Method and system for learning blood pressure estimation model using light volume change signal

Abstract

One embodiment of the present invention relates to a system and a method for learning a blood pressure estimation model using a light volume change signal (PPG), the method for learning a blood pressure estimation model comprising the steps of: preprocessing raw data including light volume change signal data and blood pressure signal data collected from a subject; and constructing a blood pressure presumption model based on the preprocessed light volume change signal data and the blood pressure signal data, wherein the original data can comprise the blood pressure signal data of which the distribution degree is greater than a preset value.

Description

Technical Field

[0001] An embodiment of the present invention relates to a method and system for learning a blood pressure estimation model using a photoplethysmography signal. In particular, the present invention relates to a method and system for learning a blood pressure estimation model that can achieve highly reliable blood pressure estimation by learning data with a large deviation between subjects and data with a large deviation within a subject, even when the blood pressure variables of the subject are large. Background Art

[0002] Recently, with the development of medical technology, the aging of society, the westernization of lifestyle and diet have led to an increase in the incidence of hypertension. As a major indicator of kidney disease and serious cardiovascular diseases, hypertension is one of the most dangerous risk factors for death. Therefore, treatment and management are crucial.

[0003] To prevent, detect and treat hypertension, it is most important to continuously measure blood pressure in daily life. However, this goal has not been achieved yet.

[0004] Typically, blood pressure monitoring mainly uses a method of measuring blood pressure by wearing a cuff and based on the pressure change injected into the cuff. During the measurement process, the method using a cuff causes discomfort due to pressure, and even if a portable product is purchased, since it needs to include a cuff, there is a disadvantage that it is actually inconvenient to carry. The method using a cuff is not suitable for real-time blood pressure monitoring due to the disadvantages in terms of both convenience and portability. Therefore, there is a need to actively research a method for measuring blood pressure without a cuff without restraint.

[0005] In particular, recently, research on applying optical sensors of photoplethysmography (PPG) has been increasing continuously. Vascular elasticity information can be obtained based on photoplethysmography signals using eigenvalues such as percussion wave and tidal wave. Since the correlation between vascular elasticity information and blood pressure is very large, blood pressure can be estimated using it.

[0006] Another reason for measuring blood pressure through photoplethysmography (PPG) signals is that PPG optical sensors, as personal cuff-less blood pressure measurement devices, are of great significance. Recently, most wearable devices such as smart bracelets and watches are commonly equipped with PPG optical sensors. This means that most wearable devices can have the function of measuring blood pressure by simply installing a program, without the need to add new sensors. When measuring blood pressure through the PPG optical sensors of wearable devices, the existing blood pressure measurement method can be greatly improved in terms of portability and convenience. In terms of convenience, wearable devices such as smart bracelets and smart watches are worn on the wrist, which has the advantage that the wearer will not feel discomfort. To sum up, by using PPG signals, not only can blood pressure be predicted with high accuracy, but also the convenience and portability problems of existing devices can be solved by applying them to wearable devices.

[0007] Recently, research on predicting blood pressure based on PPG signals using artificial intelligence technology has been actively carried out because there is an incompletely clear correlation between photoplethysmography signal data and blood pressure. There have been attempts to clarify this unclear correlation through artificial intelligence. However, since the current learning-based partial systems are modeled and experimented in an "object-dependent" manner, there are limitations in accurately inferring the high-variable blood pressure of the examinee. Summary of the Invention

[0008] The present invention is generated under the above background conditions. The object of the present invention is to provide a learning method for a blood pressure estimation model that can highly reliably estimate blood pressure even in the case of high-variable blood pressure using a Convolutional Neural Network.

[0009] The object to be achieved by the present invention is not limited to the above-mentioned (multiple) objects. Those of ordinary skill in the technical field to which the present invention pertains can clearly understand other (multiple) objects not mentioned through the following description.

[0010] To this end, the learning method for a blood pressure estimation model using a Photoplethysmography (PPG) signal according to an embodiment of the present invention is executed in each step by a learning system for a blood pressure estimation model using artificial intelligence. The method includes the following steps: preprocessing the original data, where the original data includes photoplethysmography signal data and blood pressure signal data collected from an examinee; and constructing a blood pressure estimation model based on the preprocessed photoplethysmography signal data and blood pressure signal data, and the original data may include blood pressure signal data with a distribution degree of a preset value or more.

[0011] A blood pressure estimation model learning system using a Photoplethysmography signal (PPG) according to another embodiment of the present invention includes: a preprocessing unit for preprocessing raw data, where the raw data includes photoplethysmography signal data and blood pressure signal data collected from a subject; and a model construction unit for constructing a blood pressure estimation model based on the preprocessed photoplethysmography signal data and blood pressure signal data, and the raw data may include blood pressure signal data with a distribution degree above a preset value.

[0012] According to the present invention, since a learning model is constructed using blood pressure signal data with a distribution degree above a preset value, a highly reliable blood pressure estimation model can be constructed even in various environments where the blood pressure of the subject is unstable.

[0013] Moreover, the blood pressure estimation model according to an embodiment of the present invention is not dependent on an object, and thus can highly reliably estimate the highly variable blood pressure of a subject or the like.

[0014] The effects of the present invention are not limited to the above-mentioned effects, and those of ordinary skill in the technical field to which the present invention pertains can clearly understand other effects not mentioned through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 To briefly show the structural diagram of a blood pressure estimation model learning system using a photoplethysmography signal according to an embodiment of the present invention.

[0016] Figure 2 To illustrate the conceptual diagram of a blood pressure estimation model learning system using a photoplethysmography signal according to an embodiment of the present invention.

[0017] Figure 3 To illustrate the block diagram of the structure of a preprocessing unit according to an embodiment of the present invention.

[0018] Figure 4 To illustrate the conceptual diagram of a preprocessing method according to an embodiment of the present invention.

[0019] Figure 5 To illustrate the graph of the standard deviation of object calibration centering according to an embodiment of the present invention.

[0020] Figure 6 To illustrate the conceptual diagram of a blood pressure estimation model according to an embodiment of the present invention.

[0021] Figure 7 To illustrate the flowchart of the construction method of a blood pressure estimation model according to an embodiment of the present invention.

[0022] Figure 8 To illustrate the block diagram of the structure of a blood pressure estimation device according to an embodiment of the present invention. Detailed Implementation Modes

[0023] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. The structure and the effects thereof of the present invention can be understood through the following detailed description. Before describing the present invention in detail, even if the same structural elements are shown in different drawings, the same reference numerals are used as much as possible. When it is determined that the known structure may obscure the gist of the present invention, its specific description will be omitted.

[0024] Figure 1 To briefly show the structural diagram of a blood pressure estimation model learning system using a photovolume change signal according to an embodiment of the present invention, Figure 2 It is a conceptual diagram for explaining a blood pressure estimation model learning system using a photovolume change signal according to an embodiment of the present invention.

[0025] Referring to Figure 1 and Figure 2 , a blood pressure estimation model learning system 10 using a photovolume change signal according to an embodiment of the present invention may include a blood pressure estimation model learning server 100, a database 200, and an input / output device 300.

[0026] The blood pressure estimation model learning server 100 preprocesses the raw data measured from the examinee, and learns a model for estimating blood pressure based on the user's photovolume change signal by learning the preprocessed data. The raw data includes photovolume change signal (Photoplethysmography signal, hereinafter referred to as PPG) data and blood pressure signal data.

[0027] The raw data is divided into training data and validation data based on the examinee (subject-independent).

[0028] The blood pressure estimation model learning server 100 may include a preprocessing unit 110 and a model construction unit 130.

[0029] The preprocessing unit 110 preprocesses the photovolume change signal data and blood pressure signal data of the examinee pre-stored in the database 200. The preprocessing is to improve the learning time and learning effect of the learning result and improve the reliability of the learning result.

[0030] Explanation of Reference Numerals

[0031] 10: Blood pressure estimation model learning system

[0032] 100: Blood pressure estimation model learning server

[0033] 110: Preprocessing unit

[0034] 111: Abnormal object removal module

[0035] 112: Downsampling and Segmentation Module

[0036] 113: Abnormal Segment Removal Module

[0037] 114: Normalization Module

[0038] 115: Fragment Number Equalization Adjustment Module

[0039] 130: Model Construction Unit

[0040] 200: Database

[0041] 300: Input / Output Unit

[0042] Embodiments of the Invention

[0043] In one embodiment, the preprocessing unit 110 removes abnormal data. To remove abnormal data, when collecting the original data, additional information of the person being examined may be collected together. For example, the additional information of the person being examined may include information such as weight, height, age, whether pregnant, surgical time log, and electrocardiogram, etc.

[0044] In one embodiment, the preprocessing unit 110 may perform the following 5 steps for preprocessing the data.

[0045] First, remove abnormal objects from the original data.

[0046] Second, perform downsampling and segmentation of the original data.

[0047] Third, remove abnormal segments from the original data.

[0048] Fourth, perform normalization of the original data.

[0049] Fifth, adjust the balance of the number of segments.

[0050] Later, refer to Figure 3 and Figure 4 for a detailed description of the preprocessing method.

[0051] The model construction unit 130 learns a blood pressure estimation method based on the preprocessed data.

[0052] The model for learning the blood pressure estimation model may include two one-dimensional convolutional neural networks. Among them, one one-dimensional convolutional neural network extracts the temporal features of the PPG collected based on the original data, and the other one-dimensional convolutional neural network extracts morphological features based on the PPG difference. The convolutional neural network layer may be composed of multiple kernels, and the "rectified linear unit (ReLU)" may be used as the activation function.

[0053] Later, refer toFigure 5 and Figure 6 Describe a model structure and a learning method for learning a blood pressure estimation model.

[0054] The database 200 can be used to store information, programs, etc. required for the blood pressure estimation model learning server 100 to construct or learn a blood pressure estimation model using the photoplethysmogram signal, or to implement digitalization. For example, the database 200 stores the original data for constructing the blood pressure estimation model, or can store the data preprocessed by the blood pressure estimation model learning server 100. In addition, it can store the intermediate data generated for constructing the blood pressure estimation model by the blood pressure estimation model learning server 100 and the constructed blood pressure estimation model. Such a database 200 may include at least one of a storage, a database server, or a file server.

[0055] On the other hand, the original data of an embodiment of the present invention includes photoplethysmogram signal data and blood pressure signal data collected from the examinee.

[0056] The original data includes blood pressure signal data with a distribution degree above a preset value. Among them, the distribution degree depends on the standard deviation (SDS) of subject-calibration centering.

[0057] The standard deviation (SDS) of subject-calibration centering related to the blood pressure signal data is calculated by Equation 1.

[0058] Equation 1

[0059]

[0060] where Ni is the number of segments of subject i, S i,n is calculated by Equation 2, is calculated by Equation 3.

[0061] Equation 2

[0062] s i,n = x i,n - x i,c

[0063] where x i,n is the arterial blood pressure (ABP) of the nth segment of subject i, x i,cFor calibrating the arterial blood pressure of subject i. Artery Blood Pressure (ABP) refers to the hemodynamic index that guides clinicians for treatment arbitration, and it can measure the blood pressure borne by the arterial wall. It mainly measures the brachial artery. A systolic blood pressure of 120 mmHg or less is a normal value, and a diastolic blood pressure of 80 mmHg or less is a normal value.

[0064] Mathematical formula 3

[0065]

[0066] In one embodiment, blood pressure signal data with a distribution degree above a preset value is used as raw data. Thus, even in various environments where the blood pressure of the subject under examination is unstable, a blood pressure estimation model with high reliability can be constructed.

[0067] The input / output device 300 may include an input unit and an output unit. The input unit includes an input device through which a user can perform operations such as data input and data selection. The input device may include a general keyboard, a mouse, etc. When the input unit is a touch screen capable of touch input, it can also be integrated with the output unit.

[0068] The output unit refers to a structure that displays various information related to the operation of blood pressure estimation model learning according to the control of the blood pressure estimation model learning server 100.

[0069] For example, such an output unit can be a liquid crystal display device (LCE), a light-emitting diode (LED), an organic light-emitting diode (OLED), a projector, or other currently available, previously available, or future available display devices. For example, the output unit can display an interface page for providing information or an information providing result page.

[0070] Figure 3 It is a block diagram for explaining the structure of the preprocessing unit in an embodiment of the present invention. Figure 4 It is a conceptual diagram for explaining the preprocessing method in an embodiment of the present invention.

[0071] Refer to Figure 3 and Figure 4 , the preprocessing unit 110 may include an abnormal object removal module 111, a downsampling and segmentation module 112, an abnormal segment removal module 113, a normalization module 114, and a segment number equalization adjustment module 115.

[0072] The abnormal object removal module 111 removes abnormal objects for preprocessing data.

[0073] The abnormal object removal module 111 removes abnormal and duplicate data from the collected data. Among them, for example, the abnormal data is the abnormal data of the examinee under exceptional conditions, and the almost identical arterial blood pressure and photoplethysmogram signal data are removed.

[0074] Among them, the criteria C1-1, C1-2, and C1-3 for exceptional conditions may include the following three types.

[0075] The first criterion C1-1 for exceptional conditions may include the weight, height, and pregnancy status of the examinee. For example, a normal examinee who meets the first criterion is a person with 10 kg ≤ weight ≤ 100 kg, 100 cm ≤ height ≤ 200 cm, 18 years ≤ age ≤ 100 years, and not pregnant. That is, the data of the examinee who deviates from the normal examinee criteria can be removed as abnormal data.

[0076] The second criterion C1-2 for exceptional conditions is based on the necessary information of the examinee and may include the surgical time log, electrocardiogram, photoplethysmogram signal, systolic blood pressure (ART-SBP), diastolic blood pressure (ART-DBP), and mean blood pressure (ART-MBP).

[0077] The third criterion C1-3 for exceptional conditions can be noise. The abnormal object removal module 111 can remove the photoplethysmogram signal or arterial blood pressure waveform including noise.

[0078] That is, the abnormal object removal module 111 removes the data of the examinee who violates any one of the criteria C1-1, C1-2, and C1-3 of the exceptional conditions.

[0079] Next, the downsampling and segmentation module 112 performs downsampling and segmentation of the original data.

[0080] For example, the downsampling and segmentation module 112 downsamples the arterial blood pressure and photoplethysmogram signal data sampled at 500 Hz according to a preset first criterion. For example, after downsampling to 50 Hz, it is segmented into multiple segments composed of each preset second criterion. Among them, for example, the preset second criterion can be 500 points (that is, each segment is 10 seconds of data). In other variants, the segment may be segmented into 8-second lengths for designing ANN16 and LRCN24, or may be segmented into 10-second lengths for designing SVR.

[0081] The abnormal segment removal module 113 removes abnormal segments from the segmented segments. The abnormal segments may include invalid pulse counts, abnormal systolic / diastolic blood pressure variations, or segments with unspecified pulses. The arterial blood pressure segments with normal systolic blood pressure are 70 mmHg ≤ mean systolic blood pressure ≤ 180 mmHg, and segments deviating from this can be removed. The normalization module 114 is used to perform normalization.

[0082] The systolic blood pressure and diastolic blood pressure of the A-line are composed of the average of the systolic peak pressure and the diastolic blood pressure in each A-line pulse. The systolic blood pressure value and the diastolic blood pressure value can be normalized by the mean and standard deviation of the overall training set.

[0083] The segment number equalization adjustment module 115 is used to adjust the equalization of the segment numbers.

[0084] To adjust the equalization of the segment numbers, the segment number equalization adjustment module 115 can remove the normalization objects with the number of segments below the preset minimum number. And if there are segments with the number of objects above the preset maximum number, the segment number equalization adjustment module 115 can randomly select only 100 segments. Therefore, each object can include segments with a number above the minimum number and less than the maximum number. Among them, the minimum number is 50, and the maximum number can be 100, but it is not limited thereto. With the adjustment of the equalization of the segment numbers, all objects can have a fair impact on learning and verification.

[0085] Figure 5 It is a graph for explaining the standard deviation of object calibration centering in an embodiment of the present invention.

[0086] The blood pressure estimation learning model learns the characteristics of the dynamically changing photoplethysmography signal based on the blood pressure change for a new object. Therefore, as the number of objects used for modeling increases, the BP estimation accuracy based on the photoplethysmography signal will be improved.

[0087] In one embodiment, when the photoplethysmography signal samples of the same object are used for the modules of training and test data, the model may overfit to the object. Therefore, in one embodiment, a subject-independent data set is used. That is, the data set for training and the data set for testing are composed of different objects. And the hold-out method can be used for non-exhaustive cross-validation and testing. The hold-out method is a known method in the art, so the detailed description will be omitted.

[0088] On the other hand, when the blood pressure variation within the subject is low, the accuracy may overfit.

[0089] Refer to Figure 5 , Case A shows an example where the blood pressure deviation between subjects is high and the deviation within the subject is small. On the contrary, Case B shows an example where the blood pressure deviation between subjects is high and the deviation within the subject is also high.

[0090] In one embodiment of the present invention, it not only includes Case A, but also includes data with a relatively high blood pressure deviation among the tested persons shown in Case B.

[0091] Figure 6 It is a conceptual diagram for explaining the blood pressure estimation model according to one embodiment of the present invention.

[0092] Refer to Figure 6 , the model construction unit ( Figure 1 130) can learn using a learning model including two one-dimensional convolutional neural networks, one multilayer perceptron (MLP), and one fully connected layer (FCL).

[0093] The two one-dimensional convolutional neural networks have the same structure and parameters as the main feature extraction network. One one-dimensional convolutional neural network receives the target photoplethysmogram signal and extracts the time series features of the photoplethysmogram signal waveform using multiple filters. Another one-dimensional convolutional neural network of 1×500 receives the calibrated photoplethysmogram signal for training and extracts various features from the calibrated photoplethysmogram signal waveform using multiple filters.

[0094] Among them, the one-dimensional convolutional neural network includes four hidden convolutional neural network layer groups, an average pooling layer, and a dropout layer. Each hidden convolutional neural network layer group consists of a convolutional layer, a batch normalization layer, and a rectified linear unit (ReLU) layer. The batch normalization between the convolutional layer and the rectified linear layer normalizes the input of the hidden layer and solves the problem caused by the change of the input distribution. The rectified linear layer is used to achieve faster and better learning at the end of each hidden layer.

[0095] After the four hidden convolutional neural network layer groups, the waveform is sampled through the average pooling layer, which maintains the necessary information of the function and reduces the complexity of the network. 30% of the output data of the average pooling layer is discarded through the dropout layer by randomly removing 30% of the neurons during the training process (for example, set to 0). In the case where the dropout is set to 0, the hyperparameter dropout ratio is 0.3. Dropout reduces overfitting and improves generalization by preventing meaningless work from overly relying on specific inputs.

[0096] After the dropout layer, each batch passes through the fully connected layer in units of 8, and is normalized in the batch normalization layer to make the mean and distribution become 0 and 1 respectively in order to improve the convergence speed and learning performance.

[0097] The output sequences of the two one-dimensional convolutional neural networks and their absolute value differences are finally input into the fully connected layer module and activated through the rectified linear function.

[0098] The multi-layer perceptron is used to support feature extraction from the systolic blood pressure value and diastolic blood pressure value of Line A for supervised learning. The calibrated systolic blood pressure value and the calibrated diastolic blood pressure value are input into the multi-layer perceptron and are respectively provided to two fully connected layers. After each fully connected layer, a batch normalization layer and a rectified linear unit layer are processed in batches. The output of each rectified linear unit layer is input into a connection layer. The output of the connection layer is input into a fully connected layer, and the target systolic blood pressure and diastolic blood pressure are finally output.

[0099] The features output from two one-dimensional convolutional neural networks, the difference between them, and the multi-layer perceptron are connected. The single output sequence of the connection layer is provided to a fully connected layer, and a batch normalization layer and a rectified linear unit layer are processed in batches. The output of the rectified linear unit layer generates the target systolic blood pressure and diastolic blood pressure through other fully connected layers.

[0100] In the preprocessing process, the blood pressure estimation model of an embodiment of the present invention obtains the data of 4,185 examinees from 25,779 surgical cases in the preprocessing process. Among them, 80% is used as learning data, and 20% is used for hold-out validation for the performance evaluation of the model. And, in order to prevent the model from overfitting, 10% of the learning data is randomly selected as validation data. And, as the sphygmomanometer certification standard, the BHS and AAMI standards are used to verify the application possibility of the proposed model in medical devices.

[0101] Figure 7 It is a flowchart for explaining the construction method of the blood pressure estimation model of an embodiment of the present invention.

[0102] Refer to Figure 7 The construction method of the blood pressure estimation model described Figures 1 to 6 can be executed by the blood pressure estimation model learning system described

[0103] In step S110, the raw data is preprocessed. The raw data includes the photoplethysmogram signal data and blood pressure signal data collected from the examinee. The raw data includes blood pressure signal data with a distribution degree above a preset value. The distribution degree is the standard deviation of subject-calibration centring.

[0104] The standard deviation (SDS) of subject-calibration centring can be calculated by the mathematical formula 1, mathematical formula 2, and mathematical formula 3.

[0105] First, to preprocess the raw data, abnormal objects (subjects) that do not meet the predetermined conditions are removed from the raw data. Next, the raw data is downsampled and divided into multiple segments. Subsequently, predetermined abnormal segments are removed from the divided segments. Then, the raw data from which the abnormal segments have been removed is normalized. Finally, the balance of the number of segments is adjusted by removing normalized objects smaller than a preset minimum number and randomly selecting the maximum number of segments for objects above the preset maximum number.

[0106] In step S120, a blood pressure estimation model is constructed based on the preprocessed photoplethysmogram signal data and blood pressure signal data.

[0107] Figure 8 It is a block diagram for explaining the structure of a blood pressure estimation device according to an embodiment of the present invention.

[0108] Referring to Figure 8 , the blood pressure estimation device 20 according to an embodiment of the present invention can predict the abnormal blood pressure (e.g., low blood pressure or high blood pressure) of a patient by using the acquired photoplethysmogram signal data. For example, the blood pressure estimation device 20 can use the reference Figures 1 to 6 constructed blood pressure estimation model to predict the blood pressure of the patient.

[0109] In one embodiment, the blood pressure estimation device 20 can be a surgical monitoring server, a computer, or a medical device, and a dedicated program for setting the blood pressure estimation model and blood pressure prediction method can be installed. For example, the blood pressure estimation device 200 can include: a data acquisition unit 21 capable of acquiring photoplethysmogram signal data; a blood pressure estimation unit 22 that estimates blood pressure by using the acquired data and the blood pressure estimation model; and a database 23 capable of storing the blood pressure estimation model, blood pressure estimation values, etc. as big data.

[0110] The data acquisition unit 21 can perform A / D conversion on the photoplethysmogram signal acquired from the patient to obtain photoplethysmogram signal data. For example, the data acquisition unit 21 receives data from an external device or a user (e.g., medical staff), or can receive the photoplethysmogram signal from a photoplethysmogram signal sensor. The photoplethysmogram signal data is obtained from the waveform of the photoplethysmogram signal. The photoplethysmogram signal data can include characteristic values such as the duration of each waveform of the photoplethysmogram signal, the interval between each waveform, the amplitude of each waveform, and kurtosis. And, in addition to the waveform, the photoplethysmogram signal data can also include representative values such as the average value, the maximum value, or the minimum value.

[0111] The data acquisition unit 21 can sample the photoplethysmogram signal data as needed.

[0112] The blood pressure estimation unit 22 refers to Figures 1 to 7The constructed blood pressure estimation model takes in the photoplethysmogram signal data obtained from the data acquisition unit 21 and outputs a blood pressure estimation value.

[0113] On the other hand, although not shown, the blood pressure estimation device 20 may further include an electrocardiogram sensor.

[0114] In this case, it should be understood that each block in the flowchart and combinations of the flowchart can be executed by computer program instructions. Such computer program instructions can be loaded onto the processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device. Therefore, the instructions executed by the processor of the computer or other programmable data processing device can generate a means for performing the functions described in the (multiple) flowchart blocks. Such computer program instructions can also be stored in a computer-usable memory or computer-readable memory capable of directing a computer or other programmable data processing device, and can also produce a manufactured article including the instruction means for performing the functions described in the (multiple) flowchart blocks stored in the computer-usable memory or computer-readable memory. The computer program instructions can also be loaded onto the computer or other programmable data processing device. Therefore, a series of operating steps executed on the computer or other programmable data processing device can generate a process executed by the computer. Thus, the instructions executed by the computer or other programmable data processing device can also provide the steps for performing the functions described in the (multiple) flowchart blocks.

[0115] Moreover, each block may represent a module, a segment, or a part of code including one or more executable instructions for performing the (multiple) specific logical functions. And it should be noted that in multiple alternative instances, the functions mentioned in the blocks may occur in a different order. For example, substantially, two consecutively shown blocks may be executed simultaneously, or the blocks may also be executed in the reverse order according to the corresponding functions.

[0116] In this case, the term "~ part" used in this embodiment represents a software structural element or a hardware structural element such as a field programmable gate array (FPGA) or an application specific integrated circuit (ASIC), and the "~ part" performs a certain function. However, the meaning of the "~ part" is not limited to software or hardware. The "~ part" can be configured in an accessible storage medium and can also be regenerated on one or more processors. Therefore, as an example, the "~ part" can include structural elements such as software structural elements, object-oriented software structural elements, class structural elements, and task structural elements, processes, functions, attributes, procedures, subroutines, program code segments, drivers, firmware, microcode, circuits, data, databases, data structures, tables, arrays, and variables. The functions provided within the structural element and the "~ part" can be formed by combining a smaller number of structural elements and "~ parts", or can be further divided into additional structural elements and "~ parts". Moreover, the structural element and the "~ part" can also be regenerated on one or more central processors within a device or a secure multimedia card.

[0117] It should be understood that those of ordinary skill in the art to which the present invention pertains can implement the present invention through other specific embodiments without changing the technical idea or essential features of the present invention. Therefore, the embodiments described above are merely examples in all aspects and should not be construed as having a limiting meaning. Compared with the above detailed description, the scope of the present invention should be defined based on the appended claims, and all changes or variant embodiments derived from the meaning, scope, and equivalent concepts of the claims belong to the scope of the present invention.

[0118] On the other hand, although this specification and the drawings disclose preferred embodiments of the present invention and use specific terms, these are only used in their ordinary meanings to easily explain the technical content of the present invention and help understand the present invention, and do not limit the scope of the present invention. Obviously, in addition to the embodiments disclosed herein, those of ordinary skill in the art to which the present invention pertains can implement all variant embodiments based on the technical idea of the present invention.

[0119] Industrial Applicability

[0120] The present invention can be used in industries related to healthcare.

Claims

1. A method for learning a blood pressure estimation model using photoplethysmography (PPG) signals, where each step is executed by a blood pressure estimation model learning system using artificial intelligence. The method is characterized in that, Including the following steps: Preprocess the original data, where the original data includes photoplethysmogram (PPG) signal data and blood pressure signal data collected from the person to be examined; and Construct a blood pressure estimation model based on the preprocessed PPG signal data and blood pressure signal data, The original data includes blood pressure signal data with a distribution degree above a preset value.

2. The method for learning a blood pressure estimation model using PPG signals according to claim 1, characterized in that, The distribution degree is the standard deviation of object calibration centered.

3. The method for learning a blood pressure estimation model using PPG signals according to claim 2, characterized in that, The standard deviation of object calibration centered is calculated by Mathematical Formula 1: Mathematical Formula 1 where Ni is the number of segments of object i, S i,n is calculated by Mathematical Formula 2, is calculated by Mathematical Formula 3, Mathematical Formula 2 s i，n = x i，n -x i，c where x i,n is the arterial blood pressure of the n-th segment of object i, and x i,c is for calibrating the arterial blood pressure of object i Mathematical Formula 3 4. The method for learning a blood pressure estimation model using PPG signals according to claim 1, characterized in that, The step of preprocessing the original data includes the following steps: Remove abnormal objects that do not meet the predetermined conditions from the original data; Downsample the original data and divide it into multiple segments; Remove predetermined abnormal segments from the divided segments; Normalize the original data from which the abnormal segments have been removed; And Remove normalized objects smaller than the preset minimum number and randomly select segments with the maximum number for objects above the preset maximum number to adjust the balance of the number of segments.

5. The method for learning a blood pressure estimation model using PPG signals according to claim 1, characterized in that, In the step of constructing the blood pressure estimation model, a neural network including two one-dimensional convolutional neural networks, one multi-layer perceptron, and one fully connected layer is used.

6. The method for learning a blood pressure estimation model using PPG signals according to claim 5, characterized in that, The one-dimensional convolutional neural network is stacked by four convolutional neural networks, four convolutional neural networks, an average pooling layer, a fully connected layer, and a batch normalization layer 5, and a rectified linear unit layer 5.

7. The method for learning a blood pressure estimation model using PPG signals according to claim 1, characterized in that, The original data is divided into training data and validation data based on the person to be examined.

8. A blood pressure estimation model learning system using PPG signals, characterized in that, Including: A preprocessing unit for preprocessing the original data, where the original data includes PPG signal data and blood pressure signal data collected from the person to be examined; and A model construction unit for constructing a blood pressure estimation model based on the preprocessed PPG signal data and blood pressure signal data, The original data includes blood pressure signal data with a distribution degree above a preset value.

9. The blood pressure estimation model learning system using a photoplethysmogram signal according to claim 8, wherein, The distribution degree is the standard deviation of object calibration centered.

10. The blood pressure estimation model learning system using a photoplethysmogram signal according to claim 9, wherein, The standard deviation of object calibration centered is calculated by Mathematical Formula 1: Mathematical Formula 1 where Ni is the number of segments of object i, S i,n is calculated by Mathematical Formula 2, is calculated by Mathematical Formula 3, Mathematical Formula 2 S i，n = x i，n -x i，c where x i,n is the arterial blood pressure of the n-th segment of object i, and x i,c is for calibrating the arterial blood pressure of object i Mathematical Formula 3

Images