KR102431205B1 - Apparatus for generating training data and System for symptom diagnosis to which the Artificial Intelligence data training is applied - Google Patents

Abstract

Disclosed are an apparatus for generating training data to predict mental illnesses and symptoms and a symptom diagnosis system and method to which training data is applied. According to one embodiment of the present invention, the apparatus for generating training data may include: a data receiving unit receiving biological data from a wearable device worn by a user; a data analysis unit extracting at least one feature value from the biological data; and a learning data generation unit generating training data by performing labeling on analysis data generated based on the at least one feature value.

Description

Apparatus for generating training data and System for symptom diagnosis to which the Artificial Intelligence data training is applied}

It relates to a learning data generating device for predicting mental disorders and symptoms through training of machine learning and deep learning models based on biometric data collected from a user's wearable device, and a symptom diagnosis system to which artificial intelligence data learning is applied, and a method therefor .

Conventionally, diagnosis of children's mental disorders and symptoms was performed through consultation with a specialist based on biometric data measured through an invasive device such as polysomnography (PSG).

Recently, basic biometric data of a user is collected in a non-invasive manner through a wearable device, and various diagnostic methods using the collected data are being studied. In the case of Patent Publication No. 10-2018-0029517, a method of using a wearable device for diagnosis of attention deficit disorder is disclosed.

An object of the present invention is to provide an apparatus for generating learning data for predicting mental illness and symptoms, and a system and method for diagnosing symptoms to which learning data is applied.

According to an aspect, an apparatus for generating learning data includes a data receiving unit configured to receive biometric data from a wearable device worn by a user; a data analysis unit for extracting one or more feature values from biometric data; and a training data generator configured to generate training data by labeling the analysis data generated based on one or more feature values.

The biometric data may include data on at least one of a heart rate, a step number, a metabolic rate (METs, Metabolic rates), a calorie consumption, and a sleep state.

The data analyzer may generate an estimation function y(t) by performing a cosinor analysis based on data on at least one of a heart rate and a number of steps included in the biometric data.

The data analyzer may calculate a good of fit (GOF) based on the estimation function.

The learning data generating unit may compare a check list including feature value information for each predetermined symptom with a feature value included in the analysis data to determine whether the analysis data is symptom data and perform labeling.

The training data generator may classify the training data into training data, verification data, and test data according to a predetermined ratio.

According to an aspect, a method for generating learning data includes: receiving biometric data from a wearable device worn by a user; a data analysis step of extracting one or more feature values from biometric data; and performing labeling on the analysis data generated based on one or more feature values to generate training data.

The biometric data may include data on at least one of a heart rate, a step number, a metabolic rate (METs, Metabolic rates), a calorie consumption, and a sleep state.

In the data analysis step, the estimation function y(t) may be generated by performing a cosinor analysis based on data on at least one of a heart rate and a number of steps included in the biometric data.

In the data analysis step, a good of fit (GOF) may be calculated based on the estimation function.

The learning data generation step may perform labeling by comparing a check list including feature value information for each predetermined symptom with a feature value included in the analysis data to determine whether the analysis data is data about a symptom.

The training data generation step may classify the training data into training data, verification data, and test data according to a predetermined ratio.

According to an aspect, the symptom diagnosis system may include a symptom diagnosis device and a wearable device.

According to one aspect, the symptom diagnosis apparatus may include: a data receiver configured to receive biometric data from a wearable device worn by a user; a data analysis unit for extracting one or more feature values from biometric data; and a diagnostic unit that analyzes a feature value using an analysis model learned from the learning data labeled with respect to a predetermined symptom to determine the presence or absence of a predetermined symptom.

The analysis model may be a machine learning model or a deep learning model.

The biometric data may include data on at least one of a heart rate, a step count, a metabolic rate (METs, Metabolic rates), a calorie consumption, and a sleep state.

According to one aspect, a symptom diagnosis method includes: receiving biometric data from a wearable device worn by a user; a data analysis step of extracting one or more feature values from biometric data; and a diagnosis step of determining the presence or absence of a predetermined symptom by analyzing a feature value using an analysis model learned with the learning data labeled with respect to the predetermined symptom.

Through the circadian data collected from the user's wearable device, it is possible to diagnose and evaluate children's mental disorders and symptoms early.

1 is a block diagram of an apparatus for generating learning data according to an exemplary embodiment.
2 is a flowchart illustrating a method of generating learning data according to an embodiment.
3 is a block diagram of an apparatus for diagnosing symptoms according to an exemplary embodiment.
4 is an exemplary diagram of analyzing the performance of an analysis model applied to an apparatus for generating training data in an example.
5 is a flowchart illustrating a symptom diagnosis method according to an exemplary embodiment.

Hereinafter, an embodiment of the present invention will be described in detail with reference to the accompanying drawings. In describing the present invention, if it is determined that a detailed description of a related well-known function or configuration may unnecessarily obscure the gist of the present invention, the detailed description thereof will be omitted. In addition, the terms to be described later are terms defined in consideration of functions in the present invention, which may vary according to intentions or customs of users and operators. Therefore, the definition should be made based on the content throughout this specification.

Hereinafter, embodiments of the learning data generating apparatus and the symptom diagnosis system to which the learning data is applied will be described in detail with reference to the drawings.

1 is a block diagram of an apparatus for generating learning data according to an exemplary embodiment.

According to an embodiment, the learning data generating apparatus 100 includes a data receiving unit 110 receiving biometric data from a wearable device worn by a user, a data analyzing unit 120 extracting one or more feature values from the biometric data, and one It may include a learning data generating unit 130 that generates learning data by performing labeling on the analysis data generated based on the above feature values.

According to an example, the data receiving unit 110 may receive the user's biometric data from the wearable device or the user terminal worn by the user in units of 30 seconds to 1 minute. As an example, the biometric data may include at least one of a heart rate, a step count, metabolic rates (METs), a calorie consumption, and a sleep state.

According to an example, the data analysis unit 120 may perform cosinor analysis, circadian information extraction, and data loading. If the data analyzer 120 can extract a predetermined feature point by performing a circadian analysis on the biometric data received through the data receiver 110 , it can generate a data set for each user based on the feature point.

According to an example, the data analysis unit 120 may verify the validity of the biometric data received through the data receiving unit 110 , and calculate the time at which each biometric data was recorded year, month, day, hour, minute, It can be stored separately in seconds.

According to an example, the data analysis unit 120 may perform three major classifications of biometric data based on heart rate, sleep information, and activity information on the validated biometric data file.

For example, the data analyzer 120 may calculate the minimum, maximum, and arithmetic mean per hour with respect to the heart rate.

For example, the data analyzer 120 may perform two intermediate classifications based on 30 seconds and 60 seconds for sleep information. For example, the data analysis unit 120 may calculate the duration (minutes) of 4 sleep recordings of light, deep, rem, and wake per hour for 30 seconds of sleep. In addition, the data analysis unit 120 may calculate the duration (minutes) or the number of occurrences of the three-stage sleep recording of asleep, restless, and awake per hour for 60 seconds of sleep.

As an example, the data analysis unit 120 calculates the minimum, maximum, and arithmetic mean of body intensity measurements (Intensity, METs) per hour for activity information, the sum of steps per hour, and the sum of calories consumed per hour. can be calculated

According to an example, when performing cosinor analysis, the data analysis unit 120 may use data for a predetermined time (eg, 30 minutes) or more per hour in order to check the validity of the biometric data of one hour period. For example, the data analyzer 120 may determine and use the biometric data of a user who has data for a specific time (eg, 5 hours) or more per day as valid in order to check the validity of the biometric data.

According to an example, the data analyzer 120 may perform cosinor analysis based on data on at least one of a heart rate and a number of steps included in the biometric data. For example, if the data analysis unit 120 can use biometric data of one hour period to perform cosinor analysis, it calculates MESOR (Midline Statistic Of Rhythm), Amplitude, Acrophase, and Good of fit and converts it into circadian data. can be returned

According to an embodiment, the data analysis unit 120 may perform cosinor analysis to generate an estimation function y(t) as shown in the following Equation.

Here, M is the midline estimating statistic of rhythm (MESOR), is the amplitude, is the peak time (Acrophase), is the time, T is the period. For example, the period T may be 24 hours. For example, the data analysis unit 120 may estimate y(t) by continuously listing data for a predetermined period (eg, 3 days) or longer.

According to an embodiment, the data analyzer 120 may calculate a good of fit (GOF) as shown in the following equation based on the estimation function.

는 Explained Sum of Squares를 의미하며,

와 같이 정의된다. 여기서,

과 같이 정의되면, y는 사용자의 일주기 데이터,

은 추정함수 y(t)로부터 추정된 예측값을 의미한다. here,

stands for Explained Sum of Squares,

is defined as here,

If defined as, y is the user's circadian data,

is a predicted value estimated from the estimation function y(t).

는 Total Sum of Squares를 의미하며,

와 같이 정의된다. 여기서,

은

과 같이 정의된다. For example,

stands for Total Sum of Squares,

is defined as here,

silver

is defined as

As an example, Equation 2 may be calculated as follows.

According to an example, the data analysis unit 120 calculates L5 (Least 5 hours active), M10 (Most 10 hours active), and RA (Relative Amplitude) indicators based on at least one of the user's heart rate and step count in units of one cycle. Alternatively, it can be calculated in units proportional to the entire data collection period. For example, L5 represents calculating an arithmetic mean by applying a sliding window frame by 5 hours in 1-hour period data, and calculating the minimum of this value in 1-hour period data. M10 means calculating an arithmetic mean by applying a sliding window frame by 10 hours in 1-hour period data, and calculating the maximum of this value in 1-hour period data.

According to an example, the data analyzer 120 may calculate intradaily stability (IS) and intradaily variability (IV) as shown in the following equation based on the number of steps of the user.

는 1시간 당 생체 데이터의 평균,

는 모든 생체 데이터 지점들의 평균,

는 현재 연산 중인 1시간 데이터의 인덱스,

는 하루에 존재하는 생체 데이터 지점들의 개수를 의미한다. here,

is the average of biometric data per hour,

is the average of all biometric data points,

is the index of the one-hour data currently being calculated,

denotes the number of biometric data points existing per day.

According to an example, the data analyzer 120 may calculate the basal metabolic rate by using the user's basic body information, and then calculate the difference in the circadian metabolic rate by adding or subtracting it from the daily calorie consumption. For example, the basal metabolic rate may be calculated using the following three formulas.

- Mifflin St Jeor's Equation

은 Basal Metabolic Rate이며,

는 남성의 경우 +5, 여성의 경우 -161을 할당한다. here,

is the Basal Metabolic Rate,

is assigned +5 for men and -161 for women.

- Katch-McArdle's Equation

- Harris-Benedict's Equation

According to an example, the data analyzer 120 may count the number of steps taken during the user's sleep time, and may calculate the minimum, maximum, and arithmetic mean with respect to the heart rate. For example, the sleep time may be set to 9 pm to 6 pm, sunset time to sunrise time, and the like.

According to an example, the data analysis unit 120 may count the number of steps in the user's daily time, and may calculate the minimum, maximum, and arithmetic mean with respect to the heart rate. For example, the daily time may be set to 7:00 am to 8:00 pm, sunrise time to sunset time, and the like.

According to an example, the data analyzer 120 may classify the intensity of the user's physical activity into four levels of Light, Sedentary, Moderate, and Vigorous. For example, each step represents a case in which the values of METs are in the range of <1.5, 1.5-3.0, 3.0-6.0, and >6.0, respectively, and can be continuously calculated in units of one cycle. In addition, the data analysis unit 120 may calculate and record the total duration of each step in a day in the biometric data in units of minutes.

According to an example, the data analyzer 120 may calculate the standard deviation and variance of the user's circadian heart rate. The data analysis unit 120 may calculate the total time that the user was in the sleep state in the daily time from the 30-second sleep record, and may calculate the duration of the user's four sleep phases in units of one cycle. In addition, the data analysis unit 120 may calculate the amount of time the user was in bed in a unit of one cycle as shown in the following equation.

According to an example, the data analysis unit 120 may calculate circadian sleep efficiency for each of 30 seconds sleep data and 60 seconds sleep data in the user's sleep record, as shown in the following equation.

According to an example, the data analysis unit 120 may calculate the ratio of each sleep in the 4th stage of the user's circadian cycle from the 30-second sleep record. As an example, the sleep rate may be calculated as shown in the following equation.

According to an embodiment, the learning data generation unit 130 determines whether the analysis data is data about symptoms by comparing the check list including the characteristic value information for each predetermined symptom with the characteristic values included in the analysis data to perform labeling. can be done

According to an example, the learning data generating unit 130 provides a user to the analysis data based on the K-SADS-PL (Kiddie Schedule for Affective Disorders and Schizophrenia Present and Lifetime version) DSM5 (Diagnostic and Statistical Manual of Mental Disorders) diagnostic criteria. A label can be given including whether or not the person has been diagnosed with a mental illness and has symptoms.

As an example, the mental illness or symptom may include Attention Defiant/Hyperactivity Disorder (ADHD), Obsessive-Compulsive Disorder (OCD), Oppositional Defiant Disorder (ODD), and MDD (Major Depressive Disorder) diagnosis and sleep problem symptoms. may include

According to an example, the learning data generating unit 130 may receive the K-SADS-PL based checklist data generated by the expert, and perform labeling on the analysis data based on the received checklist. For example, the learning data generating unit 130 may calculate the similarity by comparing the feature value included in the check list with the feature value included in the analysis data, and the analysis data is displayed according to whether the similarity is greater than or equal to a predetermined standard. It can be determined whether the data for

As an example, the learning data generator 130 may assign a label 1 to the analysis data having a disease and symptom to be diagnosed, and a label 0 otherwise.

For example, the learning data generating unit 130 may match the ratio between the normal group and the disease group. For example, the ratio of the normal group to the disease group may be set as 1:1 or 2:1. The training data generator 130 may use a random sampling or a synthetic minority oversampling technique (SMOTE) sampling technique to readjust the data rate.

According to an example, the training data generator 130 may perform scaling on the training data set that has undergone a sampling process. For example, the training data generator 130 uses a minimum-maximum scaling technique that converts all data into values between 0 and 1, a standard scaling technique that converts all data into values between -1 and 1, and the like. The training data can be scaled.

According to an embodiment, the training data generator 130 may classify the training data into training data, verification data, and test data according to a predetermined ratio. For example, the learning data generator 130 may adjust the training:verification:test ratio to 6:2:2, 5:3:2, 7:2:1, and the like.

2 is a flowchart illustrating a method of generating learning data according to an embodiment.

According to an embodiment, the apparatus for generating learning data may receive biometric data from a wearable device worn by the user ( S210 ).

According to an example, the learning data generating apparatus may receive the user's biometric data from the wearable device or the user terminal worn by the user in units of 30 seconds to 1 minute. As an example, the biometric data may include at least one of a heart rate, a step count, metabolic rates (METs), a calorie consumption, and a sleep state.

According to an embodiment, the apparatus for generating learning data may extract one or more feature values from the biometric data ( 220 ).

According to an example, the apparatus for generating learning data may generate an estimation function y(t) by performing a cosinor analysis based on data on at least one of a heart rate and a number of steps included in the biometric data. Also, the learning data generating apparatus may calculate a good of fit (GOF) based on the estimation function.

According to an embodiment, the apparatus for generating training data may generate training data by labeling the analysis data generated based on one or more feature values ( 230 ).

According to an example, the apparatus for generating learning data may compare a check list including feature value information for each predetermined symptom with a feature value included in the analysis data to determine whether the analysis data is data about symptoms and perform labeling. . Also, the training data generating apparatus may classify the training data into training data, verification data, and test data according to a predetermined ratio.

3 is a block diagram of an apparatus for diagnosing symptoms according to an exemplary embodiment.

The symptom diagnosis system may include a symptom diagnosis apparatus 300 and a wearable device (not shown).

According to an embodiment, the symptom diagnosis apparatus 300 includes a data receiver 310 that receives biometric data from a wearable device worn by a user, a data analyzer 320 that extracts one or more feature values from biometric data, and a predetermined symptom. The diagnostic unit 330 may include a diagnosis unit 330 for determining the presence or absence of a predetermined symptom by analyzing a feature value using an analysis model learned with the learning data labeled for .

According to an example, the analysis model may predict a child's mental illness and symptoms using machine learning and deep learning techniques based on the learning data. As an example, the machine learning technique may use a random forest, eXtreme Gradient Boosting (XGBoost), or Light Gradient Boosting Machine (LightGBM) classifier. As an example, the deep learning model may use Long-Short Term Memory (LSTM).

According to an example, in the case of a random forest classifier model, the number of repetitions of training may be set to 50 or less, n_estimator may be set to 2000 or less, max_depth may be set to 50 or less, min_samples_leaf may be set to 20 or less, and min_samples_split may be set to 50 or less.

According to an example, for the XGBoost classifier model, the number of iterations of training is 10 or less, the objective is "binary:logistic", the booster is "gbtree" and "dart", the eval_metric is "auc" and "logloss", and the maximum depth is can be set to 20 or less.

According to an example, in the case of the LightGBM classifier model, the number of iterations of training may be set to 10 or less, the objective may be set to "binary", the boosting_type to be "gbdt", the metric may be set to "auc", and max_depth may be set to 20 or less.

According to an example, in the case of the LSTM model, the training data used may be grouped by grouping the user's circadian data into a specific number of days range of 3 or more days. For example, in order to determine whether to diagnose a user's mental illness and symptoms, 14 consecutive days of circadian data may be grouped and used as an input to the model. The number of cells in the LSTM model may be set to a specified number of days, and may be configured in a many-to-one or many-to-many structure. The output of each cell may be output as a value representing the corresponding day using a global average pooling technique, or global maximum pooling may be used. In addition, all outputs of each cell are transmitted to a fully connected layer to estimate a diagnosis probability of a user's mental illness and symptoms. Here, a neural network may exist between the output of each cell and the final prediction layer.

According to an example, the analysis model may be evaluated by the following six metrics. For example, the six performance metrics include ROC/AUC, Accuracy, Sensitivity, Specificity, F1 Score, and G-Mean. Hold not used for training during performance evaluation. -Out test samples can be used to evaluate the performance of the model. For example, performance may be evaluated using test data generated by the above learning data generating device. The index for performance evaluation may be defined as follows.

-

-

-

-

-

-

Here, Receiver Operating Characteristic (ROC) indicates receiver operation characteristics, True Positive Rate (TPR) has the same meaning as Sensitivity, and False Positive Rate (FPR) has the same meaning as (1-specificity). AUC (Area Under Curve) refers to the area under the graph curve when the ROC value is displayed in graph form, and is generally used as an indicator of the performance of machine learning and deep learning models.

Also, in 15, TP stands for True Positive, TN stands for True Negative, FP stands for False Positive, and FN stands for False Negative. TP is when the actual label of the data is positive and the prediction of the model is also positive, TN is when the actual label of the data is positive but the prediction of the model is negative, FP is when the actual label of the data is negative but the prediction of the model is positive, FN represents the case where the actual label of the data is negative and the prediction of the model is also negative.

Sensitivity can be used in the same sense as recall, and when the sensitivity is high, it can be seen that the model correctly predicted the actual answer. For example, if it is necessary to predict a diagnosis of only ADHD for a certain patient A, a model with high ADHD sensitivity can be selected.

로 쓰일 수 있다. 여기서 F1 Score는 불균형한 데이터 분포를 갖는 상황에서 모델의 성능을 정확하게 하나의 값으로 평가하기 위한 지표일 수 있다. F1 Score is a numerical value representing the harmonic average of precision and recall, and the constant "2" used here depends on the label distribution.

can be used as Here, the F1 Score may be an index for accurately evaluating the performance of a model as a single value in a situation with an unbalanced data distribution.

4 is an analysis of the performance of the analysis model.

Figure 4 (a) is a table showing the size of the training data used in the three models for ADHD, the model performance of the entire sampled training data set, and the performance of the model for the hold-out data sample. As the performance of each model is a correct performance measurement method to evaluate the value for the hold-out data sample that did not participate in model training, as shown in Fig. 4(a), in the case of ADHD, ROC/AUC 0.801, accuracy 73.3 %(0.733), sensitivity 0.754, specificity 0.714, F1 Score 0.724, G-Mean 0.734, it can be seen that the performance of the LightGBM model is the best.

Figure 4(b) is a table showing the training data size used for the three models for OCD, the model performance of the entire sampled training dataset, and the model performance on the hold-out data sample. 4(b), in the case of OCD, the LightGBM model with ROC/AUC 0.799, accuracy 75.2% (0.752), sensitivity 0.785, specificity 0.707, F1 Score 0.785, and G-Mean 0.745 showed the best performance. can be checked

FIG. 4(c) is a table showing the size of the training data used in the three models for ODD, the model performance of the entire sampled training data set, and the performance of the model for the hold-out data sample. 4(c), in the case of ODD, the XGBoost model with ROC/AUC 0.844, accuracy 73.8% (0.738), sensitivity 0.769, specificity 0.711, F1 Score 0.733, and G-Mean 0.740 showed the best performance. can be checked

The reason why the data size of the random forest model in FIG. 4 is relatively small compared to the other two models is that data with missing values are discarded because the random forest model does not allow missing data.

5 is a flowchart illustrating a symptom diagnosis method according to an exemplary embodiment.

According to an embodiment, the symptom diagnosis apparatus may receive biometric data from a wearable device worn by the user ( S510 ).

According to an example, the symptom diagnosis apparatus may receive the user's biometric data from the wearable device or the user terminal worn by the user in units of 30 seconds to 1 minute. As an example, the biometric data may include at least one of a heart rate, a step count, metabolic rates (METs), a calorie consumption, and a sleep state.

According to an embodiment, the symptom diagnosis apparatus may extract one or more feature values from the biometric data ( 520 ).

According to an example, the symptom diagnosis apparatus may generate an estimation function y(t) by performing a cosinor analysis based on data on at least one of a heart rate and a number of steps included in the biometric data. Also, the learning data generating apparatus may calculate a good of fit (GOF) based on the estimation function.

According to an embodiment, the symptom diagnosis apparatus may determine the presence or absence of a predetermined symptom by analyzing a feature value using an analysis model learned from learning data labeled with a predetermined symptom ( 530 ).

According to an example, the analysis model may predict a child's mental illness and symptoms using machine learning and deep learning techniques based on the learning data. As an example, the machine learning technique may use a random forest, eXtreme Gradient Boosting (XGBoost), or Light Gradient Boosting Machine (LightGBM) classifier. As an example, the deep learning model may use Long-Short Term Memory (LSTM).

An aspect of the present invention may be implemented as computer-readable codes on a computer-readable recording medium. Codes and code segments implementing the above program can be easily inferred by a computer programmer in the art. The computer-readable recording medium may include any type of recording device in which data readable by a computer system is stored. Examples of the computer-readable recording medium may include ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical disk, and the like. In addition, the computer-readable recording medium may be distributed in network-connected computer systems, and may be written and executed as computer-readable codes in a distributed manner.

So far, the present invention has been looked at with respect to preferred embodiments thereof. Those of ordinary skill in the art to which the present invention pertains will understand that the present invention may be implemented in a modified form without departing from the essential characteristics of the present invention. Accordingly, the scope of the present invention is not limited to the above-described embodiments, and should be construed to include various embodiments within the scope equivalent to the content described in the claims.

100: training data generating device
110: data receiving unit
120: data analysis unit
130: training data generation unit
300: symptom diagnosis device
310: data receiving unit
320: data analysis unit
330: diagnostic unit

Claims (16)

a data receiver configured to receive biometric data from a wearable device worn by a user;
When the biometric data includes data for a predetermined time or longer, it is determined that the biometric data is valid, and the biometric data determined to be valid is returned as biometric data in units of one cycle, a data analysis unit for extracting the above feature values; and
And a learning data generator for generating learning data by performing labeling on the analysis data generated based on the one or more feature values,
The learning data generator receives checklist data based on Kiddie Schedule for Affective Disorders and Schizophrenia Present and Lifetime version (K-SADS-PL) including feature value information for each predetermined symptom, and based on the received checklist data to generate the learning data by performing labeling in which a label including whether the user is diagnosed with a mental illness and whether or not the user has a symptom is performed on the biometric data of the cycle unit.
The method of claim 1,
The biometric data includes data on at least one of a heart rate, a step count, a metabolic rate (METs, Metabolic rates), a calorie consumption, and a sleep state, a learning data generating device.
delete delete delete The method of claim 1,
The learning data generation unit
A learning data generating apparatus for classifying the learning data into training data, verification data, and test data according to a predetermined ratio.
receiving biometric data from a wearable device worn by a user;
When the biometric data includes data for a predetermined time or longer, it is determined that the biometric data is valid, and the biometric data determined to be valid is returned as biometric data in units of one cycle, a data analysis step of extracting the above feature values; and
A learning data generation step of generating learning data by performing labeling on the analysis data generated based on the one or more feature values;
The step of generating the learning data is,
Receive checklist data based on K-SADS-PL (Kiddie Schedule for Affective Disorders and Schizophrenia Present and Lifetime version) including feature value information for each predetermined symptom, and based on the received checklist data, A method for generating learning data, characterized in that the learning data is generated by performing labeling in which a label including whether the user is diagnosed with a mental illness and whether the user has symptoms is performed on the biometric data.
8. The method of claim 7,
The bio-data includes data on at least one of heart rate, step number, metabolic rate (METs, Metabolic rates), calorie consumption, and sleep state, learning data generation method.
delete delete delete 8. The method of claim 7,
The step of generating the training data is
Classifying the learning data into training data, verification data, and test data according to a predetermined ratio, a method of generating learning data.
a data receiver configured to receive biometric data from a wearable device worn by a user;
When the biometric data includes data for a predetermined time or longer, it is determined that the biometric data is valid, and the biometric data determined to be valid is returned as biometric data in units of one cycle, a data analysis unit for extracting the above feature values; and
a diagnosis unit for determining the presence or absence of a predetermined symptom by analyzing the feature value using an analysis model learned from the learning data labeled with the user's mental illness diagnosis and symptom status;
The learning data is based on checklist data based on K-SADS-PL (Kiddie Schedule for Affective Disorders and Schizophrenia Present and Lifetime version) including feature value information for each predetermined symptom. A symptom diagnosis device, characterized in that the label including the diagnosis and symptoms of mental illness is given and generated.
14. The method of claim 13,
The analysis model is a machine learning model or a deep learning model, symptom diagnosis apparatus.
14. The method of claim 13,
The biometric data includes data on at least one of heart rate, step number, metabolic rate (METs, Metabolic rates), calorie consumption, and sleep state, symptom diagnosis apparatus.
A method for diagnosing a symptom in a symptom diagnosis apparatus, the method comprising:
receiving biometric data from a wearable device worn by a user;
When the biometric data includes data for a predetermined time or longer, it is determined that the biometric data is valid, and the biometric data determined to be valid is returned as biometric data in units of one cycle, a data analysis step of extracting the above feature values; and
A determination step of outputting data for judging the presence or absence of a predetermined symptom by analyzing the feature value using an analysis model trained with learning data labeled with a predetermined symptom;
The learning data is based on checklist data based on K-SADS-PL (Kiddie Schedule for Affective Disorders and Schizophrenia Present and Lifetime version) including feature value information for each predetermined symptom. A method for diagnosing symptoms, characterized in that the label including whether or not a diagnosis of mental illness is diagnosed and whether or not a symptom is present is generated.

Images