Apparatus and method for predicting level of depression by using natural language processing and explainable artificial intelligence

Abstract

The present invention relates to an apparatus and method for predicting a level of depression by using natural language processing and explainable artificial intelligence. According to the present invention, the apparatus for predicting a level of depression by using natural language processing and artificial intelligence includes: a data input unit for receiving an input of patient interview data in which appeals of a target patient are recorded; a data processing unit that tokenizes the patient interview data through natural language processing; and a classification unit for classifying a severity level of depression and normality of the target patient by inputting pieces of tokenized natural language expression data into a pre-trained depression level prediction model.

Description

Device and method for predicting depression levels using natural language processing and explainable artificial intelligence

[0001] The present invention relates to a device and method for predicting depression levels using natural language processing and explainable artificial intelligence, and more particularly, to a device and method for predicting depression levels using natural language processing and explainable artificial intelligence, which can predict the severity level of a patient's depression and whether it is normal or not using natural language processing and explainable artificial intelligence.

[0002] Traditional methods of diagnosing mental illness rely on clinical data recorded by psychiatrists, along with standardized questionnaires, to categorize mental illness based on their knowledge and experience. Consequently, even among psychiatrists, there are significant discrepancies in the diagnosis of mental illness, often due to differences in knowledge and experience.

[0003] Moreover, even when the same patient is treated by the same psychiatrist, there are many cases where the diagnosis made at the first examination is different from the diagnosis made at the follow-up examination about three months later.

[0004] This results in changes in the medication administered based on the psychiatric diagnosis at the first visit and the medication administered based on the psychiatric diagnosis at the follow-up visit, which increases the patient's confusion when receiving treatment, and there is an obstacle to the realization of personalized precision medicine due to the fundamental problem of not being able to make an accurate diagnosis the first time.

[0005] Additionally, the level or scope of treatment may vary depending on the severity of depression.

[0006] Moreover, it takes a lot of time and resources to train psychiatrists with a wealth of knowledge and experience, and even if they are trained, the number of patients with mental illness is increasing significantly at present and in the future, so the demand for mental health diagnosis is expected to increase significantly, but the supply is expected to be insufficient.

[0007] Therefore, a platform capable of predicting the severity of depression using only medical data is required through the development of a specialized artificial intelligence system.

[0008] The technology underlying the present invention is disclosed in Korean Patent Publication No. 10-2023-0101328 (published on July 6, 2023).

[0009] The present invention relates to a device and method for predicting the level of depression using natural language processing and explainable artificial intelligence, which can predict the level of depression severity and normality by applying the results of natural language processing of patient interview data containing patient complaints to an artificial intelligence model.

[0010] The present invention relates to a device for predicting depression levels using natural language processing and artificial intelligence, comprising: a data input unit for receiving patient interview data in which complaints of a target patient are recorded; a data processing unit for tokenizing the patient interview data through natural language processing; and a classification unit for inputting the tokenized natural language expression data into a pre-trained depression level prediction model to classify the depression severity level of the target patient and whether it is normal.

[0011] In addition, the data input unit can receive patient interview data from a CDW (Clinical Data Warehouse) containing natural language expressions recording the complaints of the target patient.

[0012] In addition, the depression level prediction device may further include a learning unit that learns the depression classification model using patient interview data and depression diagnosis results by medical staff for multiple patients in the past, and learns the model using patient interview data and depression diagnosis results for patients from whom the same diagnosis result was obtained N or more times (N is an integer greater than or equal to 2) during a set period among all patients.

[0013] Additionally, the classification unit can classify the severity level of depression of the target patient into one of the labels of Moderate, Mild, and Normal.

[0014] In addition, the classification unit can classify the patient's depression severity level and whether it is normal by applying natural language expression data whose appearance frequency value in the patient interview data is counted to a pre-learned depression level prediction model.

[0015] In addition, the depression level prediction device may further include a provision unit that analyzes natural language expression data of a target patient input into the depression level prediction model and provides, by artificial intelligence, the analysis results of which natural language expressions were used to derive the classification results of the depression level of the target patient.

[0016] And, the present invention provides a method for predicting depression levels, which is performed by a depression level predicting device using natural language processing and artificial intelligence, comprising the steps of: receiving patient interview data in which complaints of a target patient are recorded; tokenizing the patient interview data through natural language processing; and inputting the tokenized natural language expression data into a pre-learned depression level predicting model to classify the depression severity level and normality of the target patient.

[0017] In addition, the depression classification model may be trained using patient interview data and depression diagnosis results by medical staff for multiple patients in the past, and may be trained using patient interview data and depression diagnosis results for patients who have had the same diagnosis result N or more times (N is an integer greater than or equal to 2) during a set period among all patients.

[0018] In addition, the method for predicting the level of depression may further include a step of analyzing the natural language expression data of the target patient input into the depression level prediction model and presenting, by artificial intelligence capable of explaining the analysis results regarding which natural language expressions were used to derive the classification results of the level of depression for the target patient.

[0019] According to the present invention, the level of severity of depression and whether it is normal can be predicted without a specialist's direct judgment by applying the results of natural language processing of the patient's complaint record to an artificial intelligence model.

[0020] Additionally, by using an explainable artificial intelligence model, specialists can intuitively identify words that influence severity classification, thereby streamlining the diagnostic judgment work of psychiatrists, allowing them to focus on essential treatment of patients with depression.

[0021] FIG. 1 is a diagram illustrating a depression level prediction system using natural language processing and explainable artificial intelligence according to an embodiment of the present invention.

[0022] Figure 2 is a drawing explaining the configuration of the depression level prediction device illustrated in Figure 1.

[0023] FIG. 3 is a diagram illustrating a method for predicting depression levels according to an embodiment of the present invention.

[0024] Figures 4a and 4b are diagrams showing examples of providing analysis results by explainable artificial intelligence.

[0025] Hereinafter, embodiments of the present invention will be described in detail with reference to the attached drawings so that those skilled in the art can easily practice the present invention. However, the present invention may be implemented in various different forms and is not limited to the embodiments described herein. In addition, in the drawings, parts irrelevant to the description have been omitted to clearly explain the present invention, and similar parts have been designated with similar reference numerals throughout the specification.

[0026] Throughout the specification, when a part is said to be "connected" to another part, this includes not only the cases where the parts are "directly connected" but also the cases where the parts are "electrically connected" with other elements intervening. Furthermore, when a part is said to "include" a component, this does not exclude other components, but rather includes other components, unless otherwise stated.

[0027] FIG. 1 is a diagram illustrating a depression level prediction system using natural language processing and explainable artificial intelligence according to an embodiment of the present invention.

[0028] As shown in Fig. 1, a depression level prediction system according to an embodiment of the present invention may include a depression level prediction device (100) and a user terminal (200).

[0029] The depression level prediction device (100) according to an embodiment of the present invention can be connected to a user terminal (200) via a wired, wireless, or wired / wireless combination network to exchange information with each other. The wireless network may include at least one of RF, WLAN, Wi-Fi, and Bluetooth, and may utilize various known wireless network methods.

[0030] This depression level prediction device (100) may be implemented as an online or offline platform such as a web server or app server that provides an artificial intelligence-based depression level prediction service to a network-connected user terminal (200), or may be implemented in the form of an application program, application, etc. on the user terminal (200).

[0031] A depression level prediction device (100) can provide a depression level prediction service platform implemented as an app or web to multiple network-connected user terminals (200). The service platform can be an application program running in an app or web environment.

[0032] In this way, the depression level prediction device (100) can be implemented as an application program (application) running on a platform server (Server) or a user terminal (200) that provides a depression level prediction service, and the user terminal (200) can receive related services by being connected to the system (100) via a network while the related application program is running.

[0033] The user terminal (300) may include a device that can connect to a wired or wireless network and exchange information, such as a PC, desktop, smart phone, tablet, notebook, or pad.

[0034] Figure 2 is a drawing explaining the configuration of the depression level prediction device illustrated in Figure 1.

[0035] As shown in FIG. 2, the depression level prediction device (100) according to an embodiment of the present invention includes a data input unit (110), a data processing unit (120), a classification unit (130), and may further include a provision unit (140) and a learning unit (not shown). Here, the operation of each unit (110 to 140) and the data flow between each unit may be controlled by a control unit (not shown).

[0036] This depression level prediction device (100) may be implemented as a computer device that is physically configured and includes a processor, memory, a user interface input / output device and a storage device, a network input / output unit, etc., or may be implemented as an application program running on a computer device or a user terminal.

[0037] The data input unit (110) can receive patient interview data containing the target patient's complaints from the user terminal (200). At this time, the data input unit (110) can receive patient interview data from the CDW (Clinical Data Warehouse) containing natural language expressions containing the target patient's complaints. The CDW corresponds to expert diagnosis data, and utilizing it as learning data can increase the accuracy of predicting depression levels.

[0038] The data processing unit (120) can tokenize natural language expressions by performing natural language processing (NLP) on the input patient interview data. More specifically, the data processing unit (120) can refine and tokenize natural language expressions in the patient interview data using natural language programming techniques.

[0039] The classification unit (130) can input tokenized natural language expression data for patient interview data into a pre-learned depression level prediction model to classify the depression severity level and normality of the target patient.

[0040] At this time, natural language expression data with the frequency of occurrence of the natural language expression counted can be used as input data for a depression level prediction model.

[0041] In this way, the classification unit (130) can classify the patient's depression severity level and whether it is normal by applying natural language expression data whose appearance frequency values ​​in the patient interview data are counted to a depression level prediction model.

[0042] The classification unit (130) can classify the depression severity level of the target patient into one of three labels: Moderate, Mild, and Normal, using a depression level prediction model. The depression level prediction model can be pre-trained by artificial intelligence and implemented using a deep learning-based intelligent algorithm.

[0043] A depression level prediction model can be pretrained by a learning unit (not shown). The learning unit can train a depression classification model using patient interview data and medical staff diagnoses of depression for multiple patients in the past (e.g., patients diagnosed with mild depression at all three initial or follow-up visits within a six-month period). The multiple patients may include a normal patient group, a severe depression group, and a mild depression group.

[0044] This learning unit can pre-train a depression classification model by using natural language data processed from patient interview data on multiple normal and depressed patients in the past as input data and labeling values ​​corresponding to the medical staff's diagnosis results for the corresponding patients (e.g., '2' = Moderate, '1' = Mild, '0' = Normal) as output data.

[0045] At this time, for the purpose of diagnosis accuracy, among all patients for whom patient interview data exists, training can be performed using patient interview data and depression diagnosis results for patients for whom the same diagnosis result was obtained N or more times (N is an integer greater than or equal to 2) during a set period.

[0046] For example, only clinical data collected for patients who received the same diagnosis at least three times within six months (e.g., severe depression in all three diagnoses, including the initial and follow-up visits) can be used as training data.

[0047] There are many cases where the diagnosis determined by the same specialist at the first examination for the same patient is different from the diagnosis determined at the follow-up examination three months later. In the embodiment of the present invention, classification accuracy can be increased by not utilizing such data in learning the classification model.

[0048] The provision unit (140) can output and provide the depression level (level) of the target patient classified by the depression classification model to the user terminal (200).

[0049] In addition, the provision unit (140) can analyze the natural language expression data of the target patient input into the depression level prediction model and present the analysis results on which natural language expressions were used to derive the depression level classification results for the patient through eXplainable Artificial Intelligence.

[0050] In this way, according to the present invention, a depression level prediction system based on artificial intelligence can be provided that can efficiently predict and classify the level of depression severity and whether it is normal by natural language processing patient interview data of CDW containing natural language expressions recording complaints of patients with depression and then training it with artificial intelligence.

[0051] FIG. 3 is a diagram illustrating a method for predicting depression levels according to an embodiment of the present invention.

[0052] First, the depression level prediction device (100) receives patient interview data from CDW in which the target patient's complaints are recorded (S310), and can refine and tokenize the input patient interview data through natural language processing (S320).

[0053] In this way, in the embodiments of the present invention, expert diagnostic data is utilized as input data necessary for predicting depression levels. Furthermore, since the entire data is applied in natural language processing, morphological analysis is completely unnecessary.

[0054] Next, the depression level prediction device (100) inputs tokenized natural language expression data into a pre-trained depression level prediction model, thereby classifying the target patient's depression severity level and whether it is normal (S330). At this time, the depression level prediction model can be used to classify the depression severity level into one of severe, mild, and normal.

[0055] Next, the depression level prediction device (100) can output and provide classification results through the user terminal (200) (S340). Here, the classification results can include not only depression classification classes but also probability values ​​for each class.

[0056] In addition, the depression level prediction device (100) analyzes the input natural language expressions through explainable artificial intelligence to determine the reason for the depression severity level or the probability of presenting a normal classification result, and provides the analysis results of which expressions led to such a prediction diagnosis result through explainable artificial intelligence, thereby helping to identify the cause for diagnosing depression and to determine the treatment range.

[0057] The following describes a specific example of the learning process of the prediction model.

[0058] First, the learning unit can filter out patients with mental illness from the CDW data and refine only the natural language expressions of these patients. Furthermore, it can label the patients with the mental illness they were diagnosed with. Data on patients in the normal group can be downloaded from the AI-HUB public data, which contains natural language data with normal expressions. Data on patients in the CDW data can also be used.

[0059] Table 1 exemplifies the data used to train a depression level prediction model according to an embodiment of the present invention. The training data can be structured according to the following items: PID (subject anonymized identifier), Re-examination_diagnosis, document (natural language expression), and label (diagnosed mental illness or normal group).

[0060]

[0061] Table 2 is a diagram exemplarily showing the label configuration and data source of learning data according to an embodiment of the present invention.

[0062] LabelClassData Source2ModerateDepressive EpisodeBundang Cha Hospital CDW1MildDepressive EpisodeBundang Cha Hospital CDW0NormalAI-HUB Korean Persona Dialogue Data and Multi-session Korean Dialogue Data

[0063] Data can be divided into training and test sets for AI learning. Furthermore, the natural language expressions in the documents shown in Table 1 can be tokenized using natural language programming techniques. Afterwards, the most frequent expressions can be counted for each patient within the tokenized natural language expressions.

[0064] In addition, by performing artificial intelligence learning by linking natural language expression data with counted high-frequency expressions with existing labels, an artificial intelligence model that has learned high-frequency expressions related to mental illness can be obtained.

[0065] From then on, the patient's complaint data is processed using natural language processing and then fed into a depression level prediction model trained on natural language expressions. This allows the patient's depression level to be classified into one of three categories: Moderate, Mild, and Normal. Probabilities for each category, including Moderate, Mild, and Normal, are also provided, supporting efficient diagnosis by medical professionals.

[0066] Figures 4a and 4b are diagrams illustrating examples of analysis results provided by explainable artificial intelligence. Figures 4a and 4b show data analyzing natural language expressions entered through patient interview data for the patient with the highest probability of mild depression, to explain the reason for the results.

[0067] Figure 4a numerically shows the importance of each symptom text used in the classification of depression diagnoses, while Figure 4b graphically shows the importance of each text. It also highlights relevant text within the text for natural language expressions.

[0068] According to the present invention as described above, the severity level of depression and whether it is normal can be accurately predicted and diagnosed and classified without a psychiatrist directly judging the records of patients' mental complaints.

[0069] In addition, by streamlining the diagnostic work of psychiatrists, we can focus on essential treatment for patients with depression and on researching new methods to treat patients with mental illness.

[0070] In addition, by diagnosing depression by dividing it into multiple classes, from moderate to mild and normal, it can provide medical staff with detailed assistance in determining the dose adjustment of antidepressant medication and treatment method.

[0071] The technology of the present invention can be systematically incorporated into a digital treatment platform that automates the diagnosis of depression and supports the diagnosis of patients with depression.

[0072] While the present invention has been described with reference to the embodiments illustrated in the drawings, these are merely exemplary, and those skilled in the art will understand that various modifications and equivalent alternative embodiments are possible. Therefore, the true scope of technical protection of the present invention should be determined by the technical spirit of the appended claims.

Claims

1. In a device for predicting depression levels using natural language processing and artificial intelligence, A data entry unit for entering patient interview data containing the complaints of the target patient; A data processing unit that tokenizes the above patient interview data through natural language processing; and A depression level prediction device including a classification unit that inputs tokenized natural language expression data into a pre-trained depression level prediction model to classify the depression severity level and normality of a target patient.

2. In claim 1, The above data input section, A depression level prediction device that receives patient interview data from the CDW (Clinical Data Warehouse) containing natural language expressions recording the complaints of the above target patients.

3. In claim 1, The depression classification model is trained using patient interview data and depression diagnosis results by medical staff for multiple patients in the past. A depression level prediction device further comprising a learning unit that learns by using patient interview data and depression diagnosis results for patients who have had the same diagnosis result N or more times (N is an integer greater than or equal to 2) during a set period among all patients.

4. In claim 1, The above classification section is, A depression level prediction device that classifies the depression severity level of a target patient into one of the following labels: Moderate, Mild, and Normal.

5. In claim 1, The above classification section is, A depression level prediction device that applies natural language expression data, the frequency of which is counted in the above patient interview data, to a pre-learned depression level prediction model to classify the patient's depression severity level and whether it is normal.

6. In claim 1, A depression level prediction device further comprising a provision unit that analyzes natural language expression data of a target patient input into the depression level prediction model and provides, by artificial intelligence, the analysis results of which natural language expressions were used to derive the classification results of the depression level of the target patient.

7. In a method for predicting depression level performed by a depression level prediction device using natural language processing and artificial intelligence, A step of entering patient interview data containing the complaints of the target patient; A step of tokenizing the above patient interview data through natural language processing; A method for predicting depression levels, comprising a step of inputting tokenized natural language expression data into a pre-trained depression level prediction model to classify the depression severity level and normality of a target patient.

8. In claim 7, The above input receiving step is, A method for predicting the level of depression by inputting patient interview data from the CDW (Clinical Data Warehouse) containing natural language expressions recording the complaints of the above-mentioned patients.

9. In claim 7, The above depression classification model is, A method for predicting depression levels, which is learned using patient interview data and depression diagnosis results by medical staff for multiple patients in the past, and which is learned using patient interview data and depression diagnosis results for patients who have had the same diagnosis result N or more times (N is an integer greater than or equal to 2) during a set period among all patients.

10. In claim 7, A depression level prediction method further comprising a step of analyzing natural language expression data of a target patient input into the depression level prediction model and presenting, by artificial intelligence, the analysis results regarding which natural language expressions were used to derive the classification results of the depression level for the target patient.

Images