System and method for predicting cognitive function test score on basis of artificial intelligence

Abstract

The present invention relates to the field of medical and artificial intelligence technology, and specifically, to a system and method for predicting a cognitive function test score by extracting feature data from the result of a user performing the contents of cognitive response, spoken language, and eye tracking, and then inputting the feature data to an artificial intelligence model.

Description

AI-based cognitive function test score prediction system and method

[0001] The present invention relates to the fields of medicine and artificial intelligence technology, and specifically to a system and method for predicting cognitive function test scores by extracting feature data from the results of a user performing cognitive response, speech language, and eye-tracking content, and inputting the data into an artificial intelligence model.

[0002]

[0003] Recently, interest in technologies for testing the cognitive functions of the elderly has been increasing in an aging society. Cognitive function tests include the Korean Dementia Screening Questionnaire (KDSQ-C), the Cognitive Screening Test (CIST), and the Mini-Mental State Exam (MMSE).

[0004] The Mini-Mental State Examination (MMSE) is a test that evaluates a subject's cognitive function within approximately 5 to 10 minutes. While the MMSE has the advantage of being able to assess a subject's cognitive function simply and quickly, it has the problem of requiring evaluation by a skilled professional or specialist because it can be easily influenced by the subject's age, education level, and emotional state. Additionally, the existing MMSE classifies subjects dichotomously into a normal group and a cognitively impaired group based on scores, making it difficult to evaluate the subject's cognitive function as a continuous spectrum.

[0005] We will review the relevant prior art.

[0006] Patent Document 1 (Japanese Registered Patent Document No. 759960) relates to a mild cognitive impairment determination system and discloses a technology for predicting Mini-Mental State Examination (MMSE) scores through an artificial intelligence model, but does not disclose a method of having users perform cognitive responses, spoken language, and eye-tracking content, and extracting user features through this.

[0007] Patent Document 2 (Korean Registered Patent Document No. 10-2686017) relates to an AI-based method for diagnosing Alzheimer's disease and predicting the degree of cognitive impairment. Although the technology disclosed involves receiving acoustic features extracted from voice data to predict the Mini-Mental State Examination (MMSE) score, it does not disclose the performance of cognitive response and eye-tracking content, which makes it difficult to accurately predict the Mini-Mental State Examination (MMSE) score.

[0008] Patent Document 3 (Korean Registered Patent Document No. 10-2508262) relates to a dementia diagnosis device and a dementia information provision method using voice and eye tracking. Although a device for diagnosing dementia using voice and eye tracking is disclosed, recognition of predicting cognitive function test scores through an artificial intelligence model is not disclosed.

[0009]

[0010] (Prior Art Literature)

[0011] (Patent Document 1) Japanese Registered Patent Document No. 759960

[0012] (Patent Document 2) Korean Registered Patent Document No. 10-2686017

[0013] (Patent Document 3) Korean Registered Patent Document No. 10-2508262

[0014]

[0015] The present invention was devised to solve the above-mentioned problems.

[0016] The present invention aims to provide a cognitive function test score prediction system and method that can predict a cognitive response test score through an artificial intelligence model when a user performs cognitive response, speech language, and eye-tracking content through a terminal, even without performing a conventional cognitive function test.

[0017] In addition, the present invention aims to provide a cognitive function test score prediction system and method capable of evaluating a user's cognitive function as a continuous spectrum by predicting the cognitive function test score itself, unlike existing cognitive response tests that classify users into a normal group and a cognitive decline group based on test results.

[0018]

[0019] An artificial intelligence-based cognitive function test score prediction system according to an embodiment of the present invention for solving the above-mentioned problem comprises: a cognitive response module (100) that selects one or more cognitive response contents from among previously stored cognitive response contents and displays them on a user terminal (a); a voice language module (200) that selects one or more voice language contents from among previously stored voice language contents and displays them on the user terminal (a); a gaze tracking module (300) that selects one or more gaze tracking contents from among previously stored gaze tracking contents and displays them on the user terminal (a); a feature extraction module (400) that receives user response data corresponding to the cognitive response content, voice language content, and gaze tracking content displayed on the user terminal (a), respectively, from the user terminal (a) and calculates feature data therefrom; and a score prediction module (500) that inputs the feature data into a pre-trained cognitive function test score prediction model to predict a cognitive function test score.

[0020]

[0021] The cognitive response content stored in the above-mentioned cognitive response module (100) may include selecting a repeating number, selecting the same number of times, selecting a pair of pictures, selecting a larger calculation formula among the presented calculation formulas, and selecting one picture among the presented pictures and then selecting a picture that was not previously selected.

[0022]

[0023] The voice language content stored in the above voice language module (200) may include reading along with and memorizing a presented sentence, calculating a presented formula, and then remembering, speaking, and recording the presented sentence.

[0024]

[0025] The eye-tracking content stored in the eye-tracking module (300) may include gazing at a point displayed on the user terminal, quickly looking at a point displayed on the user terminal, and quickly looking in the opposite direction of a point displayed on the user terminal.

[0026]

[0027] The feature data calculated by the feature extraction module (400) from user response data corresponding to the cognitive response contents may include whether the answer is correct, the cumulative correct answer rate, the amount of change in the correct answer rate, the response time, and the amount of change in the response time.

[0028]

[0029] The feature data calculated by the feature extraction module (400) from user response data corresponding to the cognitive response contents may include a score that is higher when the response time is correct and the response time is short, and lower when the response time is incorrect and the response time is long, among the user response data corresponding to the cognitive response contents, by classifying the response time based on quartiles.

[0030]

[0031] The feature data calculated by the feature extraction module (400) from user response data corresponding to the voice language content may include speech time, number of pauses, pause time, number of syllables, number of words, word accuracy rate, similarity to the context and semantic content of the sentence, and grammatical accuracy rate.

[0032]

[0033] The feature data calculated by the feature extraction module (400) from user response data regarding the eye-tracking content may include the elapsed time, the coordinates of a point displayed on the user terminal, and the user's gaze coordinates.

[0034]

[0035] An artificial intelligence-based cognitive function test score prediction method according to an embodiment of the present invention for solving the above-mentioned problem may include: (a) a step in which a cognitive response module (100) selects one or more cognitive response contents from among previously stored cognitive response contents and displays them on a user terminal (a); (b) a step in which a speech language module (200) selects one or more speech language contents from among previously stored speech language contents and displays them on the user terminal (a); (c) a step in which an eye tracking module (300) selects one or more eye tracking contents from among previously stored eye tracking contents and displays them on the user terminal (a); (d) a step in which a feature extraction module (400) receives user response data corresponding to the cognitive response content, speech language content, and eye tracking content displayed on the user terminal (a), respectively, from the user terminal (a) and calculates feature data therefrom; and (e) a step in which a score prediction module (500) inputs the feature data into a pre-trained cognitive function test score prediction model to predict the cognitive function test score.

[0036]

[0037] An artificial intelligence-based cognitive function test score prediction system according to one embodiment of the present invention has the effect of predicting a cognitive function test score without a separate cognitive response test when a user performs cognitive response, speech language, and eye-tracking content displayed on a terminal.

[0038] An artificial intelligence-based cognitive function test score prediction system according to one embodiment of the present invention has the effect of conveniently and quickly predicting a user's cognitive response test score without evaluation by an expert or specialist.

[0039] An artificial intelligence-based cognitive function test score prediction system according to one embodiment of the present invention can verify a user's cognitive function as a continuous spectrum, thereby having the effect of evaluating the user's cognitive function in more detail.

[0040]

[0041] FIG. 1 is a schematic diagram showing the configuration of an artificial intelligence-based cognitive function score prediction system according to one embodiment of the present invention.

[0042] FIG. 2 is a data flow diagram of an artificial intelligence-based cognitive function score prediction system according to one embodiment of the present invention.

[0043] Figure 3 is a diagram illustrating that a score prediction module includes multiple cognitive function test score prediction models, and that some feature data is input to each cognitive function test score prediction model.

[0044] Figure 4 is a diagram showing participant information for an experiment to verify the performance of an artificial intelligence-based cognitive function test score prediction model.

[0045] Figures 5 to 8 illustrate that in an experiment to verify the performance of an artificial intelligence-based cognitive function test score prediction model, 98 features were selected out of a total of 974 features by considering the p-value and correlation coefficient.

[0046] FIGS. 9 to 12 are drawings showing the selection of optimal features through the RFE (Recursive feature elimination) method based on the ExtraTreesRegressor model and the XGBRegressor model, where FIGS. 9 and 10 show the features and feature importance selected by the ExtraTreesRegressor model, and FIGS. 11 and 12 show the features and feature importance selected by the XGBRegressor model.

[0047] Figure 13 is a figure showing the results of validating the performance of the ExtraTreesRegressor model, the XGBRegressor model, and the ensemble model on a validation set.

[0048] Figure 14 is a diagram showing the results of evaluating the performance of the ensemble model of Figure 13 on a test set.

[0049] FIG. 15 is a flowchart of an artificial intelligence-based cognitive function score prediction method according to one embodiment of the present invention.

[0050]

[0051] In some cases, to avoid obscuring the concept of the present invention, known structures and devices may be omitted or illustrated in the form of a block diagram focusing on the core functions of each structure and device.

[0052] Throughout the specification, when a part is described as "comprising" or "including" a component, this means that, unless specifically stated otherwise, it does not exclude other components but may include additional components. Furthermore, terms such as "part," "unit," and "module" as used in the specification refer to a unit that processes at least one function or operation, and this may be implemented in hardware, software, or a combination of hardware and software. Additionally, "one (a or an)," "one," "the," and similar related terms may be used in the context describing the invention (particularly in the context of the following claims) to include both singular and plural forms, unless otherwise indicated in the specification or clearly contradicted by the context.

[0053] In describing the embodiments of the present invention, specific descriptions of known functions or configurations will be omitted if it is determined that such detailed descriptions could unnecessarily obscure the essence of the invention. Furthermore, the terms described below are defined in consideration of their functions in the embodiments of the present invention, and these definitions may vary depending on the intentions or practices of the user or operator. Therefore, such definitions should be based on the content throughout this specification.

[0054]

[0055] In the following, the “cognitive function test” is preferably the Mini-Mental State Exam (MMSE), but is not limited thereto and may be modified and administered as various cognitive function tests such as the Korean Dementia Screening Questionnaire (KDSQ-C) and the Cognitive Screening Test (CIST).

[0056] In the following, “user response data” refers to data recording the user’s responses to the cognitive responses, spoken language, and eye-tracking content described below. The values ​​obtained vary for each piece of content, and detailed information will be provided below.

[0057] In the following, “feature data” refers to data reflecting user characteristics extracted from user response data. The values ​​observed vary for each piece of content, and further details will be discussed below.

[0058]

[0059] FIG. 1 is a schematic diagram showing the configuration of an artificial intelligence-based cognitive function score prediction system according to one embodiment of the present invention, and FIG. 2 is a data flow diagram of an artificial intelligence-based cognitive function score prediction system according to one embodiment of the present invention.

[0060] Referring to FIGS. 1 and 2, an artificial intelligence-based cognitive function test score prediction system (1) according to one embodiment of the present invention may include a cognitive response module (100), a speech language module (200), an eye tracking module (300), a feature extraction module (400), and a score prediction module (500).

[0061] The cognitive response module (100) is configured to store cognitive response content. The cognitive response module (100) can select one or more of the previously stored cognitive response content and display them on the user terminal (a).

[0062] Cognitive response content may consist of selecting repeating numbers, selecting the number of identical numbers, selecting paired pictures, selecting paired pictures, selecting the larger equation among presented equations, and selecting one picture from presented pictures and choosing a picture that was not previously selected. For example, the repeating number selection content may be content where the user selects a repeating number from the numbers presented to them. If the presented number is “111,” the user may be asked to choose one of the options '1, 2, 3, 4'. As another example, the content for selecting one picture from presented pictures and choosing a picture that was not previously selected may present six pictures to the user, have them select one, swap the positions of the six pictures, and then have them select the picture that was not previously selected. However, cognitive response content is not limited to this and can be implemented by modifying it into various contents that can evaluate cognitive function using the user's cognitive response.

[0063]

[0064] The voice language module (200) is configured to store voice language content. The voice language module (200) can select one or more voice language contents from the previously stored voice language contents and display them on the user terminal (a).

[0065] The voice language content may involve reading along to memorize a presented sentence, calculating a presented formula, and then recalling, speaking, and recording the previously presented sentence. For example, the sentence “Young-hee went to school with a friend” may be displayed on the user’s device screen, and the user may be asked to read it aloud. Next, a calculation problem may be presented. Once the user completes the calculation problem, the message “Please say out loud the sentence you just remembered” may be displayed on the user’s device, prompting the user to recall, speak, and record the memorized sentence. However, the voice language content is not limited to this and can be modified and implemented as various content capable of evaluating cognitive functions using the user’s voice.

[0066]

[0067] The eye tracking module (300) is configured to store eye tracking content. The eye tracking module (300) can select one or more of the previously stored eye tracking content and display them on the user terminal (a).

[0068] Eye-tracking content may include content for gazing at a dot displayed on a user terminal, quickly looking at a dot displayed on a user terminal, and quickly looking in the opposite direction of a dot displayed on a user terminal. For example, quickly looking in the opposite direction of a dot displayed on a user terminal may involve displaying a red dot on the terminal screen and a message stating, "Please look in the opposite direction of the red dot appearing on the screen," thereby instructing the user to quickly look in the opposite direction of the red dot. However, eye-tracking content is not limited to this and can be implemented by modifying it into various types of content capable of evaluating cognitive functions by tracking the user's gaze.

[0069]

[0070] The feature extraction module (400) receives user response data corresponding to cognitive response content, voice language content, and eye tracking content displayed on the user terminal (a) from the user terminal (a) and can compute feature data therefrom. In the example described above, when the user performs repeated number selection, the user response data is the number selected by the user from the options '1, 2, 3, 4', and the feature data may be whether the selected number is correct, etc.

[0071]

[0072] The feature extraction module (400) can calculate whether the answer is correct, the cumulative correct answer rate, the change in the correct answer rate, the reaction time, and the change in the reaction time from user response data corresponding to cognitive response content. Additionally, the feature extraction module (400) can assign weights by reflecting the correct answer and reaction time of the cognitive response content. Specifically, the feature extraction module (400) can classify the reaction time based on quartiles, and among the user response data, the score can be measured higher for correct answers with shorter reaction times and lower for incorrect answers with longer reaction times. For example, first, the reaction time can be classified into grades 1, 2, and 3 based on quartiles, and then, in the case of a correct answer, converted to 1, 2, or 3 points according to the reaction time, and in the case of an incorrect answer, converted to -3, -2, or -1 points according to the reaction time and stored as feature data.

[0073]

[0074] Additionally, the feature extraction module (400) can calculate the utterance time, number of pauses, pause time, number of syllables, number of words, word accuracy rate, similarity to the context and semantic content of the sentence, and grammatical accuracy rate from user response data corresponding to the voice language content.

[0075] The similarity regarding the context and semantic content of sentences can be calculated by determining the embedding vectors for the correct answer sentence and the user utterance sentence, respectively, and calculating the cosine similarity between the embedding vectors. Additionally, the grammatical accuracy rate can be calculated by extracting the word corresponding to the correct answer from the user utterance sentence, determining the position of the corresponding word in the correct answer sentence and the position of the corresponding word in the utterance sentence, and then calculating the distance between the two positions.

[0076]

[0077] Additionally, the feature extraction module (400) can calculate the elapsed time, the coordinates of a point displayed on the user terminal, and the user's gaze coordinates from user response data corresponding to the eye-tracking content. The feature extraction module (400) can calculate the movement speed of a point using the coordinates of a point displayed on the user terminal and calculate the gaze speed using the user's gaze coordinates. The first value is the value obtained by dividing the movement speed of a point by the gaze speed, and the feature extraction module (400) can calculate the average of the first value from the user response data and store it as feature data.

[0078]

[0079] The score prediction module (500) is configured to predict cognitive function test scores through an artificial intelligence model. The score prediction module (500) may include a cognitive function test score prediction model that uses feature data as training data and cognitive function test scores as output data. Therefore, if the feature data generated from the feature extraction module (400) is input into the cognitive function test score prediction model, the cognitive function test score can be predicted. The method for generating the cognitive function test prediction model will be described later.

[0080]

[0081] Figure 3 is a diagram illustrating that a score prediction module includes multiple cognitive function test score prediction models, and that some feature data is input to each cognitive function test score prediction model.

[0082] Referring to FIG. 3, the cognitive function test score prediction model may be composed of multiple artificial intelligence models. For example, the first cognitive function test score prediction model may be composed of an ExtraTreesRegressor model, and the second cognitive function test score prediction model may be composed of an XGBoostRegressor model. Additionally, the cognitive function test score prediction model may be an ensemble model of the first and second cognitive function test score prediction models.

[0083] To reduce the dimensionality of cognitive responses and improve data processing efficiency, only a subset of feature data can be selected by considering p-values ​​and correlation coefficients. Some of the selected feature data can be input into a first artificial intelligence model, and the remaining selected feature data can be input into a second artificial intelligence model.

[0084]

[0085] Hereinafter, with reference to FIGS. 4 to 14, a method for generating a cognitive function test score prediction model and the experimental results thereof will be explained.

[0086] To generate a cognitive function test score prediction model, the inventors tested 220 participants using the cognitive response, speech language, and eye-tracking content presented in this invention and computed feature data. Of the 220 participants, 154 were in the normal group and 66 were in the cognitive impairment group. Among these, data with 30% or more missing values ​​among the total features were removed, and analysis was performed on 220 data points. Participant information is shown in Figure 4.

[0087] To prevent data leakage during the variable and algorithm selection process, modeling and validation were conducted by distinguishing between a training set for learning the model, a validation set for verifying it, and a test set for evaluating final performance. To handle missing values, the feature-unit average values ​​from the dataset excluding the test set were used.

[0088] An ensemble model of the ExtraTreesRegressor and XGBoostRegressor models was used as the machine learning model. By combining two models with different subset selection and training methods, the variability of predictions and the risk of overfitting were reduced compared to a single model. For each model, optimal feature configurations and model parameters were configured through feature selection and hyperparameter tuning.

[0089]

[0090] Feature selection algorithm

[0091] Out of a total of 974 features, 791 features were selected by removing one of the pairs if the correlation coefficient of the variable pairs was greater than 0.9.

[0092] To reduce dimensionality and improve data processing efficiency, features with low correlation to the target variable were removed from 791 features. For continuous variables, the Spearman correlation coefficient was calculated, and for binary variables, the point biserial correlation coefficient was calculated, and the top 98 variables with a p-value of less than 0.05 and high correlation coefficients were selected. Referring to Figures 5 through 8, it was confirmed that 78 of the 98 variables were extracted through feature extraction, and that significant features were extracted.

[0093]

[0094] Model optimization algorithm

[0095] Optimal features were selected through recursive feature elimination based on the ExtraTreesRegressor and XGBRegressor models. Referring to Figures 9 through 12, 46 features were finally used for the ExtraTreesRegressor and 51 features for the XGBoostRegressor.

[0096]

[0097] Model optimization algorithm

[0098] The models were optimized through hyperparameter tuning using the feature sets of the ExtraTreesRegressor and XGBRegressor models, respectively. To determine the model weights, a random search-based exploration was performed on the validation set. The optimal weights (ratios) were set to 0.64 for ExtraTreesRegressor and 0.36 for XGBoostRegressor. The results of verifying the performance of each model and the ensemble model on the validation set are shown in Fig. 13, and the results of the final evaluation on the test set are shown in Fig. 14. The mean absolute error (MAE) of the K-MMSE prediction of the model was 1.09, the root mean square error (RMSE) was 1.39, and the correlation coefficient was 0.75. Therefore, it can be confirmed that the artificial intelligence-based cognitive function test score prediction model according to one embodiment of the present invention predicts the Mini-Mental State Examination score with high accuracy.

[0099]

[0100] FIG. 15 is a flowchart of an artificial intelligence-based cognitive function test score prediction method according to one embodiment of the present invention.

[0101] Hereinafter, with reference to FIG. 15, an artificial intelligence-based method for predicting cognitive function test scores according to an embodiment of the present invention is described.

[0102] First, the cognitive response module (100) may select one or more of the previously stored cognitive response contents and display them on the user terminal (a). The cognitive response contents may include selecting a repeating number, selecting the same number count, selecting a pair of pictures, selecting a pair of pictures, selecting a larger calculation formula among the presented calculation formulas, and selecting one picture among the presented pictures and selecting a picture that has not been previously selected.

[0103] Next, the voice language module (200) may select one or more voice language contents from among the previously stored voice language contents and display them on the user terminal (a). The voice language contents may include reading along with and memorizing a presented sentence, calculating a presented formula, and then remembering, speaking, and recording a previously presented sentence.

[0104] Next, the eye tracking module (300) may select one or more of the previously stored eye tracking contents and display them on the user terminal (a). The eye tracking contents may include gazing at a point displayed on the user terminal, quickly looking at a point displayed on the user terminal, and quickly looking in the opposite direction of a point displayed on the user terminal.

[0105] Next, the feature extraction module (400) receives user response data corresponding to cognitive response content, voice language content, and eye tracking content displayed on the user terminal (a) from the user terminal (a) and can calculate feature data therefrom.

[0106] Next, the score prediction module (500) can predict the cognitive function test score by inputting the feature data into a pre-trained cognitive function test score prediction model. The cognitive function test score prediction model may be a plurality of artificial intelligence models or an ensemble model.

[0107]

[0108] For the time being, the present specification has been described with reference to embodiments illustrated in the drawings so that those skilled in the art can easily understand and reproduce the present invention; however, this is merely illustrative, and those skilled in the art will understand that various modifications and equivalent alternative embodiments are possible from the embodiments of the present invention. Accordingly, the scope of protection of the present invention should be determined by the claims.

[0109]

[0110] (Explanation of symbols)

[0111] 1 : AI-based cognitive function score prediction system

[0112] 100: Cognitive Response Module

[0113] 200: Speech Language Module

[0114] 300: Eye Tracking Module

[0115] 400: Feature extraction module

[0116] 500 : Score Prediction Module

[0117] a : User terminal

[0118]

Claims

1. A cognitive response module (100) that selects one or more of the previously stored cognitive response contents and displays them on a user terminal (a); A voice language module (200) that selects one or more voice language contents from among the previously stored voice language contents and displays them on the user terminal (a); An eye tracking module (300) that selects one or more of the previously stored eye tracking contents and displays them on the user terminal (a); A feature extraction module (400) that receives user response data corresponding to cognitive response content, voice language content, and eye tracking content displayed on the user terminal (a) from the user terminal (a), and calculates feature data therefrom; and A score prediction module (500) that predicts a cognitive function test score by inputting the above feature data into a pre-trained cognitive function test score prediction model, comprising AI-based cognitive function test score prediction system.

2. In Paragraph 1, The cognitive response content stored in the above cognitive response module (100) is, Selecting repeating numbers, selecting the same number of times, selecting paired pictures, selecting the larger formula among the presented formulas, and selecting one picture among the presented pictures and then selecting a picture that was not previously selected, AI-based cognitive function test score prediction system.

3. In Paragraph 2, The voice language content stored in the above voice language module (200) is, Includes reading along with and memorizing the presented sentences, calculating the presented formulas, and then speaking and recording while remembering the above-mentioned sentences. AI-based cognitive function test score prediction system.

4. In Paragraph 3, The eye-tracking content previously stored in the eye-tracking module (300) is, including staring at a point displayed on the user terminal, quickly looking at the point displayed on the user terminal, and quickly looking in the opposite direction of the point displayed on the user terminal. AI-based cognitive function test score prediction system.

5. In Paragraph 4, The feature data calculated by the feature extraction module (400) from the user response data corresponding to the cognitive response content includes whether the answer is correct, the cumulative answer rate, the change in the answer rate, the response time, and the change in the response time. AI-based cognitive function test score prediction system.

6. In Paragraph 5, The feature data calculated by the feature extraction module (400) from user response data corresponding to the cognitive response contents is classified based on quartiles for the response time, and includes a score that is higher when the response time is short and the answer is incorrect and the response time is long among the user response data corresponding to the cognitive response contents. AI-based cognitive function test score prediction system.

7. In Paragraph 6, The feature data calculated by the feature extraction module (400) from user response data corresponding to the voice language content includes speech time, number of pauses, pause time, number of syllables, number of words, word accuracy rate, similarity to the context and semantic content of the sentence, and grammatical accuracy rate. AI-based cognitive function test score prediction system.

8. In Paragraph 7, The feature data calculated by the feature extraction module (400) from the user response data for the eye-tracking content includes the time taken, the coordinates of a point displayed on the user terminal, and the user's gaze coordinates. AI-based cognitive function test score prediction system.

9. (a) A step in which the cognitive response module (100) selects one or more of the previously stored cognitive response contents and displays them on the user terminal (a); (b) A step in which the voice language module (200) selects one or more voice language contents from among the previously stored voice language contents and displays them on the user terminal (a); (c) A step in which the eye tracking module (300) selects one or more of the previously stored eye tracking contents and displays them on the user terminal (a); (d) A feature extraction module (400) receives user response data corresponding to cognitive response content, voice language content, and eye tracking content displayed on the user terminal (a) from the user terminal (a), and calculates feature data therefrom; and (e) a step in which the score prediction module (500) inputs the feature data into a pre-trained cognitive function test score prediction model to predict the cognitive function test score; comprising, AI-based cognitive function test score prediction method.

Images