WO2020122227A1 - うつ状態を推定する装置、方法及びそのためのプログラム - Google Patents
うつ状態を推定する装置、方法及びそのためのプログラム Download PDFInfo
- Publication number
- WO2020122227A1 WO2020122227A1 PCT/JP2019/048904 JP2019048904W WO2020122227A1 WO 2020122227 A1 WO2020122227 A1 WO 2020122227A1 JP 2019048904 W JP2019048904 W JP 2019048904W WO 2020122227 A1 WO2020122227 A1 WO 2020122227A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- data
- subject
- estimating
- unit time
- estimation model
- Prior art date
Links
Images
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B10/00—Other methods or instruments for diagnosis, e.g. instruments for taking a cell sample, for biopsy, for vaccination diagnosis; Sex determination; Ovulation-period determination; Throat striking implements
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/01—Measuring temperature of body parts ; Diagnostic temperature sensing, e.g. for malignant or inflamed tissue
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/02—Detecting, measuring or recording pulse, heart rate, blood pressure or blood flow; Combined pulse/heart-rate/blood pressure determination; Evaluating a cardiovascular condition not otherwise provided for, e.g. using combinations of techniques provided for in this group with electrocardiography or electroauscultation; Heart catheters for measuring blood pressure
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/16—Devices for psychotechnics; Testing reaction times ; Devices for evaluating the psychological state
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/22—Ergometry; Measuring muscular strength or the force of a muscular blow
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F1/00—Details not covered by groups G06F3/00 - G06F13/00 and G06F21/00
- G06F1/16—Constructional details or arrangements
- G06F1/1613—Constructional details or arrangements for portable computers
- G06F1/163—Wearable computers, e.g. on a belt
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H20/00—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance
- G16H20/10—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance relating to drugs or medications, e.g. for ensuring correct administration to patients
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H40/00—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices
- G16H40/60—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the operation of medical equipment or devices
- G16H40/63—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the operation of medical equipment or devices for local operation
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/30—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for calculating health indices; for individual health risk assessment
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients
Definitions
- the present invention relates to an apparatus, method and program for estimating a depressive state, and more specifically, an apparatus, method and method for estimating the depression state of depression or manic depression based on biometric data. Regarding the program.
- Mood disorders are broadly classified into “depressive disorders” only for depressive episodes and “bipolar disorders” with symptoms called “manic episodes”.
- Patent Document 1 a predetermined conditional expression based on the amount of activity and pulsation interval of the subject is provided. Techniques for determining bipolar disorder when satisfied are disclosed.
- An object of the present invention is to provide a new apparatus, method, and program for estimating a depression state using a wearable device.
- a first aspect of the present invention is a method of estimating a depression state of a subject based on biometric data, in which biometric data including a plurality of data types, A step of converting into unit time data of a predetermined time unit, a step of extracting one or a plurality of feature amounts based on the unit time data, and a predetermined number of at least a part of the one or a plurality of feature amounts as inputs Estimating the depression state using the estimated model.
- a second aspect of the present invention is characterized in that, in the first aspect, the biometric data is data measured by a wearable device worn by the subject or data corresponding thereto.
- a third aspect of the present invention is characterized in that, in the first or second aspect, the biometric data is data measured over a period of 48 hours or more or data corresponding thereto.
- a fourth aspect of the present invention is characterized in that, in any one of the first to third aspects, the predetermined time is 1 hour.
- a fifth aspect of the present invention is characterized in that, in any one of the first to fourth aspects, the plurality of data types includes skin temperature.
- a sixth aspect of the present invention in the fifth aspect, wherein the plurality of data types further include at least one of steps, energy consumption, body movement, heart rate, sleep state, and ultraviolet ray level. Is characterized by.
- a seventh aspect of the present invention is characterized in that, in any one of the first to sixth aspects, the one or more feature amounts are the quantile of the unit time data of each data type, the quantile of each unit of the data type. It is characterized in that it includes at least one of a standard deviation of unit time data and a correlation coefficient of each combination of a plurality of data types.
- An eighth aspect of the present invention is characterized in that, in any one of the first to seventh aspects, the at least a part of the one or more feature amounts is selected by regularization.
- a ninth aspect of the present invention is characterized in that, in any one of the first to eighth aspects, the estimation model is a model for estimating the presence or absence of a depressive state generated by machine learning.
- a tenth aspect of the present invention is characterized in that, in any of the first to eighth aspects, the estimation model is a model for estimating the severity of a depression state generated by machine learning. ..
- the eleventh aspect of the present invention is characterized in that, in the tenth aspect, the severity is a HAMD score.
- a twelfth aspect of the present invention is characterized in that, in the tenth or eleventh aspect, the biometric data is data measured over a period of 72 hours or more or data corresponding thereto.
- a thirteenth aspect of the present invention is a program for causing a computer to execute a method for estimating a depression state of a subject based on biometric data, wherein the method stores biometric data including a plurality of data types. For each data type, a step of converting into unit time data of a predetermined time unit, a step of extracting one or more characteristic amounts from the unit time data, and an input of at least a part of the one or more characteristic amounts , Estimating a depression state using a predetermined estimation model.
- a fourteenth aspect of the present invention is a device for estimating a depression state of a subject based on biometric data, wherein biometric data including a plurality of data types is unit time data of a predetermined time unit for each data type. And extracting one or a plurality of feature quantities from the unit time data, inputting at least a part of the one or a plurality of feature quantities, and estimating a depression state using a predetermined estimation model.
- a fifteenth aspect of the present invention is a method of generating an estimation model for estimating a depression state based on biometric data of a plurality of subjects, wherein biometric data including a plurality of data types is data for each subject. For each type, a step of converting into unit time data of a predetermined time unit, a step of extracting one or more characteristic amounts from the unit time data for each subject, and a step of extracting the one or more characteristic amounts of each subject. And a step of generating the estimation model by machine learning using teacher data in which at least a part of the input vector is used as a label and a diagnosis result by an expert for each subject is used as a label.
- a sixteenth aspect of the present invention is characterized in that, in the fifteenth aspect, the machine learning is ensemble learning.
- a seventeenth aspect of the present invention is based on the fifteenth or sixteenth aspect, wherein the one or more characteristic amounts are quantiles of the unit time data of each data type, and the unit time data of each data type. Standard deviation and/or correlation coefficient of each combination of a plurality of data types.
- the eighteenth aspect of the present invention is characterized in that, in any one of the fifteenth to seventeenth aspects, at least a part of the one or more feature amounts is selected by regularization.
- a nineteenth aspect of the present invention is a program for causing a computer to execute a method for generating an estimation model for estimating a depression state based on biological data of a plurality of subjects, wherein the generation method comprises: A step of converting biometric data including a plurality of data types for each subject into unit time data of a predetermined time unit for each data type; and extracting one or a plurality of feature quantities from the unit time data for each subject And a step of generating the estimation model by machine learning using at least a part of the one or more feature amounts of each subject as input and using teacher data with a diagnosis result by an expert for each subject as a label. It is characterized by including and.
- a twentieth aspect of the present invention is a device that generates an estimation model for estimating a depression state based on biometric data of a plurality of subjects, and obtains biometric data including a plurality of data types for each subject. For each data type, it is converted into unit time data of a predetermined time unit, one or more characteristic quantities are extracted from the unit time data for each subject, and at least a part of the one or more characteristic quantities of each subject is extracted. Is input, and the estimation model is generated by machine learning using teacher data in which a diagnosis result by an expert for each subject is used as a label.
- biometric data including a plurality of data types is acquired using a wearable device, and a depression state is estimated by one or a plurality of feature amounts based on the biometric data, thereby estimating a conventional estimation technique. Can be improved.
- a wristband type product called Silmee (registered trademark) W20 including an acceleration sensor, a pulse sensor, an ultraviolet sensor and a temperature sensor is used.
- the acceleration sensor obtains biological data regarding the number of steps, energy consumption, body movements and sleep state
- the ultraviolet sensor obtains an ultraviolet level
- the temperature sensor obtains biological data relating to the skin temperature.
- the biological data regarding the sleep state obtained from the acceleration sensor has a predetermined value calculated from the acceleration in each axis direction of the triaxial acceleration sensor, as in the method known as actigraphy. It can be obtained by determining that it is awake if it exceeds the threshold of, and by determining that it is sleeping if it does not exceed.
- the sleep state at each time or each time when the value of the acceleration sensor is recorded is represented by 1 if the sleeping state is a sleep state, and is represented by 0 if the awake state is awake state. It is possible to evaluate the time when the state in which 1 is 1 is continuous as the sleep time and evaluate the start time and the end time as the bedtime and the wakeup time, respectively. Further, as an example, in the sleep state evaluation, a value such as a heart rate other than the value of the acceleration sensor may be reflected.
- the biological data relating to the skin temperature (skin temperature) obtained from the temperature sensor can be obtained, more specifically, by measuring the skin temperature of a part such as a wrist on which the wearable device 110 is worn.
- Accurate body temperature is called core temperature and is obtained by measuring rectal temperature, axillary temperature, etc.
- the skin temperature measured in this embodiment does not impose a burden on the subject for measurement, and continuous data is obtained. Makes it possible to obtain
- FIG. 1 shows an apparatus according to this embodiment.
- the apparatus 100 is connected to the wearable device 110 in a wired or wireless manner and receives biometric data obtained by the wearable device 110.
- the wearable device 110 is connected to a mobile terminal such as a smartphone and receives data via the mobile terminal, or the wearable device 110 is not connected to the Internet via the wearable device 110, instead of directly receiving the biometric data as illustrated.
- It may be connected to a server on the IP network such as, and receive data via the server, or input the obtained biometric data to the apparatus 100 via the USB cradle.
- the apparatus 100 includes a processing unit 101 such as a processor and a CPU, a storage unit 102 including a storage device or a storage medium such as a memory and a hard disk, and a communication unit 103 such as a communication interface for communicating with other devices in a wired or wireless manner. And a program for performing each process is executed by the processing unit 101.
- the device 100 may include one or a plurality of devices, a computer or a server, the program may include one or a plurality of programs, and the program may be recorded in a computer-readable storage medium. Can be a sex program product.
- an estimation model for estimating the presence/absence of a depressed state is generated by machine learning. Then, a new subject's depression state is estimated using the estimation model.
- the generation of the estimation model and the estimation using the estimation model will be described separately. Although both are described as processing in the apparatus 100, it is assumed that the generation of the estimation model and the estimation using the estimation model can be performed by different computers.
- the method according to the present embodiment is useful for screening for a depressed state, and can be particularly used for medical examinations in industrial health.
- the biometric data for m days (m is a positive number) from 1 person (l is a positive integer) for each data type is a predetermined time unit data such as one hour (hereinafter, “unit time data ( (unit time data)”)) (S201).
- the data types included in the biometric data include steps, energy consumption, body movement, heart rate, sleep state, skin temperature, and ultraviolet ray level.
- the data type may also be referred to as a modality.
- the number of steps, energy consumption, body movement, and sleep state, that is, sleep time is a value accumulated over a predetermined time period
- heart rate, skin temperature, and UV level are the average values at the predetermined time period. It can be time data. Further, it may be a value obtained by integrating the heart rate, the skin temperature, and the ultraviolet ray level over a predetermined time.
- the case where the unit time data is one hour is taken as an example.
- the standard deviation of the distribution of the obtained 24 m sample data is calculated for each subject and for each data type (S203).
- a correlation coefficient such as a Pearson correlation coefficient is calculated for each combination of data types (S204). If there are seven data types as described above, there are 21 combinations. Although the order of the quantile, standard deviation, and Pearson correlation coefficient is described here, they can be performed in any order.
- the feature quantities such as quantiles, standard deviations, and correlation coefficients extracted from the biometric data of each subject in this way are input or input vectors, and the presence or absence of depression state by a specialist such as a doctor for each subject (presence). or absense) is used as a label, and the estimation model for the classification problem of whether or not the patient is depressed is trained by machine learning using the results of l-1 persons as teacher data (S206).
- the generated estimation model is stored in the storage unit 102 of the device 100 that performs estimation using the estimation model or a storage medium or a storage device accessible from the device 100.
- the subset that is a part of the extracted feature amount may be selected by, for example, L1 regularization, L2 regularization, which is one of the regularization methods, or Elastic Net that combines these (S205), and this may be used as the input vector.
- machine learning can be performed by ensemble learning such as XGBoost, which is one method of gradient boosting.
- k-fold cross validation is a method of repeating verification k times using any one set of data obtained by dividing teacher data by k. Since the verification method affects the data used for learning and thus the accuracy of the estimation model to be generated, the k-divided cross verification generally mitigates the problem of overfitting as compared with LOOCV.
- the biometric data of m′ days (m′ is a positive number) of the subject for whom the presence/absence of depression is determined is converted into unit time data for each data type (S301).
- the conversion to the unit time data is performed by further converting the biometric data measured by the wearable device 110 or the data corresponding to the biometric data processed by the wearable device 110 or the apparatus 100 into 1-minute data.
- Various methods can be used, such as conversion into time unit data. This point is the same when the prediction model is generated. However, it is desirable that the unit time at the time of estimation and the unit time at the time of generation are the same.
- the feature amount is extracted and stored for each data type (S302). If necessary, a subset of the feature amount is selected, and the presence or absence of the depression state is estimated using the estimation model generated in advance using the feature amount or the subset as an input or an input vector (S303).
- the specific product is given as an example of the wearable device, but any device that includes an acceleration sensor, a pulse sensor, an ultraviolet sensor, and a temperature sensor may be used, and it greatly contributes to the high accuracy of the prediction model. It is preferable that the wearable device includes at least a temperature sensor considered to be.
- method etc. has an aspect of performing an operation different from the operation described in this specification, each of the present invention
- the aspect is intended for the same operation as any one of the operations described in the present specification, and the existence of an operation different from the operation described in the present specification means that the method etc. It is added that it does not fall outside the scope of the embodiment.
- Example 1-1 Biological data for two or more days for 62 subjects were acquired using Silveree W20, and an estimation model was generated by the method according to the present embodiment. The estimation model was verified by leave-one-out cross-validation. The following table shows the result of comparison between the estimation result of the presence/absence of a depression state using the estimation model and the diagnosis result by the doctor.
- the biometric data two or more days out of the six days before the diagnosis date are used, and subset selection by Elastic Net and machine learning by XGBoost are performed.
- the unit time data is a value in units of 1 hour.
- HAMD Hamilton Rating Scale
- the depression state is estimated with high accuracy, and that it is possible to screen a healthy person and a depressed patient.
- the biometric data includes two or more data types including the skin temperature and at least one of the number of steps, energy, body movement, heart rate, sleep state, and ultraviolet ray level.
- Example 1-2 Under the same conditions as in Example 1-1, biometric data for three or more days out of six days before the diagnosis date was acquired for 55 subjects, and the estimation model generation and the depression state were performed by the method according to the present embodiment.
- the following table shows the results of the estimation.
- the accuracy rate was 0.855, the sensitivity was 0.862, and the specificity was 0.846.
- the depression state is estimated with high accuracy, but no significant improvement in precision is observed as compared with the estimation of the presence or absence of the depression state performed on the biometric data for two days. It can be said that the method according to the present embodiment can obtain sufficiently high accuracy by using biometric data for 2 days or 48 hours. This leads to a reduction in the period in which the subject has to wear the wearable device 110, and a short wearing time and a small burden on the subject promote the spread of screening for early detection of depression.
- Example 1-3 Biological data for three or more days for 86 test subjects were acquired using Silveree W20, and the estimation model was generated by the method according to the present embodiment. The estimation model was verified by 10-fold cross validation. The following table shows the result of comparison between the estimation result of the presence/absence of a depression state using the estimation model and the diagnosis result by the doctor.
- As the biometric data 228 sample data obtained for 86 people were divided into 10. Here, the data of the same test subject was made to fit in the same set (fold). In addition, we obtained multiple datasets from some subjects. Machine selection by XGBoost is performed without selecting a subset.
- the unit time data is a value in units of 1 hour.
- the accuracy rate was 0.737, the sensitivity was 0.661, and the specificity was 0.807. It is considered that the relaxation of overfitting caused by changing the verification method from LOOCV to 10-fold cross verification causes a slight decrease in the accuracy rate, etc., but the estimation is still performed with high accuracy.
- the following table shows the result of calculating the importance of each feature amount indicating which feature amount contributes significantly to the estimation result.
- the importance of each feature is estimated by calculating how much the model accuracy improves when is changed, and the service "Azure Machine Learning (trademark)" provided by Microsoft was used. The importance is 1 when all feature quantities are summed up.
- the correlation coefficient between sleep state and skin temperature, the 50th percentile of skin temperature, and the standard deviation of body movement are highly important feature values, and it is confirmed that the contribution of skin temperature is large. It was Further, the contribution of the correlation coefficient between the skin temperature and the sleep state is large, and it can be said that it is preferable that the biological data includes the sleep state in addition to the skin temperature.
- the second embodiment of the present invention makes it possible to estimate not only the presence/absence of a depressive state, but also the presence/absence of a depressive state in order to estimate the severity of depression.
- an estimation model for estimating the severity of a depressive state is generated by machine learning. Then, the severity of the new subject's depressed state is estimated using the estimation model.
- the generation of the estimation model in this embodiment is similar to that described with reference to FIG. 2 in the first embodiment, and the diagnosis result of the doctor used in machine learning is not the presence/absence of depression but the depression. It differs in that it is a score of severity and that the algorithm used in machine learning is not an algorithm suitable for a classification problem but an algorithm suitable for a regression problem.
- the estimation using the estimation model in the present embodiment is the same as that described in the first embodiment with reference to FIG. 3, and the output is not the presence or absence of the depression but the severity of the depression. Different.
- the Hamilton Rating Scale HAMD can be used for the severity of depression. There are several versions of HAMD17, and experts such as doctors evaluate 17 items, and there are 3 to 5 points for each item, for example, 7 points are normal, 8 to 13 points are mild, 14 A diagnosis of 18 to 18 is moderate, 19 to 22 is severe, and 23 or more is most severe.
- the method of the present embodiment since it is possible to determine the severity of the depression state, it is possible to contribute to the precise determination necessary for treatment, such as the determination of the treatment policy for the depression state, the change of the prescription drug, etc. Highly useful in assisting diagnostic treatment in psychiatry.
- Example 2-1 Biological data for two or more days for 62 subjects were acquired using Silveree W20, and an estimation model was generated by the method according to the present embodiment. The estimation model was verified by leave-one-out cross-validation.
- FIG. 4 shows the result of comparison between the estimation result of the presence/absence of a depression state using the estimation model and the diagnosis result by the doctor.
- the biometric data two or more days out of the six days before the diagnosis date are used, and subset selection by Elastic Net and machine learning by XGBoost are performed.
- the unit time data is a value in units of 1 hour.
- the average absolute error was 4.11, the correlation coefficient was 0.604, and the p-value was 2.04 ⁇ 10 ⁇ 7 .
- R 2 is 0.341.
- Example 2-2 Under the same conditions as in Example 2-1, biometric data of 55 subjects for 3 days or more were acquired, and an estimation model was generated and the severity of depression was estimated by the method according to the present embodiment. The result is shown in FIG.
- the average absolute error was 3.29, the correlation coefficient was 0.763, and the p-value was 1.22 ⁇ 10 -11 .
- R 2 is 0.570.
- This result shows a higher correlation coefficient than that of Example 2-1, and it is necessary to wear it for 3 days or 72 hours or more in order to obtain a high estimation accuracy exceeding 0.700 in severity estimation. It turns out that there is a high possibility that This means that, in contrast to Examples 1-1 and 1-2, the model for estimating the severity of the depressive state is longer than the model for estimating the presence/absence of the depressive state in the biological data measured. It means that it is preferable to generate by using.
- Example 2-3 Biological data for 7 or more days for 86 test subjects were acquired using Silveree W20, and the estimation model was generated by the method according to the present embodiment. The estimation model was verified by 10-fold cross validation. Then, FIG. 6 shows a result of comparison between the estimation result of the presence or absence of the depression state using the estimation model and the diagnosis result by the doctor.
- the biometric data 236 sample data obtained for 86 people were divided into 10 parts. Here, the data of the same test subject was made to fit in the same set (fold). In addition, we obtained multiple datasets from some subjects. Machine selection by XGBoost is performed without selecting a subset. The unit time data is a value in units of 1 hour.
- the average absolute error was 4.94
- the correlation coefficient was 0.610
- the p-value was 2.20 ⁇ 10 -16 .
- R 2 is 0.372.
- the following table shows the results of calculating the importance of each feature amount that indicates which feature amount contributes significantly to the estimation result. Specifically, this is provided by Microsoft. The service “Azure Machine Learning (trademark)" was used. The importance is 1 when all feature quantities are summed up.
- the 95th percentile of the skin temperature, the correlation coefficient between the sleep state and the skin temperature, and the 50th percentile of the skin temperature were found to be the features of high importance, Similarly, it was confirmed that the contribution of skin temperature was large. Further, as in Example 1-3, the contribution of the correlation coefficient between the skin temperature and the sleep state is large, and it can be said that it is preferable that the biological data includes the sleep state in addition to the skin temperature.
- apparatus 101 processing unit 102 storage unit 103 communication unit 110 wearable device
Abstract
Description
本発明の第1の実施形態では、ウェアラブルデバイスとして、リストバンド型で加速度センサ、脈拍センサ、紫外線センサ及び温度センサを備えるSilmee(登録商標)W20という製品を用いた。加速度センサから歩数、消費エネルギー、体動及び睡眠状態に関する生体データが得られ、紫外線センサからは紫外線レベルが得られ、温度センサから肌温度に関する生体データが得られる。
l人(lは正の整数)の被験者からのm日分(mは正の数)の生体データをデータタイプごとに、1時間等のあらかじめ定めた時間単位のデータ(以下「単位時間データ(unit time data)」と呼ぶ。)に変換する(S201)。生体データに含まれるデータタイプとして、歩数、消費エネルギー、体動、心拍数、睡眠状態、肌温度及び紫外線レベルが挙げられる。データタイプは、モダリティ(modality)とも呼ぶことがある。一例として、歩数、消費エネルギー、体動、及び睡眠状態、すなわち睡眠時間については、あらかじめ定めた時間にわたって積算した値、心拍数、肌温度及び紫外線レベルについては、あらかじめ定めた時間における平均値を単位時間データとすることができる。また、心拍数、肌温度及び紫外線レベルについて、あらかじめ定めた時間にわたって積算した値としてもよい。
うつ状態の有無の判定を行う被験者のm’日分(m’は正の数)の生体データをデータタイプごとに単位時間データに変換する(S301)。単位時間データへの変換は、ウェアラブルデバイス110が測定した生体データ又はそれに対してウェアラブルデバイス110若しくは装置100において処理を施した当該生体データに対応するデータを1分単位のデータに変換した後にさらに1時間単位のデータに変換して行うなど、さまざまな手法を用いることができる。この点は、予測モデルの生成時においても同様である。ただし、推定時の単位時間と生成時の単位時間は等しいことが望ましい。
62人の被験者について2日分以上の生体データをSilmee W20を用いて取得し、本実施形態にかかる手法によって推定モデルを生成した。推定モデルの検証は、leave-one-out交差検証により行った。そして、当該推定モデルを用いたうつ状態の有無の推定結果と医師による診断結果と比較した結果が以下の表である。生体データは、診断日前の6日間のうちの2日分以上を用いており、Elastic Netによるサブセットの選択及びXGBoostによる機械学習がなされている。単位時間データは、1時間単位の値である。
実施例1-1と同様の条件下で、55人の被験者について診断日前の6日間のうちの3日分以上の生体データを取得し、本実施形態にかかる手法によって推定モデルの生成及びうつ状態の推定を行った結果が以下の表である。
86人の被験者について3日分以上の生体データをSilmee W20を用いて取得し、本実施形態にかかる手法によって推定モデルを生成した。推定モデルの検証は、10-fold交差検証により行った。そして、当該推定モデルを用いたうつ状態の有無の推定結果と医師による診断結果と比較した結果が以下の表である。生体データは、86人について得られた228個のサンプルデータを10分割した。ここで、同一の被験者のデータは同一の組(fold)に収まるようにした。また、一部の被験者からは複数のデータセットを取得している。サブセットの選択は行わず、XGBoostによる機械学習がなされている。単位時間データは、1時間単位の値である。
本発明の第2の実施形態は、うつ状態の有無のみならず、あるいはうつ状態の有無に代替して、うつ状態の重症度(severity)を推定可能とする。まず、複数の被験者からウェアラブルデバイス110により得られる生体データを用いて、うつ状態の重症度を推定するための推定モデルを機械学習によって生成する。そして、当該推定モデルを用いて新たな被験者のうつ状態の重症度を推定する。
62人の被験者について2日分以上の生体データをSilmee W20を用いて取得し、本実施形態にかかる手法によって推定モデルを生成した。推定モデルの検証は、leave-one-out交差検証により行った。そして、当該推定モデルを用いたうつ状態の有無の推定結果と医師による診断結果と比較した結果が図4である。生体データは、診断日前の6日間のうちの2日分以上を用いており、Elastic Netによるサブセットの選択及びXGBoostによる機械学習がなされている。単位時間データは、1時間単位の値である。
実施例2-1と同様の条件下で、55人の被験者について3日分以上の生体データを取得し、本実施形態にかかる手法によって推定モデルの生成及びうつ状態の重症度の推定を行った結果が図5である。
86人の被験者について7日分以上の生体データをSilmee W20を用いて取得し、本実施形態にかかる手法によって推定モデルを生成した。推定モデルの検証は、10-fold交差検証により行った。そして、当該推定モデルを用いたうつ状態の有無の推定結果と医師による診断結果と比較した結果が図6である。生体データは、86人について得られた236個のサンプルデータを10分割した。ここで、同一の被験者のデータは同一の組(fold)に収まるようにした。また、一部の被験者からは複数のデータセットを取得している。サブセットの選択は行わず、XGBoostによる機械学習がなされている。単位時間データは、1時間単位の値である。
101 処理部
102 記憶部
103 通信部
110 ウェアラブルデバイス
Claims (20)
- 被験者が身に付けたウェアラブルデバイスにより計測された生体データに基づいて被験者のうつ状態を推定する方法であって、
複数のデータタイプを含む生体データをデータタイプごとに、所定の時間単位の単位時間データに変換するステップと、
前記単位時間データに基づいて1又は複数の特徴量を抽出するステップと、
前記1又は複数の特徴量の少なくとも一部を入力として、あらかじめ定めた推定モデルを用いてうつ状態を推定するステップと
を含み、
前記1又は複数の特徴量は、各データタイプの前記単位時間データの分位数及び複数のデータタイプの各組合せの相関係数を含むことを特徴とする方法。 - 前記所定の時間単位は1時間単位であることを特徴とする1に記載の方法。
- 前記複数のデータタイプは、肌温度を含むことを特徴とする請求項1から4のいずれかに記載の方法。
- 前記1又は複数の特徴量は、肌温度の分位数を含むことを特徴とする請求項3に記載の方法。
- 前記複数のデータタイプは、睡眠状態を含むことを特徴とする請求項3又は4に記載の方法。
- 前記1又は複数の特徴量は、肌温度と睡眠状態との相関係数を含むことを特徴とする請求項5に記載の方法。
- 前記1又は複数の特徴量は、各データタイプの前記単位時間データの標準偏差をさらに含むことを特徴とする請求項1から6のいずれかに記載の方法。
- 前記1又は複数の特徴量の前記少なくとも一部は、正則化によって選択することを特徴とする請求項1から7のいずれかに記載の方法。
- 前記推定モデルは、機械学習により生成されたうつ状態の有無を推定するモデルであることを特徴とする請求項1から8のいずれかに記載の方法。
- 前記推定モデルは、機械学習により生成されたうつ状態の重症度を推定するモデルであることを特徴とする請求項1から8のいずれかに記載の方法。
- 前記重症度は、HAMDのスコアであることを特徴とする請求項10に記載の方法。
- 前記推定モデルは、機械学習により生成されたうつ状態の有無を推定する第1のモデル及び機械学習により生成されたうつ状態の重症度を推定する第2のモデルを含み、
前記第2のモデルは生成に用いる生体データは、前記第1のモデルの生成に用いる生体データよりも長期間計測されたものであることを特徴とする請求項1から8のいずれかに記載の方法。 - コンピュータに、被験者が身に付けたウェアラブルデバイスにより計測された生体データに基づいて被験者のうつ状態を推定する方法を実行させるためのプログラムであって、前記方法は、
複数のデータタイプを含む生体データをデータタイプごとに、所定の時間単位の単位時間データに変換するステップと、
前記単位時間データから1又は複数の特徴量を抽出するステップと、
前記1又は複数の特徴量の少なくとも一部を入力として、あらかじめ定めた推定モデルを用いてうつ状態を推定するステップと
を含み、
前記1又は複数の特徴量は、各データタイプの前記単位時間データの分位数及び複数のデータタイプの各組合せの相関係数を含むことを特徴とする方法。 - 被験者が身に付けたウェアラブルデバイスにより計測された生体データに基づいて被験者のうつ状態を推定する装置であって、
複数のデータタイプを含む生体データをデータタイプごとに、所定の時間単位の単位時間データに変換し、
前記単位時間データから1又は複数の特徴量を抽出し、
前記1又は複数の特徴量の少なくとも一部を入力として、あらかじめ定めた推定モデルを用いてうつ状態を推定し、
前記1又は複数の特徴量は、各データタイプの前記単位時間データの分位数及び複数のデータタイプの各組合せの相関係数を含むことを特徴とする装置。 - 複数の被験者の生体データに基づいてうつ状態を推定するための推定モデルの生成方法であって、
被験者ごとに、複数のデータタイプを含む各被験者が身に付けたウェアラブルデバイスにより計測された生体データをデータタイプごとに、所定の時間単位の時間単位データに変換するステップと、
被験者ごとに、前記時間単位データから1又は複数の特徴量を抽出するステップと、
各被験者の前記1又は複数の特徴量の少なくとも一部を入力ベクトルとし、各被験者についての専門家による診断結果をラベルとする教師データを用いた機械学習により、前記推定モデルを生成するステップと
を含み、
前記1又は複数の特徴量は、各データタイプの前記単位時間データの分位数及び複数のデータタイプの各組合せの相関係数を含むことを特徴とする生成方法。 - 前記機械学習は、アンサンブル学習であることを特徴とする請求項15に記載の方法。
- 前記1又は複数の特徴量は、各データタイプの前記時間単位データの標準偏差をさらに含むことを特徴とする請求項15又は16に記載の方法。
- 前記1又は複数の特徴量の前記少なくとも一部は、正則化によって選択することを特徴とする請求項15から17のいずれかに記載の方法。
- コンピュータに、複数の被験者の生体データに基づいてうつ状態を推定するための推定モデルの生成方法を実行させるためのプログラムであって、前記生成方法は、
被験者ごとに、複数のデータタイプを含む各被験者が身に付けたウェアラブルデバイスにより計測された生体データをデータタイプごとに、所定の時間単位の単位時間データに変換するステップと、
被験者ごとに、前記単位時間データから1又は複数の特徴量を抽出するステップと、
各被験者の前記1又は複数の特徴量の少なくとも一部を入力とし、各被験者についての専門家による診断結果をラベルとする教師データを用いた機械学習により、前記推定モデルを生成するステップと
を含み、
前記1又は複数の特徴量は、各データタイプの前記単位時間データの分位数及び複数のデータタイプの各組合せの相関係数を含むことを特徴とするプログラム。 - 複数の被験者の生体データに基づいてうつ状態を推定するための推定モデルを生成する装置であって、
被験者ごとに、複数のデータタイプを含む各被験者が身に付けたウェアラブルデバイスにより計測された生体データをデータタイプごとに、所定の時間単位の単位時間データに変換し、
被験者ごとに、前記単位時間データから1又は複数の特徴量を抽出し、
各被験者の前記1又は複数の特徴量の少なくとも一部を入力とし、各被験者についての専門家による診断結果をラベルとする教師データを用いた機械学習により、前記推定モデルを生成し、 前記1又は複数の特徴量は、各データタイプの前記単位時間データの分位数及び複数のデータタイプの各組合せの相関係数を含むことを特徴とする装置。
Priority Applications (4)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US17/413,476 US20220059226A1 (en) | 2018-12-14 | 2019-12-13 | Device and Method for Inferring Depressive State and Program for Same |
CN201980082589.5A CN113164163A (zh) | 2018-12-14 | 2019-12-13 | 推断抑郁状态的装置、方法及其程序 |
JP2020559336A JPWO2020122227A1 (ja) | 2018-12-14 | 2019-12-13 | うつ状態を推定する装置、方法及びそのためのプログラム |
CA3123192A CA3123192A1 (en) | 2018-12-14 | 2019-12-13 | Device and method for inferring depressive state and program for same |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2018234966 | 2018-12-14 | ||
JP2018-234966 | 2018-12-14 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2020122227A1 true WO2020122227A1 (ja) | 2020-06-18 |
Family
ID=71075648
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2019/048904 WO2020122227A1 (ja) | 2018-12-14 | 2019-12-13 | うつ状態を推定する装置、方法及びそのためのプログラム |
Country Status (5)
Country | Link |
---|---|
US (1) | US20220059226A1 (ja) |
JP (1) | JPWO2020122227A1 (ja) |
CN (1) | CN113164163A (ja) |
CA (1) | CA3123192A1 (ja) |
WO (1) | WO2020122227A1 (ja) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20220164149A (ko) * | 2021-06-04 | 2022-12-13 | 연세대학교 원주산학협력단 | 인공지능 모델을 이용한 우울증 발생 예측방법 및 이를 기록한 컴퓨터 판독 가능한 기록매체 |
WO2023033286A1 (ko) * | 2021-09-03 | 2023-03-09 | 삼성전자 주식회사 | 정신 건강 관리를 위한 전자 장치 및 이의 제어 방법 |
WO2023224085A1 (ja) * | 2022-05-19 | 2023-11-23 | 塩野義製薬株式会社 | 情報処理システムおよび情報処理方法 |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115496105B (zh) * | 2022-09-28 | 2023-10-24 | 广东省新黄埔中医药联合创新研究院 | 睡眠预测模型训练方法、睡眠状况预测方法及相关装置 |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2016163698A (ja) * | 2015-02-27 | 2016-09-08 | 公立大学法人首都大学東京 | 精神状態判定方法及び精神状態判定プログラム |
US20170238858A1 (en) * | 2015-07-30 | 2017-08-24 | South China University Of Technology | Depression assessment system and depression assessment method based on physiological information |
JP2018015327A (ja) * | 2016-07-28 | 2018-02-01 | 株式会社リコー | 診断装置、診断方法、プログラム |
JP2018524137A (ja) * | 2015-06-15 | 2018-08-30 | メディバイオ リミテッドMedibio Limited | 心理状態を評価するための方法およびシステム |
KR101911516B1 (ko) * | 2017-12-26 | 2018-10-24 | (주)에프앤아이 | 사용자의 심리 상태 데이터를 획득하고 사용자의 심리 상태를 판단하고 피드백을 제공하는 방법 및 이를 이용한 서버 |
CN108888281A (zh) * | 2018-08-16 | 2018-11-27 | 华南理工大学 | 精神状态评估方法、设备及系统 |
Family Cites Families (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2011109716A2 (en) * | 2010-03-04 | 2011-09-09 | Neumitra LLC | Devices and methods for treating psychological disorders |
WO2011159945A2 (en) * | 2010-06-16 | 2011-12-22 | Afraxis, Inc. | Methods for treating neurological conditions |
EP2661225A1 (fr) * | 2011-01-04 | 2013-11-13 | Jabourian, Artin Pascal | Systeme automatise de mesure du degre de motivation chez un individu et procede associe |
WO2014071281A1 (en) * | 2012-11-02 | 2014-05-08 | The Johns Hopkins University | Dna methylation biomarkers of post-partum depression risk |
WO2016031650A1 (ja) * | 2014-08-26 | 2016-03-03 | 東洋紡株式会社 | うつ状態判定方法、及びうつ状態判定装置 |
CN106175800A (zh) * | 2016-07-10 | 2016-12-07 | 北京工业大学 | 一种基于人体运动状态数据的心境状态量化评估方法 |
AU2017314831C1 (en) * | 2016-08-26 | 2023-01-05 | Akili Interactive Labs, Inc. | Cognitive platform coupled with a physiological component |
JP6784368B2 (ja) * | 2016-09-01 | 2020-11-11 | 東洋紡株式会社 | うつ状態判別方法及びうつ状態判定装置 |
-
2019
- 2019-12-13 JP JP2020559336A patent/JPWO2020122227A1/ja active Pending
- 2019-12-13 US US17/413,476 patent/US20220059226A1/en active Pending
- 2019-12-13 CN CN201980082589.5A patent/CN113164163A/zh active Pending
- 2019-12-13 WO PCT/JP2019/048904 patent/WO2020122227A1/ja active Application Filing
- 2019-12-13 CA CA3123192A patent/CA3123192A1/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2016163698A (ja) * | 2015-02-27 | 2016-09-08 | 公立大学法人首都大学東京 | 精神状態判定方法及び精神状態判定プログラム |
JP2018524137A (ja) * | 2015-06-15 | 2018-08-30 | メディバイオ リミテッドMedibio Limited | 心理状態を評価するための方法およびシステム |
US20170238858A1 (en) * | 2015-07-30 | 2017-08-24 | South China University Of Technology | Depression assessment system and depression assessment method based on physiological information |
JP2018015327A (ja) * | 2016-07-28 | 2018-02-01 | 株式会社リコー | 診断装置、診断方法、プログラム |
KR101911516B1 (ko) * | 2017-12-26 | 2018-10-24 | (주)에프앤아이 | 사용자의 심리 상태 데이터를 획득하고 사용자의 심리 상태를 판단하고 피드백을 제공하는 방법 및 이를 이용한 서버 |
CN108888281A (zh) * | 2018-08-16 | 2018-11-27 | 华南理工大学 | 精神状态评估方法、设备及系统 |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20220164149A (ko) * | 2021-06-04 | 2022-12-13 | 연세대학교 원주산학협력단 | 인공지능 모델을 이용한 우울증 발생 예측방법 및 이를 기록한 컴퓨터 판독 가능한 기록매체 |
KR102608408B1 (ko) * | 2021-06-04 | 2023-11-30 | 주식회사 커넥티드인 | 인공지능 모델을 이용한 우울증 발생 예측방법 및 이를 기록한 컴퓨터 판독 가능한 기록매체 |
WO2023033286A1 (ko) * | 2021-09-03 | 2023-03-09 | 삼성전자 주식회사 | 정신 건강 관리를 위한 전자 장치 및 이의 제어 방법 |
WO2023224085A1 (ja) * | 2022-05-19 | 2023-11-23 | 塩野義製薬株式会社 | 情報処理システムおよび情報処理方法 |
Also Published As
Publication number | Publication date |
---|---|
CN113164163A (zh) | 2021-07-23 |
US20220059226A1 (en) | 2022-02-24 |
CA3123192A1 (en) | 2020-06-18 |
JPWO2020122227A1 (ja) | 2021-12-16 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2020122227A1 (ja) | うつ状態を推定する装置、方法及びそのためのプログラム | |
Boursalie et al. | M4CVD: Mobile machine learning model for monitoring cardiovascular disease | |
Mohino-Herranz et al. | Assessment of mental, emotional and physical stress through analysis of physiological signals using smartphones | |
US20150148621A1 (en) | Methods and systems for creating a preventative care plan in mental illness treatment | |
KR20230137869A (ko) | 기계 학습 보조 인지 평가 및 치료를 위한 시스템들 및 방법들 | |
Hong et al. | Prediction of cardiac arrest in the emergency department based on machine learning and sequential characteristics: model development and retrospective clinical validation study | |
Kelly et al. | Automatic prediction of health status using smartphone-derived behavior profiles | |
Ho et al. | Imputation-enhanced prediction of septic shock in ICU patients | |
Pinto-Bernal et al. | A data-driven approach to physical fatigue management using wearable sensors to classify four diagnostic fatigue states | |
Kheirkhahan et al. | Actigraphy features for predicting mobility disability in older adults | |
Ali et al. | Multitask deep learning for cost-effective prediction of patient's length of stay and readmission state using multimodal physical activity sensory data | |
Ko et al. | Predicting sarcopenia of female elderly from physical activity performance measurement using machine learning classifiers | |
Chen et al. | Artificial Intelligence‐Based Medical Sensors for Healthcare System | |
Cosoli et al. | Methods for the metrological characterization of wearable devices for the measurement of physiological signals: state of the art and future challenges | |
Eom et al. | Deep learning-based optimal smart shoes sensor selection for energy expenditure and heart rate estimation | |
JP2020513274A (ja) | バランスを定量化するための多重パラメータ法 | |
Wu et al. | Personal sleep pattern visualization using sequence-based kernel self-organizing map on sound data | |
JP5911840B2 (ja) | 診断データ生成装置および診断装置 | |
Fonseca et al. | A computationally efficient algorithm for wearable sleep staging in clinical populations | |
CN116019429A (zh) | 基于生理指标的健康监测方法、装置、设备及存储介质 | |
Tsiouris et al. | Mining motor symptoms UPDRS data of Parkinson's disease patients for the development of Hoehn and Yahr estimation decision support system | |
WO2022216220A1 (en) | Method and system for personalized prediction of infection and sepsis | |
JP2021189093A (ja) | 自己免疫性水疱症の診断装置 | |
Wang et al. | A convenient non-harm cervical spondylosis intelligent identity method based on machine learning | |
JP2012228492A (ja) | 生体状態診断方法 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 19897220 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 2020559336 Country of ref document: JP Kind code of ref document: A |
|
ENP | Entry into the national phase |
Ref document number: 3123192 Country of ref document: CA |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 19897220 Country of ref document: EP Kind code of ref document: A1 |