Prediction method for vital signs
The method generates a personalized prediction model on a smartphone for future blood sugar levels by extracting feature values and correcting for time and unit mismatches, addressing inefficiencies in conventional methods by enabling accurate predictions and timely alerts without server reliance.
Patent Information
- Application Number
- JP2024527594
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2021-12-03
- Filing Date
- 2022-06-02
- Publication Date
- 2024-11-08
AI Technical Summary
Conventional methods for predicting future blood sugar levels require large memory and calculation resources, necessitating server communication and relying on blood sugar history information of both the user and surrounding users, which is inefficient and burdensome for diabetic patients.
A method for predicting future blood sugar levels using a communication terminal like a smartphone, generating a personalized prediction model based on user biometric history information, extracting first and second feature values from preprocessed biometric data, and correcting for time and unit mismatches, with the ability to relearn or regenerate the model based on prediction errors.
Enables accurate prediction of future blood sugar levels with minimal memory and calculation, providing personalized predictions without server connection, and addressing errors through relearning or regeneration, ensuring timely hypoglycemia or hyperglycemia alerts.
Smart Images

Figure 2024541344000001_ABST
Abstract
Description
[Technical field]
[0001] The present invention relates to a method for predicting a vital sign value in a blood glucose measurement system, and more particularly, to a method for predicting a vital sign value in which a prediction model is generated through a communication terminal having a small memory and a small amount of calculation, such as a smartphone, which a user always carries with him / her to manage his / her vital signs, and the generated prediction model is applied to the user's vital signs to predict the user's future vital signs, and the present invention relates to a method for predicting a vital sign value in which a prediction model personalized for the user is generated based on the user's vital signs history information, and the future vital signs of the user are predicted without requiring the vital signs of other nearby users and without connecting to a server. [Background technology]
[0002] Diabetes is a chronic disease that is common among modern people, affecting more than 2 million people in Korea, or 5% of the total population.
[0003] Diabetes occurs when there is an absolute or relative deficiency of insulin produced by the pancreas due to various causes such as obesity, stress, poor eating habits, and congenital genetics, which causes the blood to quickly balance the sugar content, resulting in an absolute excess of sugar in the blood.
[0004] Blood normally contains a certain concentration of glucose, from which tissue cells obtain their energy.
[0005] However, if glucose levels increase more than necessary, it cannot be properly stored in the liver, muscles, or fat cells and instead accumulates in the blood. As a result, diabetics maintain blood sugar levels much higher than normal people, and the excess blood sugar passes directly through the tissues and is excreted in the urine, causing a shortage of sugar, which is absolutely necessary for each tissue of the body, and causing abnormalities in each tissue.
[0006] Diabetes is characterized by the fact that in the early stages there are almost no noticeable symptoms, but as the disease progresses, specific symptoms of diabetes such as excessive drinking, eating, urination, weight loss, general fatigue, skin itching, and persistent wounds on the hands and feet that do not heal appear. As the disease progresses further, advanced complications such as vision problems, high blood pressure, kidney disease, stroke, periodontal disease, muscle spasms, neuralgia, and psoriasis appear.
[0007] In order to diagnose and manage diabetes so that it does not progress to complications, systematic blood glucose monitoring and treatment must be carried out simultaneously.
[0008] Diabetes requires constant monitoring of blood glucose levels to be managed, and the demand for blood glucose monitoring and related devices is constantly increasing. It has been confirmed through various studies that when diabetic patients strictly control their blood glucose levels, the occurrence of diabetic complications is significantly reduced. Therefore, it is very important for diabetic patients to regularly monitor their blood glucose levels to control their blood glucose levels.
[0009] Generally, blood glucose measuring devices using the finger prick method are used to manage blood glucose levels of diabetics. Although this method helps diabetics manage blood glucose levels, it has a problem that it is difficult to accurately grasp blood glucose levels that change frequently because it only shows the results at the time of measurement. In addition, blood glucose measuring devices require blood to be drawn every time to measure blood glucose levels at any time throughout the day, which places a heavy burden on diabetics.
[0010] Diabetic patients generally go through periods of hyperglycemia and hypoglycemia, but emergency situations occur during hypoglycemia. Hypoglycemia occurs when sugar is not sustained for a long period of time, and can result in loss of consciousness or, in the worst case, loss of life. Therefore, it is very important for diabetic patients to detect hypoglycemia immediately. However, blood glucose measuring devices that measure blood glucose intermittently have clear limitations.
[0011] To overcome the limitations of blood sampling type blood glucose measuring devices, a Continuous Glucose Monitoring System (CGMS) was developed that is inserted into the human body to measure blood glucose every few minutes, and can be used to easily manage diabetic patients and respond to emergency situations.
[0012] The continuous blood glucose measuring system includes a body-attached unit that is inserted into the human body to measure blood glucose using a body fluid such as the user's blood, and a communication terminal that communicates with the body-attached unit and displays the blood glucose level measured by the body-attached unit.
[0013] The body-attached unit is made up of a sensor that is partially inserted into the body for a certain period of time, for example up to three months (7 weeks, 15 days, 1 month, etc.), to generate a biosignal indicative of the user's blood glucose level from bodily fluids, and a transmitter that transmits the biosignal received by the sensor to the communication terminal in real time, periodically, or when requested by the communication terminal.
[0014] The body-attached unit generates a biosignal and transmits it to the communication terminal. The communication terminal is equipped with a blood glucose management application that receives the biosignal from the body-attached unit, performs pre-processing such as noise removal on the received biosignal, calibrates the received current value biosignal with a blood glucose level, and calibrates it using a reference blood glucose level, and then outputs it for the user to check.
[0015] In this way, the continuous blood glucose monitoring system can not only measure the user's biological signals in real time and output the user's blood glucose level, but also predict the user's future blood glucose level based on the user's blood glucose level history, user's activity history, meal history, medication administration history such as insulin, etc.
[0016] Such prediction of the user's future blood glucose level can be performed in various ways, but typically, a prediction model is generated based on the user's blood glucose level history and the blood glucose level history of other surrounding users, and the user's current blood glucose level is applied to the generated prediction model to predict the user's future blood glucose level.
[0017] Such a technology for predicting a user's future blood glucose level has the effect of predicting the possibility that the user may fall into hypoglycemia or hyperglycemia in advance, and enabling the user to take measures before the user actually reaches a hypoglycemic or hyperglycemic state.
[0018] However, conventional prediction models used to predict future blood glucose levels are generated using a large amount of blood glucose history information, meal history information, activity history, etc. of not only the user but also other surrounding users, so they require a very large memory space to store the information necessary to generate the prediction model, and have problems in that a large amount of calculation is required to apply the user's biological signals to the prediction model to predict future blood glucose levels.
[0019] Therefore, the generation and modification of such a prediction model, and the prediction of future blood glucose levels using the prediction model must mainly be performed through a separate server, and there is a problem in that the communication terminal must always communicate with the server to predict future blood glucose levels. DISCLOSURE OF THEINVENTION [Problem to be solved by the invention]
[0020] The present invention is intended to solve the problems of the conventional method for predicting future blood glucose level mentioned above, and an object of the present invention is to provide a method for predicting a vital sign value, which can predict a user's future blood glucose level with a small memory and calculation amount through a communication terminal such as a smartphone that the user always carries with them to manage their blood glucose.
[0021] Another object of the present invention is to provide a method for generating a prediction model personalized for a user based on the user's biometric history information, thereby predicting the user's future biometric values even without connecting to a server.
[0022] Another object of the present invention is to provide a method for predicting biometric values, which can accurately predict a user's future biometric values by calculating a prediction error from a user's predicted biometric value and an actual biometric value, determining an expression characteristic of the prediction error, and re-learning or regenerating a prediction model according to the expression characteristic of the prediction error.
[0023] Another object of the present invention is to provide a method for generating a prediction model personalized for a user using a first feature value extracted from pre-processed biometric information and a second feature value extracted from unit-corrected biometric information obtained by time-correcting the pre-processed biometric information, and for accurately predicting a user's biometric value through the generated prediction model. [Means for solving the problem]
[0024] In order to achieve the object of the present invention, a method for predicting a user's biometric value using biometric information measured from a sensor of a body-attachable unit includes a step of extracting a first feature value from the measured biometric information of the user, a step of correcting the measured biometric information of the user and extracting a second feature value from the corrected biometric information, a step of reducing and combining the first feature value and the second feature value to generate a feature vector value, and a step of applying the generated feature vector value to a prediction model to predict the user's biometric value.
[0025] The sensor is characterized in that it is partially inserted into the user's body for a certain period of time to measure the user's biological information.
[0026] Preferably, the method for predicting biometric information according to the present invention further includes a step of pre-processing the measured biometric information by removing noise from the measured biometric information, and the first feature value and the second feature value are extracted from the pre-processed biometric information.
[0027] Here, the first feature value is extracted directly from the preprocessed biometric information, and the second feature value is extracted from corrected biometric information generated by correcting the preprocessed biometric information for time delay and unit mismatch.
[0028] Here, the unit discrepancy is corrected based on pre-processed vital signs or reference vital signs.
[0029] In one embodiment of the present invention, the unit mismatch is corrected by applying an ascending or descending weight to the pre-processed biometric information.
[0030] In another embodiment of the present invention, the unit discrepancy is corrected by a weight assigned according to a difference between a vital sign value determined from measured vital signs and a reference vital sign value.
[0031] Preferably, the method for predicting a vital sign value according to the present invention further comprises the steps of calculating a prediction error from a difference between a predicted vital sign value at a first prediction time point and a vital sign value actually measured at the first prediction time point, and determining whether to re-learn a prediction model based on the prediction error.
[0032] Here, when the prediction error is greater than a critical value or a critical rate, the prediction model is retrained.
[0033] In the case where the prediction model is determined by re-learning in the method for predicting a biometric value according to the present invention, the prediction model is re-learned using a subsequent data set generated from the user's biometric information measured up to the present time in addition to the previous data set used to generate the prediction model.
[0034] Preferably, the method for predicting a vital sign value according to the present invention further comprises the step of determining whether to regenerate a prediction model based on an expression characteristic of a prediction error during a unit time.
[0035] Here, the expression feature is characterized by at least one of the number of times that the prediction error continuously exceeds a critical value or a critical ratio during a unit time and the total number of times that the prediction error exceeds a critical value or a critical ratio during a unit time. Effect of the Invention
[0036] The method for predicting vital signs according to the present invention has the following advantages.
[0037] First, the method for predicting a biometric value according to the present invention generates a prediction model through a communication terminal having a small memory and a small amount of calculation, such as a smartphone, which a user always carries with him / her to manage his / her biometric values, and predicts the user's future biometric value by applying the generated prediction model to the user's biometric information.
[0038] Second, the method for predicting vital signs according to the present invention creates a prediction model personalized for a user based on the user's vital signs history information, and can predict the user's future vital signs without needing the vital signs of other nearby users or connecting to a server.
[0039] Third, the method for predicting a biometric value according to the present invention calculates a prediction error from a predicted biometric value and an actual biometric value of a user, determines an expression characteristic of the prediction error, and re-learns or regenerates a prediction model according to the expression characteristic of the prediction error, thereby accurately predicting a user's future biometric value.
[0040] Fourth, the method for predicting a biometric value according to the present invention generates a prediction model personalized for a user using a first feature value extracted from pre-processed biometric information and a second feature value extracted from corrected biometric information obtained by time-correcting the pre-processed biometric information and performing unit correction, thereby making it possible to accurately predict the biometric value of the user through the generated prediction model. [Brief description of the drawings]
[0041] [Figure 1] FIG. 1 is a schematic diagram showing a blood glucose measuring system according to an embodiment of the present invention. [Diagram 2] FIG. 2 is a functional block diagram for explaining the health value predicting device according to the present invention. [Diagram 3] FIG. 3 is a functional block diagram for explaining an example of a feature value generating unit according to the present invention. [Figure 4]FIG. 4 is a functional block diagram for explaining an example of a learning unit according to the present invention. [Diagram 5] FIG. 5 is a flow chart for explaining the method for predicting a vital sign value according to the present invention. [Figure 6] FIG. 6 is a flow chart for explaining an example of the step of re-learning a prediction model in the present invention. [Figure 7] FIG. 7 is a flow chart for explaining an example of the steps for regenerating a predictive model in the present invention. [Figure 8] FIG. 8 is a diagram for explaining an example of time delay correction. [Figure 9] FIG. 9 is a diagram for explaining an example of a method for calibrating unit inconsistency. [Figure 10] FIG. 10 is a diagram for explaining another example of a method for calibrating unit inconsistency. [Figure 11] 11 and 12 are diagrams for explaining an example of determining whether or not to re-learn a prediction model. [Figure 12] 11 and 12 are diagrams for explaining an example of determining whether or not to re-learn a prediction model. [Figure 13] FIG. 13 is a diagram for explaining an example of regenerating a prediction model. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0042] It should be noted that the technical terms used in the present invention are merely used to describe a specific embodiment and are not intended to limit the present invention. In addition, unless otherwise specifically defined in the present invention, the technical terms used in the present invention should be interpreted in a way that is generally understood by a person having ordinary knowledge in the technical field to which the present invention belongs, and should not be interpreted in an overly comprehensive or overly narrow sense. In addition, when the technical terms used in the present invention are incorrect technical terms that cannot accurately express the idea of the present invention, they should be replaced and understood by technical terms that can be correctly understood by a person skilled in the art.
[0043] In addition, singular expressions used in the present invention include plural expressions unless the context clearly dictates differently. In the present invention, terms such as "comprise" or "include" should not be interpreted as including all of the components or steps described in the invention, but rather as including some components or steps that may not be included, or may further include additional components or steps.
[0044] In addition, it should be noted that the attached drawings are intended to facilitate understanding of the concept of the present invention, and should not be construed as limiting the concept of the present invention.
[0045] The method for predicting a vital sign value according to the present invention will now be described in more detail with reference to the accompanying drawings.
[0046] Hereinafter, a continuous blood glucose measurement system will be described which continuously measures biometric information indicating blood glucose levels through a body-attached unit attached to a user's body for a certain period of time and transmits the measured biometric information to a communication terminal. However, depending on the field to which the present invention is applied, the body-attached unit may measure various types of biometric information and transmit the measured biometric information to a communication terminal, which is within the scope of the present invention.
[0047] FIG. 1 is a schematic diagram showing a blood glucose measuring system according to an embodiment of the present invention.
[0048] Referring to FIG. 1, a blood glucose measuring system 1 according to an embodiment of the present invention includes a body-attachable unit 10 and a communication terminal 30.
[0049] The body-attachable unit 10 is attached to the body, and when the body-attachable unit 10 is attached to the body, one end of the sensor of the body-attachable unit 10 is inserted into the skin and continuously measures biological information indicating the user's blood glucose level using body fluids, etc., during the period the sensor is in use.
[0050] The communication terminal 30 is a terminal capable of receiving biometric information from the body-attached unit 10 and displaying the received biometric information to a user, and may be, for example, a mobile terminal such as a smartphone, tablet PC, or notebook that can communicate with the body-attached unit 10. Of course, the communication terminal 13 is not limited thereto, and may be any type of terminal that includes a communication function and can be installed with programs and applications.
[0051] The body-attached unit 10 transmits the measured biometric information to the communication terminal 30 at the request of the communication terminal 30 or at set times. For data communication between the body-attached unit 10 and the communication terminal 30, the body-attached unit 10 and the communication terminal 30 may be connected to each other by a wired connection such as a USB cable, or may be connected to each other by a wireless communication method such as infrared communication, NFC communication, Bluetooth (registered trademark), etc.
[0052] The communication terminal 30 predicts the user's future vital signs based on the received vital signs and provides the predicted vital signs to the user. Preferably, the communication terminal 30 can provide the user with a hyperglycemia or hypoglycemia alarm based on the predicted vital signs, or provide the user with a necessary prescription along with a hypoglycemia or hypoglycemia alarm.
[0053] Here, the communication terminal 30 can store the received biometric information for a certain period of time, and can generate a prediction model using the stored biometric information.
[0054] In addition, the communication terminal 30 can monitor and compare the predicted vital signs value for a certain time in the future with the actual vital signs value actually measured after a certain time has passed, and re-learn or re-generate the prediction model generated based on the prediction error between the predicted vital signs value and the actual vital signs value or based on the expression characteristics of the prediction error.
[0055] FIG. 2 is a functional block diagram for explaining the health value prediction device according to the present invention.
[0056] The biometric value prediction device according to the present invention can be implemented in a communication terminal, and more specifically, referring to Fig. 2, the transceiver 110 receives the biometric information of the user from the body-attached unit, and the storage unit 130 stores the received biometric information. Here, the storage unit 130 can map the received biometric information to a reception time and store it.
[0057] The biometric value determining unit 180 determines the biometric value of the user from the received biometric information and outputs the determined biometric value to the user or stores the determined biometric value in the storage unit 130 after mapping it with the reception time.
[0058] Here, various types of biometric information can be measured through a sensor inserted into the user's body for a certain period of time, and the biometric information below can be an example of biometric information indicating the user's blood glucose information, and the biometric value can be a blood glucose level determined from the biometric information.
[0059] The storage unit 130 can store biometric information received from the body-attached unit as well as biometric values determined from the biometric information, and the feature value generating unit 150 generates a feature value vector used to determine predicted biometric values at a future time point using the biometric information.
[0060] The prediction unit 170 determines a predicted vital sign value after a certain time has elapsed based on the current time point using the prediction model stored in the storage unit 130, and the prediction unit 170 determines the predicted vital sign value by applying the generated feature value vector to the prediction model. Here, the predicted vital sign value is a predicted vital sign value that the user will have after a certain time has elapsed based on the current time point, and the predicted vital sign value can be used to predict the user's vital sign value after a certain period of time in advance, and an alarm can be provided to the user based on the predicted vital sign value, or a necessary prescription can be provided.
[0061] For example, if the user's predicted blood sugar level three hours later is determined to be below 60 mg / dL based on the current time point, the user may be provided with a hypoglycemia alarm or a prescription to eat.
[0062] Meanwhile, the storage unit 130 stores the predicted vital sign value determined by the prediction unit 170 and the actual vital sign value determined by the vital sign value determination unit 180, and the learning unit 190 monitors the actual vital sign value determined after a certain time has elapsed and the predicted vital sign value of the user predicted before a certain time, and compares them with each other.
[0063] The learning unit 190 can re-learn the prediction model stored in the storage unit 130 based on the magnitude or ratio of the prediction error between the predicted vital signs value and the actual vital signs value, or can re-learn the prediction model from the beginning based on the expression characteristics of the prediction error to regenerate a new prediction model, and update the prediction model stored in the storage unit 130 with the generated new prediction model.
[0064] FIG. 3 is a functional block diagram for explaining an example of a feature value generating unit according to the present invention.
[0065] More specifically, referring to FIG. 3, the pre-processing unit 151 removes or reduces noise from the biometric information received through the transceiver unit.
[0066] The biological information measured through or received from the body-attachable unit may include noise. For example, since the sensor of the body-attachable unit is partially inserted into the human body, the sensor may move whenever the person moves, and such sensor movement may cause noise to be included in the biological information data measured by the body-attachable unit.
[0067] Also, when the biometric information is transmitted from the body-attached unit to the communication terminal, it may be affected by surrounding electromagnetic waves, etc., and as a result, noise may be included in the biometric information received by the communication terminal.
[0068] The pre-processing unit 151 performs an outlier processing filtering process on the received biometric information, or performs a low-pass filtering process on the biometric information after the singular value processing.
[0069] The first feature value extraction unit 153 extracts a first feature value from the preprocessed biometric information, and the second feature value extraction unit 157 extracts a second feature value from the calibrated biometric information calibrated by the calibration unit 155. The preprocessed biometric information received from the body-attached unit is current value information that varies differently depending on the biometric value of the user, and the calibrated biometric information means biometric information obtained by performing time delay calibration or unit mismatch calibration on the preprocessed biometric information in the calibration unit 155.
[0070] Preferably, the calibration unit 155 calibrates the pre-processed biometric information and provides the calibrated biometric information to the second feature value extraction unit 157, where the calibration unit calibrates the time delay of the pre-processed biometric information or calibrates the current value biometric information in units of biometric values, for example, blood glucose levels.
[0071] The feature vector value generating unit 159 generates a feature vector value by simply combining the first feature value and the second feature value, or by reducing and combining the first feature value and the second feature value using a method such as resource reduction.
[0072] FIG. 4 is a functional block diagram for explaining an example of a learning unit according to the present invention.
[0073] More specifically, referring to Fig. 4, the prediction error calculation unit 191 monitors the actual biometric value determined after a certain time has elapsed and the predicted biometric value of the user previously predicted before a certain time, compares them, and calculates a prediction error between the actual biometric value and the predicted biometric value. For example, the predicted biometric value after t time has elapsed based on the current time point is determined from the feature vector value, and calculates a prediction error between the actual biometric value determined using biometric information after t time has elapsed.
[0074] The prediction error characteristic determination unit 195 determines the expression characteristics of the prediction error, such as the number of times that the prediction error exceeds a critical value or a critical ratio during a unit time, or the number of times that the prediction error continuously exceeds a critical value or a critical ratio during a unit time.
[0075] The condition decision unit 193 decides whether the re-learning condition is satisfied based on whether the prediction error exceeds a set critical value or whether the ratio of the prediction error to the actual biometric value or the predicted biometric value exceeds a set critical ratio, or decides whether the condition for regenerating the prediction model is satisfied based on the expression characteristics of the prediction error.
[0076] If the re-learning conditions are satisfied, the re-learning unit 197 re-learns the prediction model stored in the storage unit and updates the prediction model with the re-learned prediction model, and if the re-generation conditions are satisfied, the generation unit 199 newly generates the prediction model stored in the storage unit and updates the prediction model with the newly generated prediction model.
[0077] FIG. 5 is a flow chart illustrating the method for predicting a vital sign value according to the present invention.
[0078] More specifically, referring to Fig. 5, the biometric information of the user is received (S110). Here, the biometric information of the user may be received from a body-attachable unit that is attached to the user's body for a certain period of time and continuously measures the biometric information of the user.
[0079] The communication terminal pre-processes the received biometric information (S130). More specifically, the pre-processing process searches for data outside a predetermined range among the received biometric information for singular value processing filtering, and processes the corresponding biometric information. At this time, if it is determined that the biometric information has a singular value, the corresponding biometric information can be removed and processed. However, the present invention is not limited to this, and the biometric information having a singular value can be calibrated and used as necessary. Low-band filtering can be performed on the biometric information processed by the singular value. Low-band filtering can remove components corresponding to a high band and leave only biometric information corresponding to a low band. Using the low-band filtered biometric information, an average value for the low-band filtered biometric information can be calculated and processed, or a cut average value can be used. Various methods of pre-processing are possible depending on the field to which the present invention is applied, and these are within the scope of the present invention.
[0080] A first feature value is extracted from the pre-processed biometric information (S140). The pre-processed biometric information may be a biometric value of the user measured by the body-attached unit, for example, a current value indicating a blood glucose level. Here, the first feature value may be a feature value such as difference, slope, deviation, average, effective value, or sharpness of the biometric information extracted by a statistical technique, or a feature value extracted by analyzing the biometric information in the frequency domain, such as a Fourier transform or a wavelet transform.
[0081] Meanwhile, the pre-processed biometric information is calibrated to generate calibrated biometric information (S150), and a second feature value is extracted from the calibrated biometric information (S150). Here, the pre-processed biometric information may be calibrated for time delay or unit inconsistency contained in the pre-processed biometric information before extracting the second feature value.
[0082] A time delay occurs between the time when the user's biometric information is actually measured by the body-attached unit and the time when the user's biometric information is received by the communication terminal. This time delay may be caused by the physical structure of the body-attached unit's sensor measuring and generating the biometric information from the user's body fluids, or may be caused by the calculation time required to generate the biometric information.
[0083] In addition, since the preprocessed vital information is current value information showing the vital value of the user, and not the vital value itself, the unit discrepancy must be calibrated from the current value to the actual vital value of the user. The unit discrepancy can be calibrated using a reference vital value measured through a blood sampling vital value measuring device using a separate sensor strip (blood measurement test paper) and collected body fluid. For example, if the current value of the vital information is 10nA and the reference vital value measured at this time is 100mg / dL, the calibration slope (A) is set to 10 and the slope is applied to the current value of the vital information to determine the user's vital value.
[0084] Depending on the field to which the present invention is applied, when calibrating the unit discrepancy, a weighting value can be assigned depending on whether the biometric value is increasing or decreasing, or how much difference there is between the measured biometric value and the reference biometric value, to calibrate the unit discrepancy.
[0085] A second feature value is extracted using the calibrated biometric information (S160). Here, the second feature value is a feature value such as difference, slope, deviation, average, effective value, or sharpness of the calibrated biometric information extracted by a statistical technique, or a feature value extracted by analyzing the calibrated biometric information in the frequency domain, such as by Fourier transform or wavelet transform.
[0086] A feature vector value is generated from the first and second feature values by simply combining the first and second feature values or by reducing the resources (S170).
[0087] The generated feature vector value is applied to a prediction model to generate a predicted vital sign value after a certain time (S190).
[0088] The prediction model is used to generate a predicted biometric value after a certain time period using the feature vector value. For example, if the number of components constituting the feature vector value (x) generated at a certain interval is 10, and the label (y) for each feature vector value is the predicted biometric value after a certain time period, the input (x) and label (y) of each feature vector value are expressed as in Equation 1 below.
[0089] [Mathematical formula 1]
number
[0090] Based on this, a feature vector value x=[x1.....x 10 ] into the prediction model, the predicted vital sign value y' from a certain time onwards is output.
[0091] Here, the prediction model may be generated using machine learning such as SVM (Support Vector Machine) and GMM (Gaussian Mixture Model), deep learning such as CNN (Convolution Neural Network) and RNN (Recurrent Neural Network), reinforcement learning such as model-free RL and model-based RL, and deep reinforcement learning such as DQN (Deep Q Network) by using the feature value vector of the user as training data. The prediction model generated in the present invention may be retrained using additional or new training data. Various well-known techniques may be used to generate the prediction model, and detailed description thereof will be omitted.
[0092] The first feature value and the second feature value have a large number of features. When trying to generate a prediction model using such data, the learning speed is slow and the performance is likely to be poor due to the large dimension of the data. For this reason, the first feature value and the second feature value can be simply combined, or features can be selected or reduced from among them to generate a prediction model. Projection and manifold learning, which are methods for reducing the dimension of data, and principal component analysis (PCA), which is a representative dimension reduction algorithm, can be used, or a feature selection algorithm such as Lasso can be used. Various well-known techniques can be used to generate a feature vector value from the first feature value and the second feature value, and detailed description thereof will be omitted.
[0093] In the present invention, a prediction model is generated by using all of the first and second feature values generated from the biometric information of a user, thereby making it possible to generate a prediction model that is personalized for the user and can accurately predict the biometric value of the user without using data of other surrounding users, and by applying a feature vector value generated from the first and second feature values to the prediction model, it is possible to accurately predict the biometric value of the user.
[0094] FIG. 6 is a flow diagram illustrating an example of the step of retraining a predictive model in the present invention.
[0095] More specifically, referring to FIG. 6, a prediction error is calculated from a difference between a predicted health value at a first prediction time point generated using a feature vector value and an actual health value measured using an actual health value at the first prediction time point (S211).
[0096] The prediction error is compared with a preset critical value or critical rate to determine whether the prediction error exceeds the critical value or critical rate (S213).
[0097] If the prediction error exceeds a critical value or a critical rate, it is determined whether a re-learning requirement is satisfied (S215). Here, the re-learning requirement may be satisfied if the prediction error exceeds a critical value or a critical rate, or if the prediction error exceeds a critical value or a critical rate and the average value of the prediction error for a preset first time period thereafter exceeds a critical average value, or if the prediction error exceeds a critical value or a critical rate and the prediction error continues to exceed a critical value or a critical rate for a second time period thereafter.
[0098] If the re-learning requirements are satisfied, the prediction model is re-learned and the previously used prediction model is updated with the re-learned prediction model (S217). Here, the re-learning of the prediction model is characterized in that the prediction model stored in the storage unit is generated using a future data set generated from the user's biometric information measured up to the present time in addition to the previous data set used.
[0099] Here, the data set is characterized in that it is a feature vector value generated from the first feature value and the second feature value.
[0100] FIG. 7 is a flow chart for explaining an example of the steps for regenerating a predictive model in the present invention.
[0101] More specifically, referring to Fig. 7, the prediction error expression characteristic is determined (S231). Here, the prediction error expression characteristic may be an average value of the prediction error during a unit time, the number of times that the prediction error exceeds a critical value or a critical rate during a unit time, the number of times that the prediction error continuously exceeds a critical value or a critical rate during a unit time, the proportion of time that the prediction error exceeds a critical value or a critical rate versus a unit time, etc.
[0102] It is determined whether a condition for regenerating the prediction model is satisfied based on the occurrence characteristics of the prediction error (S233). Here, the condition for regenerating the prediction model may be that the average value of the prediction error during the unit time exceeds a first critical average value, or the number of times that the prediction error exceeds a critical value or a critical rate during the unit time exceeds a first critical number, or the number of times that the prediction error continuously exceeds a critical value or a critical rate during the unit time exceeds a second critical number, or the proportion of time that the prediction error over the unit time exceeds a critical value or a critical rate exceeds a first critical rate, or a combination thereof.
[0103] If the conditions for regenerating the prediction model are satisfied, the prediction model is regenerated using the training data, and the previously used prediction model is updated with the regenerated prediction model (S235). Here, the regeneration of the prediction model is characterized in that a future data set generated from the user's biometric information measured up to the present time is used in addition to the previous data set used to generate the prediction model stored in the storage unit.
[0104] Here, the data set is characterized in that it is a feature vector value generated from the first feature value and the second feature value.
[0105] FIG. 8 is a diagram for explaining an example of time delay correction.
[0106] As shown in Fig. 8, a time delay occurs between the biometric information actually measured by the body-attached unit (shown by the dotted line) and the biometric information received by the communication terminal (shown by the solid line). This time delay can be calibrated using the well-known Kalman filter, ARIMA, etc.
[0107] FIG. 9 is a diagram for explaining an example of a method for calibrating unit inconsistency.
[0108] As shown in FIG. 9, in order to ensure accuracy, the biological information received from the body-attached unit in the continuous blood glucose measuring system must be calibrated for unit discrepancies using reference biological values at regular time intervals (t1, t2, t3, t4...), for example, every 12 hours or 24 hours.
[0109] When calibrating the unit mismatch, a weighting value can be assigned considering whether the biometric information (shown by a solid line) is rising or falling, and the calibration slope (A) can be calibrated by multiplying the assigned weighting value. Here, if the biometric information is rising, the weighting value can be assigned inversely proportional to the rate of rise, and if the biometric information is falling, the weighting value can be assigned proportionally to the rate of fall. For example, if the biometric information is rising, the weighting value can be assigned at a low value inversely proportional to the rate of rise (the higher the rate of rise, the lower the value, such as 0.90, 0.80, 0.70...), and if the biometric information is falling, the weighting value can be assigned at a high value in proportion to the rate of fall (the higher the rate of fall, the higher the value, such as 1.10, 1.20, 1.30...).
[0110] FIG. 10 is a diagram for explaining another example of a method for calibrating unit inconsistency.
[0111] As shown in FIG. 10, for accuracy, the biometric information received from the body-attached unit in the continuous blood glucose measuring system must be calibrated for unit discrepancies using reference biometric values at regular time intervals (t1, t2, t3, t4...), for example, every 12 hours or 24 hours.
[0112] The unit mismatch can be calibrated based on the difference between the reference biometric value and the measured biometric value (shown by the solid line) during calibration. Weights can be assigned to reduce the difference based on the difference between the reference biometric value and the measured biometric value (for example, weights can be calculated and assigned so that the calibration slope has the average value of the reference biometric value and the measured biometric value), or weights can be assigned to reduce the difference only when the difference between the reference biometric value and the measured biometric value falls outside a critical range.
[0113] 11 and 12 are diagrams for explaining an example of determining whether or not to re-learn a prediction model.
[0114] As shown in FIG. 11, when the prediction error (d1) between the predicted vital sign value predicted at the first prediction time (t1) and the measured vital sign value actually measured at the first prediction time falls outside a critical value or critical range, or when the prediction error (d2) between the predicted vital sign value predicted at the second prediction time (t2) and the measured vital sign value actually measured at the second prediction time falls outside a critical value or critical range, the prediction model can be re-trained at the first prediction time and the second prediction time, respectively.
[0115] Meanwhile, as shown in FIG. 12, if the average prediction error between the predicted vital signs value predicted for a certain time (td) and the measured vital signs value actually measured at the corresponding time exceeds the critical average value, the prediction model can be re-trained.
[0116] FIG. 13 is a diagram for explaining an example of regenerating a prediction model.
[0117] As shown in FIG. 13, the prediction model can be regenerated when the total number of times that the prediction error between the predicted vital sign value predicted for a certain unit time (tD) and the measured vital sign value actually measured at the corresponding time exceeds a critical value or critical range, or the number of times that the prediction error between the predicted vital sign value predicted for a certain unit time (tD) and the measured vital sign value actually measured at the corresponding time continuously exceeds a critical value or critical range exceeds a critical number, or when all of these combinations are satisfied.
[0118] Meanwhile, the above-described embodiments of the present invention can be created as a program that can be executed by a computer, and can be embodied in a general-purpose digital computer that runs the program using a computer-readable recording medium.
[0119] The computer-readable recording medium includes storage media such as magnetic storage media (e.g., ROM (Read Only Memory), floppy disks, hard disks, etc.), optically readable media (e.g., CD-ROM, DVD, etc.) and carrier waves (e.g., transmission over the Internet).
[0120] The present invention has been described with reference to the embodiments shown in the drawings, but these are merely illustrative, and those skilled in the art will appreciate that various modifications and equivalent embodiments are possible. Therefore, the true technical scope of the present invention should be determined by the technical ideas of the appended claims.
Claims
1. A method for predicting a biometric value of a user by using biometric information measured by a sensor, comprising: extracting a first feature value from the measured biometric information of the user; correcting the measured biometric information of the user and extracting a second feature value from the corrected biometric information; generating a feature vector value by reducing and combining the first feature value and the second feature value; and applying the generated feature vector value to a prediction model to predict the user's vital sign value.
2. 2. The method of claim 1, wherein the sensor is a sensor that is partially inserted into the user's body for a certain period of time to continuously measure the user's biological information.
3. The method for predicting biological information is The method further includes a step of pre-processing the measured biometric information by removing noise from the measured biometric information, The method for predicting a vital sign value according to claim 2 , wherein the first and second feature values are extracted from pre-processed vital sign information.
4. the first feature value is extracted directly from the pre-processed biometric information; The method of claim 3, wherein the second feature value is extracted from corrected biometric information generated by correcting time delay and unit inconsistency of pre-processed biometric information.
5. The unit discrepancy is 5. The method for predicting vital signs as claimed in claim 4, wherein the vital signs are corrected based on pre-processed vital signs or reference vital signs.
6. The unit discrepancy is The method of claim 5, wherein the pre-processed vital sign information is corrected by adding an ascending or descending weighting value.
7. The unit discrepancy is 6. The method of claim 5, wherein the health value is corrected by a weighting value assigned according to a difference between the health value determined from the measured health information and the reference health value.
8. The method for predicting a vital sign comprises: calculating a prediction error from a difference between a predicted vital sign value at a first prediction time point and a vital sign value actually measured at the first prediction time point; The method of claim 4, further comprising the step of determining whether to retrain the prediction model based on the prediction error.
9. The method of claim 8, wherein if the prediction error is greater than a critical value or a critical rate, the prediction model is retrained.
10. The method for predicting a vital sign comprises: The method of claim 8, further comprising determining whether to regenerate the prediction model based on an expression characteristic of a prediction error during a unit time.
11. The expression characteristic is The method of claim 10, wherein the number of times the prediction error continuously exceeds a critical value or a critical rate during the unit time and / or the total number of times the prediction error exceeds a critical value or a critical rate during the unit time are at least one of the following.
12. The method for predicting a vital sign comprises: The method for predicting vital signs according to claim 8, characterized in that the re-learning of the prediction model or the regeneration of the prediction model uses a subsequent data set generated from the user's vital signs information measured up to the present time in addition to the previous data set used to generate the prediction model.
Citation Information
Patent Citations
Fault discrimination and responsive processing method, system, and apparatus based on data and context
JP2020168437A
Systems and methods for patient monitoring using HCP-specific devices
JP2020516324A
Methods, systems, and devices for calibration and optimization of glucose sensors and sensor outputs
JP2020520264A
Systems, devices and methods for compensating for temperature effects on sensors
JP2021511094A
Automated analyte sensor calibration and error detection
JP2021528117A
