data including the plurality of items. For example, the HAMD result data generator 130 may generate HAMD result data for each item using a plurality of decision tree models included in the artificial intelligence model.

The generated HAMD result data is stored in the database 300 and can be transmitted to the user terminal 200 according to the user's request (step S400).

Figure 5 is a flowchart specifically showing the steps for generating the artificial intelligence model of Figure 4. It will be described below with reference to FIGS. 1 to 3.

In Figure 5, the data learning unit 121 can extract BDI data and HAMD data for each questionnaire item and arrange them in order starting from data with high correlation or correlation (step S110). The data learning unit 121 may machine learn data about a plurality of BDI data, a plurality of HAMD data, and the correlation between the BDI data and the HAMD data. The data learning unit 121 may organize BDIs that will become root nodes (or reference values) for each HAMD item based on the correlation between BDI data and HAMD data. The root node of the decision tree model generated for each item of each HAMD result data can be determined as the BDI data with the highest correlation to the item. The data learning unit 121 can organize BDI data that can become a root node.

The oversampling decision unit 122 may perform oversampling on items with insufficient classes among the HAMD data (step S120). The oversampling decision unit 122 may determine oversampling for items lacking classes among HAMD items. For example, the oversampling decision unit 122 may perform oversampling to four HAMD items with one class.

The artificial intelligence model provider 120 may generate first decision tree models for each HAMD questionnaire item (step S130).

The accuracy calculation unit 123 may calculate accuracy for each of the first decision tree models. For example, the accuracy calculation unit 123 may calculate accuracy using the Gini coefficient and entropy. The depth determination unit 124 may determine the depth of the decision tree models for each item based on the calculated accuracy. That is, the artificial intelligence model providing unit 120 may generate a second decision tree model from the first decision tree models through the accuracy calculation unit 123 and the depth determination unit 124 (step S140). The second decision tree model can prevent overfitting and obtain more accurate results than the first decision tree model through pruning.

Figure 6 is a diagram showing the correlation (correlation) between BDI data and HAMD data according to an embodiment of the present invention. Figures 7 and 8 are diagrams illustrating a decision tree for calculating HAMD result data for specific items according to an embodiment of the present invention.

Figure 6 is a table showing the relationship (correlation) between BDI questionnaire items and HAMD questionnaire items. In Figure 6, HAMD03, HAMD05, and HAMD14 correspond to some of the plurality of HAMD questionnaire items. BDI01 to BDI21 correspond to the BDI questionnaire items. In Figure 6, the correlation of BDI questionnaire items with some HAMD questionnaire items is shown in numbers. For example, the HAMD03 item may have the highest correlation with the BDI09 item at 0.78. HAMD03 is an item about suicide, and BDI09 may also be an item about suicide. In the case of HAMD05, it appears to have the highest correlation with BDI16. For example, HAMD05 may be an item about insomnia, and BDI16 may be an item about changes in life expectancy. In this case, the correlation can show the highest value of 0.71. HAMD14, an item about sexual symptoms, may have the highest correlation with BDI21, an item about loss of interest in sex.

Figure 7 shows the decision tree model for the HAMD03 item. In Figure 7, each node is shown as a rectangle with rounded corners. In this example, the depth of the decision tree model for item HAMD03 is 6. However, it is not necessarily limited thereto, and may be determined to be a depth of 6 or less depending on the Gini coefficient and/or entropy. In this embodiment, the root node 21 is determined as the BDI09 item based on the previously examined correlation. If the proposition in the root node 21 that the class of the BDI09 item is 0.5 or less is true, the process can proceed to the second node 22. In the case of the root node 21, the Gini coefficient is 0.624 and can include 79 sample data. The class is determined to be 0. The decision tree model according to one embodiment may determine the BDI item with the highest correlation as the root node, and the BDI item with the second highest correlation as the second node. That is, the decision tree for a specific HAMD item can be sequentially separated from high to low correlation of a plurality of BDI items with the corresponding HAMD item.

In the case of the second node 22, if the proposition that the class is 2.5 or less for the BDI15 item is true, the class of HAMD03 can be maintained as 0 in the next node as well. In this case, the Gini coefficient could be 0.111. It can be seen that the Gini coefficient of the second node 22 has decreased compared to the Gini coefficient of the root node 21. In the case of the third node 23, the Gini coefficient appears as 0. The third node 23 has a depth of 4. It can be seen that in the nodes below the third node 23, the Gini coefficient remains constant at 0 even as the depth increases. That is, in the case of the decision tree model shown in FIG. 7, the depth can be adjusted to 4 according to pruning.

Figure 8 shows the decision tree model for the HAMD05 item. In Figure 8, the depth of the decision tree model for item HAMD05 is shown as 6. In the case of the root node 31, it can be determined by the BDI16 item. If the result of the questionnaire on BDI16 at the root node 31 is 0.5 or less, the process can proceed to the second node 32. If the questionnaire result of BDI07 at the second node 32 appears to be 1.5 or less, the class of HAMD05 may be maintained at 0. Continuing, when the third node 33 is reached, the Gini coefficient converges to 0. The third node 33 may be determined to be a terminal node. If the second node 32 determines that the result for BDI07 is not 1.5 or more, the class of HAMD05 is determined to be 1. In this case, the Gini coefficient converges to 0, so the class for HAMD05 can be determined immediately.

The decision tree models in FIGS. 7 and 8 are only an example, and it does not exclude that decision tree models different from this embodiment are created for items HAMD03 and HAMD05. For example, for HAMD03, depth may be determined as 4, and for HAMD05, depth may be determined as 5.

The method according to the above-described embodiment of the present invention may be implemented in the form of program instructions that can be executed through various computer components and recorded on a computer-readable recording medium. A computer-readable recording medium may include program instructions, data files, data structures, etc., singly or in combination. Program instructions recorded on a computer-readable recording medium may be specially designed and configured for embodiments of the present invention, or may be known and usable by those skilled in the computer software field. Computer-readable recording media include magnetic recording media such as hard disks, floppy disks, and magnetic tapes, optical recording media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, ROM, RAM, and flash memory. , includes hardware configured to store and execute program instructions. Program instructions include machine language code created by a compiler and high-level language code that can be executed on a computer using an interpreter. The hardware may be configured to operate as one or more software modules for processing the method according to the invention and vice versa.

The method according to an embodiment of the present invention may be executed in an electronic device in the form of program instructions. Electronic devices include portable communication devices such as smartphones and smart pads, computer devices, portable multimedia devices, portable medical devices, cameras, wearable devices, and home appliances.

The method according to an embodiment of the present invention may be provided and included in a computer program product. Computer program products are commodities and can be traded between sellers and buyers. Computer program products may be distributed in the form of machine-readable recording media or online through an application store. In the case of online distribution, at least a portion of the computer program product may be at least temporarily stored or created temporarily in a storage medium such as the memory of a manufacturer's server, an application store's server, or a relay server.

Each component, such as a module or program, according to an embodiment of the present invention may be composed of a single or a plurality of sub-components, and some of these sub-components may be omitted or other sub-components may be added. may be included. Some components (modules or programs) may be integrated into one entity and perform the same or similar functions performed by each corresponding component before integration. Operations performed by modules, programs, or other components according to embodiments of the present invention are executed sequentially, in parallel, repeatedly, or heuristically, or at least some operations are executed in a different order, omitted, or other operations are added. It can be.

The description of the present invention described above is for illustrative purposes, and those skilled in the art will understand that the present invention can be easily modified into other specific forms without changing the technical idea or essential features of the present invention. will be. Therefore, the embodiments described above should be understood in all respects as illustrative and not restrictive. For example, each component described as unitary may be implemented in a distributed manner, and similarly, components described as distributed may also be implemented in a combined form.

The scope of the present invention is indicated by the claims described below, and all changes or modified forms derived from the meaning and scope of the claims and their equivalent concepts should be construed as being included in the scope of the present invention.

100: Clinical outcome prediction system
200: user terminal
300: database
110: data receiving unit
120: Artificial intelligence model provision department
130: HAMD result data generation unit
140: Interface providing unit

Claims (15)

A BDI questionnaire result receiving unit that receives the user's BDI (Beck Depression Inventory) questionnaire result input from the user terminal;
An artificial intelligence model providing unit that provides a machine-learned decision tree-based artificial intelligence model to output corresponding HAMD (Hamilton Depression Rating Scale) result data from the input BDI result data; and
A HAMD result data generator that generates the user's HAMD result data based on the user's BDI questionnaire results using the provided artificial intelligence model,
A clinical outcome prediction system in which the HAMD result data includes a plurality of items, and the artificial intelligence model provides a plurality of decision tree models based on correlations between the HAMD items and a plurality of BDI items. The clinical result prediction system according to claim 1, further comprising an interface provider providing an interface for inputting the BDI questionnaire results and outputting the HAMD result data to the user terminal. The clinical outcome prediction system of claim 1, further comprising a database in which the generated HAMD result data is stored. The method of claim 1, wherein the artificial intelligence model provider,
A data learning unit that learns BDI data and HAMD data; and
A clinical outcome prediction system including a decision tree model generator that generates the decision tree-based artificial intelligence model based on the learned BDI data and the HAMD data. The method of claim 4, wherein the artificial intelligence model provider further includes an oversampling determination unit that determines whether to perform oversampling on a specific item of the HAMD data to adjust class imbalance between the BDI data and the HAMD data. Clinical outcome prediction system. The clinical outcome prediction system of claim 4, wherein the artificial intelligence model providing unit further includes an accuracy calculation unit that calculates accuracy based on the Gini coefficient and entropy for each of the plurality of decision tree models. The method of claim 6, wherein the artificial intelligence model providing unit further includes a depth determination unit that determines a depth for each of the plurality of decision tree models based on the accuracy calculated through the accuracy calculation unit, A clinical outcome prediction system that determines decision tree models of different depths for HAMD items. The clinical outcome prediction system of claim 7, wherein the artificial intelligence model provider arranges the BDI data with the greatest correlation among the plurality of BDI data at the highest depth for each of the plurality of items. The clinical outcome prediction system of claim 8, wherein the HAMD outcome data generator generates the HAMD outcome data for each of the plurality of items using the plurality of decision tree models. Based on Beck Depression Inventory (BDI) data and Hamilton Depression Rating Scale (HAMD) data, generating a machine-learned decision tree-based artificial intelligence model to output corresponding HAMD result data from the input BDI result data;
Receiving the user's BDI questionnaire results from the user terminal;
Inputting the BDI questionnaire results of the user into the artificial intelligence model to generate HAMD result data for the user; and
A clinical result prediction method comprising the step of storing the generated HAMD result data for the user in a database and transmitting it to the user terminal according to the user's request. The method of claim 10, wherein the step of generating the artificial intelligence model includes generating a plurality of decision tree models each generating HAMD result data for a plurality of different items. The method of claim 11, wherein generating the plurality of decision tree models includes calculating accuracy based on a Gini coefficient and entropy for each of the plurality of decision tree models. The clinical result of claim 12, wherein generating the plurality of decision tree models further includes determining a depth for each of the plurality of decision tree models based on the calculated accuracy. Prediction method. The method of claim 13, wherein the step of generating the plurality of decision tree models includes, for each of the plurality of items, arranging the BDI data with the greatest correlation among the plurality of BDI data at the highest depth. Methods for predicting clinical outcomes. The method of claim 11, wherein the step of generating the plurality of decision tree models determines whether to perform oversampling to adjust class imbalance between the BDI data and the HAMD data for a specific item among the plurality of items. A clinical outcome prediction method further comprising the steps of:

Images