Blood pressure estimation model learning method and system using optical volume change signals

Abstract

The present invention provides a blood pressure estimation model training method using a photoplethysmography signal, which includes the steps of preprocessing raw data including photoplethysmography signal data and blood pressure signal data collected from a subject, and constructing a blood pressure estimation model based on the preprocessed photoplethysmography signal data and blood pressure signal data, wherein the raw data includes blood pressure signal data having a degree of variance equal to or greater than a predetermined value.

Description

[Technical Field]

[0001] The present invention relates to a blood pressure estimation model training method and system using a photoplethysmography signal, and relates to a blood pressure estimation model training method and system that learns data with a large degree of inter-subject deviation and data with a large intra-subject deviation, and estimates blood pressure with high reliability and accuracy even when there is a large inter-subject and intra-subject variance (variate) in blood pressure. [Background technology]

[0002] In recent years, the prevalence of hypertension has increased due to the aging of society, the Westernization of lifestyles and dietary habits, and the advancement of medical technology. Hypertension is a major indicator of kidney disease and serious cardiovascular disease, and is one of the most dangerous causes of death, so treatment and management are essential.

[0003] The most important thing for preventing, recognizing, and treating hypertension is to continuously monitor blood pressure in daily life, but this is not currently being done.

[0004] Traditionally, blood pressure has been monitored by wearing a cuff and measuring changes in the pressure injected into the cuff. The cuff-based method has drawbacks: it can cause discomfort during the measurement process, and even if a portable device is purchased, the cuff itself must be carried, making it difficult to actually carry around. Due to the drawbacks in both convenience and portability, the cuff-based method is not suitable for real-time blood pressure monitoring. Therefore, active research is being conducted into devices that can measure blood pressure without a cuff and without restraints.

[0005] In particular, research using photoplethysmography (PPG) optical sensors has been increasing recently because it is possible to obtain vascular elasticity information using characteristic values ​​such as percussion wave and tidal wave from optical volume change signals.Since vascular elasticity information has a high correlation with blood pressure, it can be used to estimate blood pressure.

[0006] Another reason for the increasing number of attempts to measure blood pressure using optical volumetric change signals is the significant value of PPG optical sensors as personal cuffless blood pressure measurement devices. Most wearable devices, such as smart bands and smart watches, are now equipped with PPG optical sensors. This means that most wearable devices can be equipped with blood pressure measurement functionality simply by installing a program, without the need for additional sensors. Measuring blood pressure using PPG optical sensors in wearable devices would significantly improve existing blood pressure measurement methods in terms of portability and convenience. Wearable devices, such as smart bands and smart watches, are worn on the wrist, eliminating significant inconvenience for the wearer. In summary, optical volumetric change signals can predict blood pressure with high accuracy, and their application to wearable devices can address the issues of convenience and portability of existing devices.

[0007] In recent years, there have been many attempts to predict blood pressure from optical volume change signals using artificial intelligence technology. This is because there exists an incomplete correlation between optical volume change signals and blood pressure. Attempts have been made to investigate this incomplete correlation using artificial intelligence. For example, Patent Document 1 discloses a method of inputting the difference between an ECG signal and a PPG signal into a neural network and estimating blood pressure based on temporal and morphological features. However, some existing learning-based systems have had limitations in that they cannot accurately estimate a subject's highly variable blood pressure because they perform modeling and experiments dependent on the subject. [Prior art documents] [Patent documents]

[0008] [Patent Document 1] Korean Patent Publication No. 10-2022-0105092 Summary of the Invention [Problem to be solved by the invention]

[0009] The present invention has been invented in light of the above-mentioned problems, and aims to provide a method and system for learning a blood pressure estimation model that is highly reliable even for blood pressures with high variability, using a convolutional neural network. [Means for solving the problem]

[0010] The blood pressure estimation model training method using a photoplethysmography signal (hereinafter also referred to as "PPG signal") of the present invention includes a step of preprocessing raw data including photoplethysmography signal data and blood pressure signal data collected from a subject, and a step of constructing a blood pressure estimation model based on the preprocessed photoplethysmography signal data and blood pressure signal data, wherein the raw data includes blood pressure signal data having a degree of variance equal to or greater than a predetermined value.

[0011] The blood pressure estimation model learning system using a light volume change signal of the present invention includes a preprocessing unit that preprocesses raw data consisting of light volume change signal data and blood pressure signal data collected from a subject, and a model construction unit that constructs a blood pressure estimation model based on the preprocessed light volume change signal data and blood pressure signal data, wherein the raw data includes blood pressure signal data whose degree of variance is equal to or greater than a predetermined value. [Effects of the Invention]

[0012] According to the present invention, a learning model is constructed using blood pressure signal data whose degree of variance is equal to or greater than a predetermined value, making it possible to construct a highly reliable blood pressure estimation model even in various environments where the subject's blood pressure is unstable.

[0013] Furthermore, the blood pressure estimation model constructed according to the present invention is independent of the subject, and therefore can estimate blood pressure with high reliability even for highly variable blood pressures of the subject. [Brief explanation of the drawings]

[0014] [Figure 1] 1 is a diagram illustrating a schematic configuration of a blood pressure estimation model learning system using a photovoltaic volume change signal according to an embodiment of the present invention. [Figure 2] 1 is a conceptual diagram illustrating a blood pressure estimation model learning system using a photovoltaic volume change signal according to an embodiment of the present invention. FIG. [Figure 3] FIG. 2 is a block diagram illustrating a configuration of a preprocessing unit according to an embodiment of the present invention. [Figure 4] FIG. 2 is a conceptual diagram illustrating a preprocessing method according to an embodiment of the present invention. [Figure 5] FIG. 10 is a diagram illustrating the standard deviation of subject-calibration centering according to an embodiment of the present invention. [Figure 6] FIG. 1 is a conceptual diagram for explaining a blood pressure estimation model according to an embodiment of the present invention. [Figure 7] 1 is a flowchart illustrating a method for constructing a blood pressure estimation model according to an embodiment of the present invention. [Figure 8] 1 is a block diagram illustrating a configuration of a blood pressure estimation device according to an embodiment of the present invention. BEST MODE FOR CARRYING OUT THE INVENTION

[0015] Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. The configuration and effects of the present invention will be clearly understood through the following detailed description. Prior to the detailed description of the present invention, identical components will be designated by the same reference numerals as much as possible even if they are shown in different drawings, and detailed descriptions of well-known components will be omitted if it is determined that they may obscure the gist of the present invention.

[0016] FIG. 1 is a diagram illustrating the configuration of a blood pressure estimation model learning system using a light volume change signal according to one embodiment of the present invention, and FIG. 2 is a conceptual diagram illustrating the blood pressure estimation model learning system using a light volume change signal according to one embodiment of the present invention.

[0017] The blood pressure estimation model learning system 10 using an optical volume change signal may include a blood pressure estimation model learning server 100, a database 200, and an input / output device 300.

[0018] The blood pressure estimation model learning server 100 preprocesses raw data consisting of photovolume change signal data and blood pressure signal data measured from a subject, learns the preprocessed raw data, and learns a blood pressure estimation model from the user's photovolume change signal.

[0019] The raw data is used for training and validation. However, it is desirable that the raw data of the same subject is not used for training and validation (subject-independent). This is because a blood pressure estimation model with a subject-dependent data set will be overfitted to that subject. Therefore, it is desirable to use a combination of raw data from different subjects as training data and validation data, for example, raw data from subject A as training data and raw data from subject B as validation data.

[0020] The blood pressure estimation model learning server 100 may include a preprocessing unit 110 and a model construction unit 130.

[0021] The preprocessing unit 110 preprocesses the photovolume change signal data and blood pressure signal data of the subject that are pre-stored in the database 200. The preprocessing is performed to improve the learning time and learning effect of the learning results and to increase the reliability of the learning results.

[0022] In this embodiment, the pre-processing unit 110 removes abnormal data. To remove abnormal data, additional information about the subject must be collected along with the raw data. The additional information about the subject may include, for example, weight, height, age, pregnancy status, surgery time log, and electrocardiogram.

[0023] In this embodiment, the preprocessing unit 110 can perform the following five steps to preprocess the raw data.

[0024] First, abnormal raw data is removed from the raw data.

[0025] Second, the remaining raw data from which the abnormal raw data has been removed is downsampled and segmented.

[0026] Third, anomalous segments are removed from the downsampled and segmented raw data.

[0027] Fourth, normalization of the raw data is performed, removing outlier segments.

[0028] Fifth, balance the number of segments.

[0029] A specific pre-processing method will be described with reference to FIGS.

[0030] The model construction unit 130 learns how to estimate blood pressure based on the pre-processed photovoltaic signal data and blood pressure signal data.

[0031] The model used to train the blood pressure estimation model includes two 1D-CNNs. One 1D-CNN extracts temporal features from PPG signals collected as raw data, and the other 1D-CNN extracts morphological features from the difference between PPG signals. The CNN layer consists of multiple kernels and uses the ReLU (rectified linear unit) as the activation function.

[0032] The structure of the model used to train the blood pressure estimation model and the training method will be described with reference to FIGS.

[0033] The database 200 stores or databases information, programs, etc. required for the blood pressure estimation model learning server 100 to construct or learn a blood pressure estimation model using a light volume change signal. For example, the database 200 can store raw data for constructing a blood pressure estimation model, or store light volume change signal data and blood pressure signal data preprocessed by the blood pressure estimation model learning server 100. It can also store intermediate data generated by the blood pressure estimation model learning server 100 to construct a blood pressure estimation model, and the constructed blood pressure estimation model. The database 200 includes at least one of a storage, a DB server, and a file server.

[0034] Meanwhile, the raw data according to this embodiment includes photovoltaic signal data and blood pressure signal data collected from the subject.

[0035] The raw data includes blood pressure signal data whose degree of dispersion is equal to or greater than a set value, where the degree of dispersion is defined by the standard deviation (SDS) of the subject-calibration centering.

[0036] The subject-calibration centered standard deviation (SDS) for the blood pressure signal data is calculated using Equation 1.

[0037]

number

[0038] where Ni is the number of segments for subject i, and Si,n is calculated according to Equation 2: JPEG2025536701000003.jpg87 is calculated using Equation 3.

[0039]

number

[0040] Here, xi,n is the arterial blood pressure (hereinafter also referred to as "ABP") of the nth segment of subject i, and xi,c is the ABP used for calibration of subject i. ABP is a hemodynamic index that guides clinicians in providing therapeutic intervention. It is the pressure of blood exerted against the arterial walls and is measured primarily in the brachial artery. Normal values ​​are considered to be systolic blood pressure below 120 mmHg and diastolic blood pressure below 80 mmHg.

[0041]

number

[0042] By using blood pressure signal data with a degree of variance equal to or greater than a predetermined value as raw data, it becomes possible to construct a highly reliable blood pressure estimation model even in various environments where the subject's blood pressure is unstable.

[0043] The input / output device 300 may include an input unit and an output unit. The input unit includes input means that allows a user to operate, such as inputting or selecting data. The input means may include a general keypad, a mouse, etc. If the input unit is configured as a touch screen that allows touch input, it may be configured integrally with the output unit.

[0044] The output unit is configured to display various information related to the learning operation of the blood pressure estimation model under the control of the blood pressure estimation model learning server 100.

[0045] The output unit may be implemented by, for example, a liquid crystal display (LCD), a light emitting diode (LED), an organic light emitting diode (OLED), a projector, or any other display device currently available, previously available, or future available. The output unit may display, for example, an interface page for providing information or an information provision result page.

[0046] FIG. 3 is a block diagram illustrating the configuration of a preprocessing unit according to an embodiment of the present invention, and FIG. 4 is a conceptual diagram illustrating a preprocessing method according to an embodiment of the present invention.

[0047] The preprocessing unit 110 may include an outlier subject removal module 111 , a downsampling and segment execution module 112 , an outlier segment removal module 113 , a normalization module 114 , and a segment number balancing module 115 .

[0048] The abnormal subject removal module 111 removes abnormal and duplicated raw data from the raw data collected from the subjects for preprocessing of the raw data. Here, the abnormal and duplicated raw data includes, for example, data from subjects that fall under "exceptional conditions" or nearly identical ABP data and PPG signal data.

[0049] Here, the "exceptional conditions" consist of the following three criteria (C1-1, C1-2, C1-3).

[0050] The first criterion C1-1 for the "exceptional condition" may include the subject's weight, height, and whether or not the subject is pregnant. For example, a normal subject based on the first criterion is one whose weight is 10 kg or less, whose height is 100 cm or less, whose age is 18 years or less, and whose age is 100 years or less, and who is not pregnant. In other words, raw data of a subject that does not meet the criteria for a normal subject can be removed as abnormal data.

[0051] The second criterion C1-2 for "exceptional condition" is based on the essential information of the subject, which may include the operation time log, electrocardiogram, PPG signal, ART-SBP (systolic blood pressure), ART-DBP (diastolic blood pressure), and ART-MBP (mean blood pressure). That is, the raw data of the subject that deviates from these normal subject standards can be removed as abnormal raw data.

[0052] The third criterion C1-3 for "exceptional conditions" is noise. Raw data from subjects with noisy PPG signals or ABP waveforms can be removed as abnormal raw data.

[0053] That is, the abnormal subject removal module 111 removes raw data of subjects who meet any one of the above-mentioned criteria C1-1, C1-2, and C1-3 of the "exceptional condition."

[0054] Next, the downsampling and segmentation module 112 performs downsampling and segmentation on the raw data.

[0055] The downsampling and segment execution module 112 downsamples each data of the ABP and PPG signals, for example, sampled at 500 Hz, to a predetermined first standard, for example, 50 Hz, and then divides the data into multiple segments, each configured according to a predetermined second standard.

[0056] Here, the preset second criterion may be, for example, 500 points (i.e., 10 seconds of data per segment). In other embodiments, the segments may be divided into 8-second segments for designing the ANN 16 and LRCN 24, or into 10-second segments for designing the SVR.

[0057] The abnormal segment removal module 113 removes abnormal segments from the segmented segments. Abnormal segments may include segments with invalid pulse rates, abnormal SBP / DBP fluctuations, or irregular pulses. Since ABP segments with normal SBP are 70 mmHg≦mean SBP≦180 mmHg, segments outside this range can be removed. The normalization module 114 performs normalization.

[0058] A-line SBP and DBP consist of the mean peak systolic and end-diastolic pressures at each A-line pulse. SBP and DBP values ​​are normalized to the mean and SD of the entire training set.

[0059] The segment number balance adjustment module 115 adjusts the balance of the number of segments.

[0060] To balance the number of segments, the segment number balancing module 115 can remove normalized subject data (ABP and PPG signal data) containing fewer than a preset minimum number of segments. Furthermore, if a subject's data contains more than a preset maximum number of segments, the segment number balancing module 115 randomly selects only 100 segments. Therefore, each subject's data contains more than the minimum number of segments and less than the maximum number of segments. Here, the minimum number may be 50, and the maximum number may be 100, but is not limited thereto. Balancing the number of segments ensures that all subjects' data have a fair impact on training and validation.

[0061] FIG. 5 is a diagram illustrating the standard deviation of subject-calibration centering according to an embodiment of the present invention.

[0062] Since the blood pressure estimation model learns PPG features that change dynamically with BP changes for each new subject, the accuracy of PPG signal-based BP estimation improves with increasing number of subjects used for modeling.

[0063] If PPG signal samples from the same subject are used for both training and validation data, the blood pressure estimation model will be overfitted to the subject. In this case, a data set independent of a specific subject will be used. Therefore, in this embodiment, the training data set used for training and the validation data set used for validation are composed of data from different subjects. In addition, a holdout method can be used for non-exhaustive cross-validation and testing. Since the holdout method is a well-known method, a detailed description will be omitted.

[0064] On the other hand, when intrasubject BP variability is low, accuracy performance is overqualified.

[0065] Case A in Figure 5 shows high between-subject BP variation but small within-subject variation, whereas case B not only shows high between-subject BP variation but also high within-subject variation.

[0066] In this example, data including not only case A but also case B having high BP deviation within the subject is included.

[0067] FIG. 6 is a conceptual diagram for explaining a blood pressure estimation model according to an embodiment of the present invention.

[0068] Referring to FIG. 6, the model construction unit (130 in FIG. 1) can perform training using a model including two 1D-CNNs, one MLP (multilayer perceptron), and one FCL (fully connected layer).

[0069] The two 1D-CNNs have the same structure and parameters as the main feature extraction network. One 1D-CNN receives the target PPG signal and uses multiple filters to extract time-series features from the PPG signal waveform. The other 1x500 1D-CNN receives the calibration PPG signal for training and uses multiple filters to extract various features from the calibration PPG signal waveform.

[0070] Here, the 1D-CNN includes four hidden CNN layer groups, an average pooling layer, and a dropout layer. Each hidden CNN layer group consists of one convolutional layer, a batch normalization layer, and a rectified linear unit (ReLU) layer. The batch normalization between the convolutional layer and the ReLU layer normalizes the hidden layer input and solves the problem caused by the change in input distribution. The ReLU layer is used at the end of each hidden layer for faster learning.

[0071] After the four hidden CNN layer groups, the waveform is sampled through an average pooling layer, which maintains essential information of the features and reduces the complexity of the network. 30% of the output data from the average pooling layer is dropped (e.g., set to 0) in the dropout layer, which randomly removes 30% of the neurons during training. When dropout is set to 0, the dropout rate in the hyperparameters becomes 0.3. Dropout prevents meaningless behavior from being too dependent on specific inputs, reducing overfitting and improving generalization.

[0072] After the dropout layer, each batch passes through the FCL in units of 8, and is normalized in a batch normalization layer so that the mean and variance are 0 and 1, respectively, to improve the convergence speed and learning performance.

[0073] The output sequences of the two 1D-CNNs and the difference between the absolute values ​​of these two are provided as inputs to the final FCL module, which is activated by the ReLU function.

[0074] The MLP is used to support feature extraction for supervised learning from the A-line SBP and DBP values. The calibration SBP and DBP values ​​are input to the MLP and provided to two FCLs, respectively. A batch normalization layer and a ReLU layer are placed after each FCL. The output of each ReLU layer is input to a connection layer. The output of the connection layer is input to the FCL, and finally the target SBP and DBP are output.

[0075] The features output from the two 1D-CNNs, their differences, and the MLP are connected. The single output sequence of the connection layer is provided to the FCL, where a batch normalization layer and a ReLU layer are placed. The output of the ReLU layer is passed through another FCL to generate the target SBP and DBP.

[0076] For the blood pressure estimation model of this example, data for 4,185 subjects from 25,779 surgical cases was obtained during the preprocessing process. 80% of the data was used as training data, and 20% was used for holdout validation to evaluate the model's performance. To prevent overfitting of the model, 10% of the training data was randomly selected and used as validation data. Furthermore, the feasibility of incorporating the proposed model into medical devices was verified using the BHS and AAMI standards, which are certification standards for blood pressure monitors.

[0077] FIG. 7 is a flowchart illustrating a method for constructing a blood pressure estimation model according to an embodiment of the present invention.

[0078] The method for constructing the blood pressure estimation model described with reference to FIG. 7 is performed by the blood pressure estimation model learning system described with reference to FIGS.

[0079] In step S110, raw data including optical volume change signal data and blood pressure signal data collected from a subject is preprocessed. The raw data includes blood pressure signal data whose degree of dispersion is equal to or greater than a predetermined value. The degree of dispersion is the standard deviation of the subject-calibration centering.

[0080] The standard deviation (SDS) of the subject-calibration centering is calculated using Equations 1, 2, and 3 above.

[0081] To preprocess the raw data, first, raw data of abnormal subjects that do not meet predefined conditions is removed from the raw data. Next, the raw data is downsampled and divided into multiple segments. After that, predefined abnormal segments are removed from the divided segments. Next, the raw data from which the abnormal segments have been removed is normalized. Finally, raw data of normalized subjects below a predetermined minimum number are removed, and for raw data of subjects above a predetermined maximum number, the maximum number of segments is randomly selected to balance the number of segments.

[0082] In step S120, a blood pressure estimation model is constructed based on the pre-processed photovoltaic volume change signal data and blood pressure signal data.

[0083] FIG. 8 is a block diagram illustrating the configuration of a blood pressure estimation device according to one embodiment of the present invention.

[0084] The blood pressure estimation device 20 can predict abnormal blood pressure (e.g., hypotension or hypertension) of a patient using the acquired PPG signal data. For example, the blood pressure estimation device 20 can predict the patient's blood pressure using a blood pressure estimation model constructed with reference to Figures 1 to 6, which uses the PPG signal data as an input value.

[0085] The blood pressure estimation device 20 may be a surgical monitoring server, a computer, or a medical device, and may have installed thereon a dedicated program for setting a blood pressure estimation model and a blood pressure prediction method. For example, the blood pressure estimation device 200 may include a data acquisition unit 21 for acquiring PPG signal data, a blood pressure estimation unit 22 for estimating blood pressure using the acquired data and a blood pressure estimation model, and a database 23 for storing the blood pressure estimation model, estimated blood pressure values, etc. as big data.

[0086] The data acquisition unit 21 can acquire PPG signal data by A / D converting the PPG signal acquired from the patient. For example, the data acquisition unit 21 can receive data from an external device, input from a user (e.g., medical staff), or a PPG signal from a PPG optical sensor. The PPG signal data can be acquired from the waveform of the PPG signal. It can include characteristic values ​​such as the time each waveform of the PPG signal is maintained, the interval between each waveform, the amplitude of each waveform, and kurtosis. In addition to the waveform, the PPG signal data can also include representative values ​​such as the average value, maximum value, or minimum value.

[0087] The data acquisition unit 21 can sample each piece of PPG signal data as needed.

[0088] The blood pressure estimation unit 22 inputs the PPG signal data acquired from the data acquisition unit 21 into the blood pressure estimation model constructed with reference to FIGS. 1 to 7, and outputs an estimated blood pressure value.

[0089] On the other hand, although not shown, the blood pressure estimation device 20 may further include an electrocardiogram sensor.

[0090] Each block of the process flow diagram and combinations thereof may be implemented by a computer program. These computer program instructions may be loaded into a processor in a general-purpose computer, special-purpose computer, or other programmable data processing device, such that the instructions, executed by the processor of the computer or other programmable data processing device, generate means for performing the functions described in the flow diagram blocks. These computer program instructions may be stored in a computer-usable or computer-readable memory that can direct the computer or other programmable data processing device to implement the functions in a particular manner, and the instructions stored in the computer-usable or computer-readable memory may produce an article of manufacture containing instruction means for performing the functions described in the flow diagram blocks. The computer program instructions may also be loaded onto a computer or other programmable data processing device, such that a series of operational steps are performed on the computer or other programmable data processing device to create a computer-implemented process, and the instructions that cause the computer or other programmable data processing device to provide steps for performing the functions described in the flow diagram blocks.

[0091] Each block may represent a module, segment, or portion of code that includes one or more executable instructions for performing the specified logical function. In some alternative embodiments, the functions shown in each block may occur out of order. For example, two blocks shown in succession may actually be performed substantially simultaneously, or the blocks may sometimes be performed in reverse order depending on the relevant function.

[0092] The term "module" as used herein refers to software or hardware components such as FPGAs or ASICs. A "module" performs a function, but is not limited to software or hardware. A "module" may be configured to reside on an addressable storage medium or to implement one or more processors. Thus, by way of example, a "module" includes components such as software components, object-oriented software components, class components, and task components, as well as processes, functions, attributes, procedures, subroutines, program code segments, drivers, firmware, microcode, circuits, data, databases, data structures, tables, arrays, and variables. The functionality provided within a component or module may be combined into fewer components and modules or further separated into additional components and modules. Furthermore, a component or module may be embodied to implement one or more CPUs within a device or security multimedia card.

[0093] A person skilled in the art to which the present specification pertains can implement the present specification in other specific forms without changing the technical idea or essential features thereof. Therefore, these examples are illustrative in all respects and are not limiting.

[0094] Meanwhile, the present specification and drawings disclose preferred embodiments of the present specification, and even if specific terms are used, they are used in general terms to simply explain the technical content of the specification and to aid in understanding the invention, and are not intended to limit the scope of the specification. It is obvious to those skilled in the art to which the present specification pertains that other modifications based on the technical ideas of the present specification can be implemented in addition to the embodiments disclosed herein. [Explanation of symbols]

[0095] 10 Blood Pressure Estimation Model Learning System 100 Blood Pressure Estimation Model Learning Server 110 Pretreatment section 111 Abnormal Subject Removal Module 112 Downsampling and Segmentation Execution Module 113 Abnormal Segment Removal Module 114 Normalization Module 115 Segment Balancing Module 130 Model Construction Department 200 databases 300 Input / output section

Claims

1. A blood pressure estimation model learning method using a photoplethysmography signal, comprising: pre-processing raw data collected from the subject, including optical volume change signal data and blood pressure signal data; and constructing a blood pressure estimation model based on the pre-processed light volume change signal data and the blood pressure signal data; A blood pressure estimation model learning method, characterized in that the raw data includes blood pressure signal data whose degree of variance is equal to or greater than a predetermined value.

2. 2. The method of claim 1, wherein the degree of dispersion of the blood pressure signal data is a standard deviation of subject-calibration centering.

3. The subject-calibration centering standard deviation (SDS) is calculated according to Equation 1: [Equation 1] where N is the number of segments for subject i, and S is calculated according to Equation 2: 【number】 is calculated by Equation 3 [Equation 2] [Equation 3] where x is the ABP of the nth segment for subject i, and x is the ABP used for calibration for subject i.

3. The blood pressure estimation model learning method using a photovoltaic volume change signal according to claim 2.

4. The step of pre-processing the raw data includes: removing raw data of abnormal subjects that do not meet predefined conditions from the raw data; downsampling and dividing the raw data into segments; removing predefined abnormal segments from the divided segments; normalizing the raw data from which the outlier segments have been removed; and removing raw data of normalized subjects that is less than a predetermined minimum number, and adjusting the balance of the number of segments for raw data of subjects that is equal to or greater than a predetermined maximum number by randomly selecting the maximum number of segments.

5. The step of constructing the blood pressure estimation model includes: The blood pressure estimation model training method using a light volume change signal according to claim 1, characterized in that a neural network including two 1D-CNNs, one MLP (multilayer perceptron), and one FCL (fully connected layer) is used.

6. 6. The blood pressure estimation model training method using a light volume change signal according to claim 5, wherein the 1D-CNN is formed by stacking four CNNs, an average pooling layer, an FCL, a batch layer 5, and a ReLU layer 5.

7. 2. The blood pressure estimation model learning method using optical volume change signals according to claim 1, wherein the raw data is used as training data and validation data, and the training data and validation data are raw data from different subjects.

8. A blood pressure estimation model learning system using a photoplethysmography signal, comprising: a preprocessing unit that preprocesses raw data including optical volume change signal data and blood pressure signal data collected from the subject; a model constructing unit that constructs a blood pressure estimation model based on the preprocessed photovoltaic volume change signal data and the blood pressure signal data, A blood pressure estimation model learning system, wherein the raw data includes blood pressure signal data whose degree of dispersion is equal to or greater than a predetermined value.

9. 9. The blood pressure estimation model learning system using an optical volume change signal according to claim 8, wherein the degree of dispersion of the blood pressure signal data is a standard deviation of subject-calibration centering.

10. The subject-calibration centering standard deviation (SDS) is calculated according to Equation 1: [Equation 1] where N is the number of segments for subject i, and S is calculated according to Equation 2: 【number】 is calculated by Equation 3 [Equation 2] [Equation 3] where x is the ABP of the nth segment for subject i, and x is the ABP used for calibration for subject i.

10. The blood pressure estimation model learning system using a light volume change signal according to claim 9.

Images