KR20210156706A - Apparatus and method for generating blood pressure estimation model and apparatus for estimating blood pressure - Google Patents

Abstract

According to an aspect, an apparatus for generating a blood pressure estimation model includes: a signal acquisition unit which receives a user's first signal and a second signal; and a processor which obtains a pulse transit time (PTT) as a first predictive variable value based on the first signal and the second signal, extracts a feature from the second signal to obtain a second predictive variable value based on the extracted feature, and generates a blood pressure estimation model based on the first predictive variable value and the second predictive variable value.

Description

Apparatus and method for generating a blood pressure estimation model, and apparatus for estimating blood pressure

It relates to a technique for estimating blood pressure without a cuff using a ballistocardiogram (BCG) signal.

Healthcare technology is receiving a lot of attention due to social problems such as rapid entry into an aging society and consequent increase in medical expenses. Accordingly, not only medical devices that can be used in hospitals or examination institutions, but also small medical devices that can be carried by individuals are being developed. In addition, such a small medical device is worn by a user and spread in the form of a wearable device that can directly measure a cardiovascular health state such as blood pressure, so that a user can directly measure and manage a cardiovascular health state. making it possible Recently, many studies have been conducted on a method of estimating biometric information by analyzing biosignals in order to increase the accuracy of biometric information estimation and at the same time to miniaturize the device.

An apparatus and method for generating a blood pressure estimation model through pulse transit time (PTT) and pulse wave analysis (PWA) analysis using BCG signals, and a technique for estimating blood pressure without a cuff using the blood pressure estimation model are presented.

According to an aspect, the apparatus for generating a blood pressure estimation model includes a signal acquisition unit receiving a user's first signal and a second signal, and a first predictive variable value for a pulse transit time (PTT) based on the first signal and the second signal processor for obtaining a second predictor variable value based on the extracted feature by extracting a feature from the second signal, and generating a blood pressure estimation model based on the first predictive variable value and the second predictive variable value may include

The signal acquisition unit may include a first sensor for measuring a first signal from the user and a second sensor for measuring a second signal from the user.

The signal acquisition unit may receive at least one of the first signal and the second signal from the external device through the communication unit.

In this case, the first signal includes at least one of electrocardiography (ECG), photoplethysmogram (PPG), electromyography (EMG), impedance plethysmogram (IPG), pressure wave, and video plethysmogram (VPG), and the second signal includes a ballistocardiogram (BCG). ) may be included.

When the first signal is a PPG signal and the second signal is a BCG signal, the processor determines a time interval between the PPG base point and any one of J-wave, H-wave, I-wave, K-wave, and L-wave of the BCG signal. is obtained by PTT, and when the first signal is an ECG signal and the second signal is a BCG signal, any one of the ECG R-wave and the J-wave, H-wave, I-wave, K-wave, and L-wave of the BCG signal A time interval with and may be obtained as the PTT.

The characteristic may include one or more of I-J time interval, J-L time interval, I-K time interval, J-K time interval, J-K amplitude, K-L amplitude, I-J amplitude of the second signal.

The processor may pre-process the input first signal and the second signal.

The processor may normalize the reference blood pressure, the first predictor value, and the characteristic.

The processor may generate a predictor variable acquisition model based on a reference blood pressure, the first predictive variable value, and the characteristic, and obtain the second predictive variable value by using the predictive variable acquisition model.

The processor may generate a residual reference blood pressure by removing a component associated with the first predictive variable value from the reference blood pressure, and may generate the predictive variable acquisition model based on the residual reference blood pressure.

The processor may remove a component associated with the value of the first predictor by subtracting a dot product of the reference blood pressure and the first predictive variable from the reference blood pressure.

The processor may generate a predictor variable acquisition model using a singular value decomposition (SVD) decomposition technique based on the residual reference blood pressure and the characteristic.

The processor may decompose a singular value of a component obtained by dot product of the residual reference blood pressure and the feature such that a covariance between the residual reference blood pressure and the feature is maximized.

The processor removes a component associated with the first predictive variable value from the reference blood pressure to generate a residual reference blood pressure, and a predictive variable using a singular value decomposition (SVD) decomposition technique based on the residual reference blood pressure and the characteristic. An acquisition model is generated, and the second predictor variable value is obtained using the predictor variable acquisition model, but the residual reference blood pressure is converted into a reference blood pressure and the above process is repeated until a predetermined number of second predictive variable values are obtained. can do.

According to an aspect, the method for generating a blood pressure estimation model includes receiving a first signal and a second signal, acquiring a PTT as a first predictor variable value based on the first signal and the second signal, and a feature from the second signal extracting , obtaining a first predictive variable value and a second predictive variable value based on the extracted features, and generating a blood pressure estimation model based on the first predictive variable value and the second predictive variable value may include

The obtaining of the second predictive variable value may include normalizing the reference blood pressure, the first predictive variable value, and the characteristic.

The obtaining of the second predictive variable value includes generating a predictor variable obtaining model based on the reference blood pressure, the first predictive variable value, and the characteristics, and obtaining the second predictive variable value using the predictive variable obtaining model. may include steps.

The generating of the predictive variable acquisition model may include generating the residual reference blood pressure by removing a component associated with the first predictive variable value from the reference blood pressure.

The generating of the residual reference blood pressure may include removing a component associated with the first predictive variable value by subtracting a dot product of the reference blood pressure and the first predictive variable value from the reference blood pressure.

The generating of the predictor variable acquisition model may include generating the predictive variable acquisition model using a singular value decomposition (SVD) decomposition technique based on the residual reference blood pressure and characteristics.

The generating of the predictor variable acquisition model may include decomposing a singular value of a component obtained by dot product of the residual reference blood pressure and the feature such that a covariance between the residual reference blood pressure and the feature is maximized.

The obtaining of the second predictive variable value includes generating a residual reference blood pressure by removing a component associated with the first predictive variable value from the reference blood pressure, and decomposing a singular value decomposition (SVD) based on the residual reference blood pressure and characteristics. generating a predictor variable acquisition model using a method, obtaining a second predictor variable value using the predictor variable acquisition model, and using the residual reference blood pressure until a predetermined number of second predictor variable values are obtained The method may include repeating the steps of generating the residual reference blood pressure by converting it into blood pressure.

According to an aspect, the blood pressure estimating apparatus includes a first sensor for measuring a first signal from a user, a second sensor for measuring a second signal from the user, and a first predictor variable for PTT based on the first signal and the second signal. It is obtained as a value, a feature is extracted from the second signal, a second predictor value is obtained based on the extracted feature using a predictor variable acquisition model, and a first predictor variable value and a second predictor variable value are obtained using a blood pressure estimation model. 2 Blood pressure can be estimated based on the predictor values.

Blood pressure can be estimated without a cuff through pulse transit time (PTT) and pulse wave analysis (PWA) analysis using the BCG signal.

1 and 2 are block diagrams of an apparatus for generating a blood pressure estimation model according to embodiments.
3 is an embodiment of the processor of FIGS. 1 and 2 ;
4 is a diagram for explaining an example of extracting PTT and features.
5 is a flowchart of a method for generating a blood pressure estimation model according to an exemplary embodiment.
6A and 6B are flowcharts of embodiments of obtaining the second predictor value of FIG. 5 .
7 is a block diagram of an apparatus for estimating blood pressure according to an exemplary embodiment.
8 is a flowchart of a method for estimating blood pressure according to an exemplary embodiment.
9 is a diagram illustrating a wearable device.
10 is a diagram illustrating a smart device.
11A to 11C are diagrams illustrating an earphone.

The details of other embodiments are included in the detailed description and drawings. Advantages and features of the described technology, and how to achieve them, will become apparent with reference to the embodiments described below in detail in conjunction with the drawings. Like reference numerals refer to like elements throughout.

Terms such as first, second, etc. may be used to describe various elements, but the elements should not be limited by the terms. The terms are used only for the purpose of distinguishing one component from another. The singular expression includes the plural expression unless the context clearly dictates otherwise. Also, when a part "includes" a certain component, it means that other components may be further included, rather than excluding other components, unless otherwise stated. In addition, terms such as “…unit” and “module” described in the specification mean a unit that processes at least one function or operation, which may be implemented as hardware or software, or a combination of hardware and software.

Various embodiments described below take blood pressure estimation as an example, but it is not limited to blood pressure estimation, and various bio-information such as blood vessel age, arteriosclerosis, aortic pressure waveform, stress index, fatigue, skin age, and skin elasticity are used. can be applied to estimation.

1 and 2 are block diagrams of an apparatus for generating a blood pressure estimation model according to embodiments.

Referring to FIG. 1 , an apparatus 100 for generating a blood pressure estimation model according to an embodiment includes a signal acquisition unit 110 and a processor 120 .

The signal acquisition unit 110 may include a first sensor 111 for measuring a first signal from the user and a second sensor 112 for measuring a second signal. The signal acquisition unit 110 measures a plurality of first and second signals from a plurality of users through the first sensor 111 and the second sensor 112 , and the measured plurality of first and second signals The signal may be transmitted to the processor 120 as learning data.

The first sensor 111 may be a PPG sensor that measures a photoplethysmogram (PPG) signal. In this case, the PPG sensor may include one or more light sources for irradiating light to the test site of the user, and one or more detectors for detecting reflected or scattered light from the test site. The light source may include a light emitting diode (LED), a laser diode, and a phosphor. The light source may be formed in one or two or more arrays, and each light source may irradiate light of a different wavelength. In addition, the detector may include a photodiode, a phototransistor, a complementary metal-oxide semiconductor (CMOS) image sensor, a charge-coupled device (CCD) image sensor, and the like, and is formed in one or two or more arrays. can be

However, the first sensor 111 is not limited to the PPG sensor, and includes other sensors measuring electrocardiography (ECG), electromyography (EMG), impedance plethysmogram (IPG), pressure wave, and video plethysmogram (VPG). can do.

The second sensor 112 may be a sensor that measures a ballistocardiogram (BCG) signal. For example, the second sensor 112 may include a displacement sensor, a velocity sensor, an acceleration sensor, a load cell sensor, a polyvinylidene fluoride (PVDF) film sensor, an electro mechanical film (EMFi) sensor, and a force sensor. Various types of sensors capable of measuring a BCG signal, such as a force sensor, may be included. However, the second signal is not limited to the BCG signal.

The processor 130 may control the first sensor 111 and the second sensor 112 and generate a blood pressure estimation model using the learning data received from the first sensor 111 and the second sensor 112 . have. In this case, reference blood pressure (eg, SBP, DBP, MBP, PP, etc.) measured from a plurality of users may be further collected as training data, and a predictor variable acquisition model and a blood pressure estimation model are generated using the collected training data. can do. For example, a pulse transit time (PTT) is obtained based on the first signal and the second signal, a plurality of features are extracted based on the second signal, and the reference blood pressure, PTT, A predictor variable acquisition model and a blood pressure estimation model may be generated based on the plurality of features.

Referring to FIG. 2 , the apparatus 200 for generating a blood pressure estimation model according to an embodiment includes a signal acquisition unit 110 , a processor 120 , a communication unit 210 , a storage unit 220 , and an output unit 230 . can do.

According to the present embodiment, in the signal acquisition unit 110 , at least one of the first sensor 111 and the second sensor 112 may be omitted, and the first signal and the second signal from the external device through the communication unit 210 . At least one of the signals may be received. Alternatively, the signal acquisition unit 110 may include both the first sensor 111 and the second sensor 112 as in the above-described embodiment, and selectively the first sensor 111 according to the control of the processor 120 . Alternatively, the second sensor 112 may be driven and the remaining non-driven sensors mounted on the external device may be controlled through the communication unit 210 . The external device may include, for example, a smartphone or wearable device equipped with a PPG or BCG signal measuring function, or a signal measuring device configured to individually measure ECG or the like.

The communication unit 210 may communicate with an external device using wired/wireless communication technology under the control of the processor 120 , and may transmit/receive a control signal of the processor 120 and/or a signal measured by the external device. Alternatively, the communication unit 210 transmits the predictor variable acquisition model or the blood pressure estimation model generated by the processor 120 to the blood pressure estimation device periodically or irregularly at the request of the blood pressure estimation device, and transmits the predictive variable acquisition model or the blood pressure estimation device. You can update the model.

At this time, wired and wireless communication technologies include Bluetooth (Bluetooth) communication, BLE (Bluetooth Low Energy) communication, near field communication (NFC), WLAN communication, Zigbee communication, infrared (Infrared Data Association, IrDA) communication, Including, but not limited to, Wi-Fi Direct (WFD) communication, ultra-wideband (UWB) communication, Ant+ communication, WIFI communication, Radio Frequency Identification (RFID) communication, 3G communication, 4G communication, and 5G communication.

The storage 220 may store programs or commands for generating a blood pressure estimation model. In addition, learning data collected from a plurality of users may be stored to generate a blood pressure estimation model, and a predictor variable acquisition model or a blood pressure estimation model generated by the processor 120 may be stored.

At this time, the storage unit 220 is a flash memory type (flash memory type), a hard disk type (hard disk type), a multimedia card micro type (multimedia card micro type), card type memory (eg, SD or XD memory, etc.) , Random Access Memory (RAM), Static Random Access Memory (SRAM), Read Only Memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), Programmable Read Only Memory (PROM), Magnetic Memory, Magnetic Disk , may include at least one type of storage medium, such as an optical disk. Also, the storage 220 may include an external storage medium such as web storage.

The output unit 230 may output the processing result of the processor 120 . For example, the output unit 230 may output the first signal, the second signal, the predictor variable acquisition model, the blood pressure estimation model, and various information related thereto. The output unit 230 may provide information to the user using various output modules, such as a display module, a speaker, and a haptic module, for example.

3 is an embodiment of the processor of FIGS. 1 and 2 ; 4 is a diagram for explaining an example of extracting PTT and features.

Referring to FIG. 3 , the processor 300 according to an embodiment includes a preprocessor 310 , a PTT acquirer 220 , a feature extractor 330 , a predictor variable acquisition model generator 340 , and a blood pressure estimation model generator (350).

The preprocessor 310 may preprocess the first signal and the second signal upon receiving the first signal and the second signal from the signal obtaining unit 110 . For example, the preprocessor 310 may remove noise such as motion noise by using various noise removal techniques such as filtering and smoothing. For example, when the first signal is an ECG signal, band-pass filtering having a cutoff frequency of 1 to 40 Hz may be performed, and in the case of a PPG signal, band-pass filtering having a cut-off frequency of 1-10 Hz may be performed. In addition, when the second signal is a BCG signal, band pass filtering having a cutoff frequency of 0.8-20 Hz may be performed.

The PTT acquisition unit 320 may acquire the PTT as the first predictor variable based on the first signal and the second signal. For example, a characteristic point may be extracted from the first signal and the second signal, and a time interval between the extracted characteristic points may be obtained as a PTT. For example, referring to FIG. 4 , when the first signal is the ECG signal 41 and the second signal is the BCG signal 43 , the R-wave (R) of the first signal and the J-wave (J) of the second signal A time interval (PTT1) of can be obtained as PTT. or PTT the time interval PTT2 between the base point F of the first signal and the J-wave J of the second signal when the first signal is the PPG signal 42 and the second signal is the BCG signal 43 can be obtained with However, the present invention is not limited thereto, and instead of the J-wave of the BCG signal 43 , H-wave, I-wave, K-wave, L-wave, etc. R-wave (R) of the ECG signal 41, PPG signal ( 42), the time interval between the base points may be obtained as PTT.

The feature extraction unit 330 may extract one or more feature points from the second signal. For example, the feature extraction unit 330 may perform G-wave (G), H-wave (H), I-wave (I), and J-wave (J) in the BCG signal 42 as shown in FIG. 4 . ), K-wave (K), L-wave (L), etc. can be extracted as feature points, and features can be extracted based on the extracted feature points. In this case, the feature extraction unit 330 may perform bit gating of the second signal to obtain a one-period signal, and extract a feature point from the obtained one-period signal. For example, the feature extractor 330 may segment one cycle signals by bit-gating the second signal based on the feature points of the first signal. As an example, when the first signal is an ECG signal, gating may be performed based on the R-wave, and as another example, when the first signal is a PPG signal, gating may be performed based on a heart rate 50% of the base point.

In addition, the feature extraction unit 330 uses the extracted feature points to extract, for example, JL time, IK time, JK time, JK amplitude associated with a pulse pressure (PP) change, KL amplitude, and IJ amplitude associated with PTT change as features. can Also, the J-J time associated with the heart rate, that is, the time interval between the J-wave of one cycle signal and the J-wave of the next cycle signal, may be extracted as features.

The predictor variable acquisition model generator 340 may normalize the reference blood pressure, the first predictive variable value, and each characteristic. As described above, through normalization, the characteristic difference for each subject can be corrected. For example, as in Equation 1, reference blood pressure measured for a predetermined period, such as DBP (diastolic blood pressure), SBP (systolic blood pressure), PP (pulse pressure), a first predictive variable value, and the average of each feature is used. can be normalized. However, it is not limited to Equation 1 above, and instead of the average value, a reference blood pressure, a characteristic, and a PTT obtained at a time when the user is stable may be used, and various other known normalization techniques may be used.

는 i 번째 피험자의 기준 혈압을 나타낸다. 여기서, 기준 혈압은 DBP, SBP, PP 일 수 있다.

는 기준 혈압의 평균,

는 기준 혈압의 정규화값을 의미한다. i는 i번째 피험자를 나타낸다. 또한,

,

,

는 각각 i번째 피험자에 대한 k번째 특징, k번째 특징의 평균, k번째 특징의 정규화값을 의미하고, k는 추출된 각 특징을 나타내는 인덱스이다. 또한,

,

,

는 각각 i번째 피험자에 대한 제1 예측변수값 즉, PTT, PTT의 평균, PTT의 정규화값을 의미한다. 여기서, 평균은 소정 기간 동안 측정된 기준 혈압, 특징, PTT의 평균을 의미한다.here,

denotes the baseline blood pressure of the i-th subject. Here, the reference blood pressure may be DBP, SBP, or PP.

is the average of the baseline blood pressure,

is the normalized value of the reference blood pressure. i represents the i-th subject. Also,

,

,

denotes the k-th feature, the average of the k-th feature, and the normalized value of the k-th feature for the i-th subject, respectively, and k is an index indicating each extracted feature. Also,

,

,

denotes the first predictor value for the i-th subject, respectively, that is, the PTT, the average of PTT, and the normalized value of the PTT. Here, the average means the average of the reference blood pressure, the characteristic, and the PTT measured for a predetermined period.

The predictor variable acquisition model generator 340 may generate a predictive variable acquisition model based on the reference blood pressure, the first predictive variable value, and the characteristic. In this case, the reference blood pressure, the first predictive variable value, and the characteristic may be normalized values. For example, a predictor variable acquisition model for SBP estimation, a predictor variable acquisition model for DBP estimation, and a predictor variable acquisition for PP estimation based on the normalized first predictor variable values and features, and the reference SBP, the reference DBP, and the reference PP, respectively Each model can be created.

The predictor variable acquisition model generator 340 may generate a residual reference blood pressure by removing a blood pressure component associated with the first predictive variable value from the reference blood pressure. For example, as shown in Equation 2 below, the residual reference blood pressure may be generated by subtracting the dot product of the reference blood pressure and the first predictive variable value from the reference blood pressure.

는 잔여 기준 혈압으로 PTT에 의해 설명될 수 없는 성분 즉 PTT에 직교한 혈압(SBP, DBP, PP) 성분을 나타낸다. P는 각 피험자들의 정규화된 기준 혈압(SBP, DBP, PP)을 기준 혈압 벡터로 통합한 것을 나타내며, τ는 각 피험자들의 정규화된 제1 예측변수값(PTT)를 통합한 제1 예측변수값 벡터를 나타낸다.

는 기준 혈압(P)과 제1 예측변수값(τ)의 내적 성분을 나타낸다. here,

denotes a component that cannot be explained by PTT as a residual reference blood pressure, that is, a component of blood pressure (SBP, DBP, PP) orthogonal to PTT. P represents the integration of the normalized reference blood pressures (SBP, DBP, PP) of each subject into the reference blood pressure vector, and τ is the first predictor value vector incorporating the normalized first predictor values (PTT) of each subject. indicates

denotes the inner product component of the reference blood pressure (P) and the first predictor value (τ).

The predictor variable acquisition model generator 340 may generate a predictor variable acquisition model using the features extracted from the second signal, for example, through principal component analysis (PCA), singular value decomposition, or the like. . For example, the singular value decomposition may be performed on a component obtained by dot product of the residual reference blood pressure and each feature such that the covariance between the residual reference blood pressure and each feature is maximized.

The predictor variable acquisition model generator 340 may derive a coupling coefficient for each feature through singular value decomposition, and may generate a predictor variable acquisition model based on the derived coupling coefficient for each feature. In this case, the coupling coefficient for each feature may be derived by learning the learning data collected from a plurality of subjects by using various learning techniques such as machine learning.

For example, Equation 3 below shows an example of a predictor variable acquisition model.

, 는 특이값 분해를 통해 획득된 k번째 특징에 대한 결합계수,

는 k번째 특징을 나타낸다. 위 과정을 기준 DBP, 기준 SBP, 기준 PP별로 수행함으로써 각 특징에 대한 결합계수를 각 기준 혈압별로 획득할 수 있다.where X is the value of the second predictor,

, is the coupling coefficient for the k-th feature obtained through singular value decomposition,

represents the k-th feature. By performing the above process for each reference DBP, reference SBP, and reference PP, a coupling coefficient for each characteristic can be obtained for each reference blood pressure.

When the predictor acquisition model is generated in this way, the predictor acquisition model generation unit 340 obtains a second predictor variable value by combining each feature extracted from the second signal through the predictive variable acquisition model and a coupling coefficient for each feature. can do.

The blood pressure estimation model generator 350 uses the first predictive variable value, the second predictive variable value, and each reference blood pressure of the plurality of subjects as learning data, for example, by learning the blood pressure estimation model as shown in Equation 4 below, thereby applying the first predictive variable value to each predictive variable. coefficients can be derived. However, this blood pressure estimation model is only an example.

Here, Y denotes an estimated blood pressure value, A denotes a first predictive variable value, and X denotes a second predictive variable value. α, β, and γ are coefficients obtained through learning using learning data of a plurality of subjects. As described above, by performing the above process for each reference blood pressure, that is, DBP, SBP, and PP, a blood pressure estimation model may be generated for each reference blood pressure. Accordingly, DBP, SBP, and the like can be estimated independently.

Meanwhile, the predictor variable acquisition model generator 340 may acquire a predetermined number of second predictor variable values by performing singular value decomposition in multiple steps. For example, as described above, the process of generating the residual reference blood pressure by removing the blood pressure component associated with the first predictive variable value from the reference blood pressure and obtaining the second predictive variable value using the residual reference blood pressure is repeatedly performed. From the process of acquiring the value of the second predictor variable, the residual reference blood pressure generated in the immediately preceding process may be converted into a new reference blood pressure and then performed. In this way, the PTT-related blood pressure component is removed from the reference blood pressure to generate the first residual reference blood pressure, and the PTT-related blood pressure component is again removed from the first residual reference blood pressure to generate the second residual reference blood pressure. By decomposing, more accurate blood pressure estimation may be possible.

Equation 5 below is an example of a blood pressure estimation model using the second predictor value obtained through multi-step singular value decomposition.

Here, Y represents the blood pressure estimate, A represents the first predictor value, X ₁ , X ₂ , ... , Xn represents the second predictor value obtained through multi-step singular value decomposition. α, β ₁ , β ₂ , … , βn, γ are coefficients obtained through learning using learning data of a plurality of subjects. β ₁ , β ₂ , … , βn may be the same value.

5 is a flowchart of a method for generating a blood pressure estimation model according to an exemplary embodiment. 6A and 6B are flowcharts of embodiments of obtaining the second predictor value of FIG. 5 .

Referring to FIG. 5 , the method for generating a blood pressure estimation model may be performed by the above-described blood pressure estimation model generating apparatus 100 . Since it has been described in detail above, it will be briefly described below.

First, a first signal and a second signal are received ( 510 ). In this case, the first signal may be one of electrocardiography (ECG), photoplethysmogram (PPG), electromyography (EMG), impedance plethysmogram (IPG), pressure wave, and video plethysmogram (VPG), and the second signal is ballistocardiogram (BCG). can In this case, the first signal and the second signal may be obtained from a plurality of subjects for learning to generate a blood pressure estimation model.

Then, based on the first signal and the second signal, the PTT may be obtained as a first predictor value ( 520 ). For example, the time interval of R-wave and J-wave when the first signal is ECG and the second signal is BCG, and the time interval between the base point and J-wave when the first signal is PPG and the second signal is BCG. It can be obtained through PTT. However, the present invention is not limited thereto, and instead of the J-wave of the BCG signal 43 , H-wave, I-wave, K-wave, L-wave, etc. R-wave (R) of the ECG signal 41, PPG signal ( 42), the time interval between the base points may be obtained as PTT.

Then, one or more features may be extracted based on the second signal ( 530 ). For example, in the BCG signal, J-L time associated with PTT change, I-K time, J-K time, J-K amplitude associated with pulse pressure (PP) change, K-L amplitude, I-J amplitude, and J-J time associated with heart rate can be extracted as features.

Next, a second predictive variable value may be obtained based on the reference blood pressure, the first predictive variable value, and the characteristic ( 540 ).

An example of obtaining the second predictor value will be described with reference to FIG. 6A .

First, the reference blood pressure, the first predictor value, and the characteristic may be normalized ( 611 ). For example, as described above, a value obtained by subtracting the average of each value from each value may be normalized by dividing the value by the average of each value.

Next, the residual reference blood pressure may be generated by removing the blood pressure component associated with the second predictive variable value from the reference blood pressure ( S612 ). For example, a blood pressure component described by PTT may be obtained by dot product of the reference blood pressure and PTT, and the blood pressure component described as PTT may be subtracted from the reference blood pressure.

Next, a predictor variable acquisition model may be generated based on the residual reference blood pressure and characteristics ( 613 ). For example, a predictor variable acquisition model may be generated by obtaining a coupling coefficient for each feature by decomposing the residual reference blood pressure and a component obtained by dot product of each feature to maximize the covariance between the residual reference blood pressure and each feature. In this case, it is possible to obtain an optimal coupling coefficient for each characteristic through learning using the learning data of a plurality of subjects.

Then, a second predictor variable value may be obtained using the predictor variable acquisition model ( 614 ). For example, as shown in Equation 3, the second predictor variable value may be obtained by summing all the products of the coupling coefficient of each feature and each feature.

An example of obtaining a plurality of second predictor values by performing a process of obtaining a second predictor value in multiple steps will be described with reference to FIG. 6B .

First, the reference blood pressure, the first predictor value, and the characteristic may be normalized ( S621 ). For example, as described above, a value obtained by subtracting the average of each value from each value may be normalized by dividing the value by the average of each value.

Next, the residual reference blood pressure may be generated by removing the blood pressure component associated with the second predictive variable value from the reference blood pressure ( S622 ). For example, a blood pressure component described by PTT may be obtained by dot product of the reference blood pressure and PTT, and the blood pressure component described as PTT may be subtracted from the reference blood pressure.

Next, a predictor variable acquisition model may be generated based on the residual reference blood pressure and characteristics ( S623 ). For example, a predictor variable acquisition model may be generated by obtaining a coupling coefficient for each feature by decomposing the residual reference blood pressure and a component obtained by dot product of each feature to maximize the covariance between the residual reference blood pressure and each feature. In this case, it is possible to obtain an optimal coupling coefficient for each characteristic through learning using the learning data of a plurality of subjects.

Next, a second predictor variable value may be obtained using the predictor variable acquisition model ( 624 ). For example, as shown in Equation 3, the second predictor variable value may be obtained by summing all the products of the coupling coefficient of each feature and each feature.

Next, it may be determined whether a predefined number of second predictor values has been obtained ( 625 ). If the desired number of second predictor variable values is not obtained as a result of the determination, the residual reference blood pressure generated in step 622 is converted into a new reference blood pressure ( 626 ), and steps 622 and below are repeated again. As a result of the determination, if a desired number of values of the second predictor is obtained, the process is terminated.

Referring back to FIG. 5 , a blood pressure estimation model may be generated based on the first predictive variable value and the second predictive variable value ( 550 ). For example, coefficients for each predictive variable may be derived by learning a blood pressure estimation model as shown in Equation 4 using the first predictive variable value, the second predictive variable value, and the reference blood pressure of the plurality of subjects as learning data.

7 is a block diagram of an apparatus for estimating blood pressure according to an exemplary embodiment.

Referring to FIG. 7 , the blood pressure estimating apparatus 700 may include a first sensor 710 , a second sensor 720 , a processor 730 , an output unit 740 , a storage unit 750 , and a communication unit 760 . can

The first sensor 710 and the second sensor 720 may be electrically connected to the processor 730 and may measure the first signal and the second signal from the user, respectively, under the control of the processor 730 . The first signal may include electrocardiography (ECG), electromyography (EMG), impedance plethysmogram (IPG), photoplethysmogram (PPG), pressure wave, and video plethysmogram (VPG). Also, the second signal may be a ballistocardiogram (BCG) signal.

The processor 730 may simultaneously control the first sensor 710 and the second sensor 720 in response to a blood pressure estimation request to measure the first signal and the second signal for a predetermined time. When the first signal and the second signal are received, the processor 730 may perform pre-processing such as band pass filtering, smoothing, and bit averaging of continuous measurement signals.

The processor 730 may obtain a first predictor variable value of the blood pressure estimation model by obtaining a PTT based on the feature points of the first signal and the second signal. In addition, the processor 730 may extract a plurality of features from the second signal, and may obtain a second predictive variable value of the blood pressure estimation model by applying a predictive variable acquisition model to the extracted plurality of features. In this case, the plurality of features, for example, J-L time, I-K time, J-K time, J-K amplitude, K-L amplitude, and I-J amplitude associated with PTT change in the BCG signal can be extracted as features. It may also include the J-J time associated with the heart rate, and the like. In addition, when the PTT and a plurality of features are obtained, the processor 730 may normalize to reflect the user's characteristics, and obtain a first predictor variable value and a second predictor variable value using the normalized PTT and feature.

When the first predictive variable value and the second predictive variable value are obtained, the processor 730 may obtain the blood pressure estimate value using the blood pressure estimation model expressed in Equation 4 or 5 described above. In addition, the processor 730 may evaluate the cardiovascular health state by comprehensively analyzing whether the current blood pressure estimate value is normal, the blood pressure estimate history, and the like. The processor 730 may generate information such as an evaluation score obtained by quantifying the cardiovascular health status evaluation result or measures to be taken by the user according to the evaluation result.

The output unit 740 may output the first signal, the second signal, the blood pressure estimate, and the health state evaluation result to provide to the user. In this case, the output unit 740 may include various output means such as a display, a speaker, and a haptic device, and may provide information to a user by utilizing one or more of various output means.

The storage unit 750 stores various types of information for blood pressure estimation. For example, a predefined blood pressure estimation model, a predictor variable acquisition model, and the like may be stored. In addition, the first and second signals measured by the first sensor 710 and the second sensor 720, the blood pressure estimate value obtained by the processor 730, the health status evaluation result, the blood pressure estimation history, etc. may be stored. have. In addition, user characteristic information serving as basic data for health condition evaluation, such as the user's age, gender, and health state, may be stored. However, it is not limited to these examples.

The communication unit 760 may communicate with an external device and may transmit/receive various data related to blood pressure estimation. For example, the predictor variable acquisition model and the blood pressure estimation model may be received from the blood pressure estimation model generating apparatus 100 , and the existing predictive variable acquisition model and the blood pressure estimation model stored in the storage unit 750 may be updated. In addition, the blood pressure estimation result may be transmitted to an external device such as a user's smartphone, tablet PC, desktop PC, notebook PC, or wearable device.

Meanwhile, the processor 730 may generate and update the personalized model by performing calibration of the predictor variable acquisition model and/or the blood pressure estimation model according to a preset period or a user's request.

For example, when the blood pressure estimation apparatus 700 is manufactured, the general-purpose predictor variable acquisition model and the blood pressure estimation model generated by the blood pressure estimation model generating apparatus 100 may be stored in the storage unit 750 . The processor 730 may initially estimate the blood pressure using the general-purpose predictor variable acquisition model and the blood pressure estimation model. When the user's blood pressure is estimated for a predetermined period using the general-purpose predictive variable acquisition model and the blood pressure estimation model, data such as the first signal, the second signal, PTT, characteristics, and reference blood pressure collected from the user during the period are used as learning data. Thus, as described above, a blood pressure estimation model and predictor variable acquisition model personalized to the user can be generated through learning. As described above, the accuracy of the blood pressure estimation model to be applied to the user can be improved by periodically accumulating previous data obtained from the user while estimating the user's blood pressure and using it as learning data.

8 is a flowchart of a method for estimating blood pressure according to an exemplary embodiment.

FIG. 8 is an embodiment of a blood pressure estimation method performed by the blood pressure estimation apparatus 700 of FIG. 7 , which will be briefly described below.

First, a first signal and a second signal are respectively measured from the user ( 810 ).

Next, a PTT is obtained based on the feature points of the first signal and the second signal to obtain a first predictor value of the blood pressure estimation model ( 820 ), and a plurality of features can be extracted from the second signal ( 830 ).

Next, a second predictor value of the blood pressure estimation model may be obtained using the extracted plurality of features using the predictor variable acquisition model ( 840 ). In this case, the extracted features may be normalized to reflect the user's characteristics.

Next, when the first predictive variable value and the second predictive variable value are obtained, an estimated blood pressure value may be obtained using the blood pressure estimation model ( S850 ). In this case, the cardiovascular health status may be evaluated by comprehensively analyzing whether the current blood pressure estimate is normal, the blood pressure estimation history, and the like, and an evaluation result may be generated.

Then, the first signal, the second signal, the blood pressure estimate value, and/or the health state evaluation result may be output ( 860 ).

9 illustrates a wearable device. Embodiments of the above-described blood pressure estimating apparatus 700 may be mounted on a wearable device.

Referring to FIG. 9 , the wearable device 900 includes a body 910 and a strap 930 .

The strap 930 may be connected to both ends of the body 910 to be flexible so as to be bent in a shape wrapped around the user's wrist. The strap 930 may include a first strap and a second strap separated from each other. One end of the first strap and the second strap may be connected to the main body 910 and the other end may be fastened to each other through a fastening means. In this case, the fastening means may be formed in a manner such as magnet fastening, velcro fastening, pin fastening, and the like, but is not limited thereto. In addition, the strap 930 is not limited thereto and may be formed integrally without being separated from each other, such as a band shape.

At this time, the strap 930 may be formed such that air is injected therein to have elasticity according to a change in pressure applied to the wrist or may include an air bag, and may transmit the pressure change of the wrist to the body 910 .

A battery for supplying power to the wearable device 900 may be built in the body 910 or the strap 930 .

A sensor unit 920 is mounted on one side of the body 910 . The sensor unit 920 may include a PPG sensor that measures a PPG signal, and the PPG sensor includes a light source that irradiates light to the skin of the wrist and a CIS-based optical sensor that detects light scattered or reflected from the subject, a photodiode, etc. of the detector may be included. Also. The sensor unit 920 may further include a force sensor or an acceleration sensor that measures a BCG signal from the subject.

A processor is mounted inside the main body 910 , and the processor may be electrically connected to components mounted on the wearable device 900 . The processor may obtain a PTT by using the PPG signal and the BCG signal measured by the sensor unit 920 , extract a plurality of features from the BCG signal, and estimate the blood pressure based on the PTT and the features. A detailed description will be omitted.

Also, a storage unit for storing reference information for estimating blood pressure and performing various functions of the wearable device 900 and information processed by various components may be mounted inside the main body 910 .

In addition, a manipulation unit 940 for receiving a user's control command and transmitting it to the processor may be mounted on one side of the main body 910 . The manipulation unit 940 may include a power button for inputting a command for turning on/off the power of the wearable device 900 .

Also, a display unit for outputting information to a user may be mounted on the front surface of the main body 910 , and the display unit may be a display including a touch screen capable of a touch input. The display unit may receive a user's touch input, transmit it to the processor, and display a processing result of the processor.

In addition, a communication unit for communicating with an external device may be mounted inside the main body 910 . The communication unit may transmit a blood pressure estimation result to an external device, for example, a user's smart phone, and may receive a blood pressure estimation model from the blood pressure estimation model generating device.

10 shows a smart device. In this case, the smart device may include a smart phone and a tablet PC. The smart device may be equipped with the function of the above-described blood pressure estimating apparatus 700 .

Referring to FIG. 10 , the smart device 1000 may have a sensor unit 1030 mounted on one surface of the main body 1010 . The sensor unit 1030 may include a PPG sensor including a light source 1031 and a detector 1032 , and a force sensor or an acceleration sensor for measuring a BCG signal. The detector 1032 may include a photodiode, a CIS-based photosensor, or the like.

In addition, a display unit may be mounted on the front surface of the main body 1010 , and the display unit may visually output a blood pressure estimation result, a health state evaluation result, and the like. The display unit may include a touch screen, and may receive information input through the touch screen and transmit it to the processor.

The body 1010 may include an image sensor 1020 as shown. The image sensor 1020 performs a function of photographing various images, and for example, when a finger contacts the sensor unit 1030 , a finger may acquire a fingerprint image. Meanwhile, when the CIS-based image sensor 1032 is mounted on the first sensor of the sensor unit 1030 , the image sensor 1020 may be omitted.

As described above, the processor may estimate the blood pressure based on the PPG signal and the BCG signal measured by the sensor unit 1030 . A detailed description will be omitted.

11A to 11C illustrate an earphone. Various embodiments of the above-described blood pressure estimating apparatus 700 may be mounted on a device such as an earphone as illustrated. Here, the earphones include wired or wireless earphones, and are not limited to their types, such as earbuds, necklace-type earphones, and ear-type earphones.

11A and 11B, the user can measure the PPG signal and the BCG signal by contacting the earphone 1110 with one finger O2 while the earphone 1110 is inserted into the ear O1. The blood pressure may be estimated using the PPG signal and the BCG signal. For example, referring to FIG. 11B , a BCG sensor 1110 for measuring a BCG signal may be mounted inside the earphone 1110 body. In addition, the PPG sensor 1120 may be disposed on the outer surface of the earphone main body so that the earphone 1110 body can measure the PPG signal from the finger when the finger is in contact with the earphone 1110 body inserted into the ear. In this case, the BCG sensor 1110 and the PPG sensor 1120 may be disposed on both the left and right main bodies of the earphone 1100 or disposed only on one side.

Referring to FIG. 11C , the PPG signal and the BCG signal can be simultaneously measured while the user inserts the earphone 1110 into the ear. For example, a BCG sensor 1110 for measuring a BCG signal is mounted inside the earphone 1110, and the earphone 1110 body can measure the PPG signal inside the contacting ear while the earphone 1110 body is inserted into the ear. A PPG sensor 1120 may be disposed on the inner surface of the . In this case, the BCG sensor 1110 and the PPG sensor 1120 may be disposed on both the left and right main body of the earphone 1100 or disposed only on one side.

Meanwhile, the processor for performing the blood pressure estimation function may be mounted inside the earphone 1100 body or the earphone controller. In this case, when the earphone 1100 is inserted into the ear, the processor may control the BCG sensor and the PPG sensor, and as described above, may estimate the biometric information using the BCG signal and the PPG signal. The processor may transmit the blood pressure estimation result to an external device connected via wire or wireless, for example, a smart phone or a wearable device.

In addition, the processor for performing the blood pressure estimation function may be mounted on an external device connected to the earphone 1100 by wire or wirelessly, for example, a smart phone or a wearable device, instead of being mounted on the earphone 1100 body. The processor of the external device may receive the BCG signal and the PPG signal from the BCG sensor 1110 and the PPG sensor 1120 of the earphone 1100 , and estimate the blood pressure using the received signals.

The earphone 1100 or the processor of the external device may visually output the blood pressure estimation result to the user through the display of the external device. Alternatively, when the blood pressure estimation is completed when the earphone 1100 is worn on the ear, the biometric information estimation result may be output by voice through the earphone 1100 .

Meanwhile, the present embodiments can be implemented as computer-readable codes on a computer-readable recording medium. The computer-readable recording medium includes all types of recording devices in which data readable by a computer system is stored.

Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage device, etc. include In addition, the computer-readable recording medium is distributed in a computer system connected to a network, so that the computer-readable code can be stored and executed in a distributed manner. In addition, functional programs, codes, and code segments for implementing the present embodiments can be easily inferred by programmers in the relevant technical field.

Those of ordinary skill in the art to which the present disclosure pertains will be able to understand that the disclosure may be implemented in other specific forms without changing the technical spirit or essential features. Therefore, it should be understood that the embodiments described above are illustrative in all respects and not restrictive.

100,200: blood pressure estimating device 110: signal acquisition unit
111: first sensor 112: second sensor
120,300: processor 210: communication unit
220: storage unit 230: output unit
310: preprocessor 320: PTT acquisition unit
330: feature extraction unit 340: predictor variable acquisition model generation unit
350: blood pressure estimation model generation unit
700: blood pressure estimating device 710: first sensor
720: second sensor 730: processor
740: output unit 750: storage unit
760: communication department

Claims (23)

a signal acquisition unit receiving a user's first signal and a second signal; and
Obtaining a pulse transit time (PTT) as a first predictive variable value based on the first signal and the second signal, extracting a feature from the second signal, and obtaining a second predictive variable value based on the extracted feature, and a processor for generating a blood pressure estimation model based on the first predictive variable value and the second predictive variable value. According to claim 1,
The signal acquisition unit
a first sensor for measuring a first signal from a user; and
A blood pressure estimation model generating apparatus including a second sensor for measuring a second signal from a user. According to claim 1,
The signal acquisition unit
An apparatus for generating a blood pressure estimation model for receiving at least one of the first signal and the second signal from an external device through a communication unit. According to claim 1,
The first signal includes at least one of electrocardiography (ECG), photoplethysmogram (PPG), electromyography (EMG), impedance plethysmogram (IPG), pressure wave, and video plethysmogram (VPG),
The second signal is an apparatus for generating a blood pressure estimation model including a ballistocardiogram (BCG). According to claim 1,
the processor is
When the first signal is a PPG signal and the second signal is a BCG signal, a time interval between the PPG base point and one of the J-wave, H-wave, I-wave, K-wave, and L-wave of the BCG signal is defined as the PTT. obtained with
When the first signal is an ECG signal and the second signal is a BCG signal, a time interval between the ECG R-wave and one of the J-wave, H-wave, I-wave, K-wave, and L-wave of the BCG signal is defined as the PTT. A device for generating a blood pressure estimation model obtained by According to claim 1,
The feature is
An apparatus for generating a blood pressure estimation model including at least one of an IJ time interval, a JL time interval, an IK time interval, a JK time interval, a JK amplitude, a KL amplitude, and an IJ amplitude of the second signal. According to claim 1,
the processor is
A blood pressure estimation model generating apparatus for pre-processing the input first and second signals. According to claim 1,
the processor is
An apparatus for generating a blood pressure estimation model that normalizes a reference blood pressure, the first predictive variable value, and the characteristic. According to claim 1,
the processor is
A blood pressure estimation model generating apparatus for generating a predictive variable acquisition model based on a reference blood pressure, the first predictive variable value, and the characteristic, and acquiring the second predictive variable value by using the predictive variable acquiring model. 10. The method of claim 9,
the processor is
A blood pressure estimation model generating apparatus for generating a residual reference blood pressure by removing a component associated with the first predictive variable value from the reference blood pressure, and generating the predictive variable acquisition model based on the residual reference blood pressure. 11. The method of claim 10,
the processor is
A blood pressure estimation model generating apparatus for removing a component associated with the first predictive variable value by subtracting a dot product of the reference blood pressure and the first predictive variable value from the reference blood pressure. 12. The method of claim 11,
the processor is
A blood pressure estimation model generating apparatus for generating the predictive variable acquisition model using a singular value decomposition (SVD) decomposition technique based on the residual reference blood pressure and the characteristic. 13. The method of claim 12,
the processor is
A blood pressure estimation model generating apparatus for decomposing a singular value of a component obtained by dot product of the residual reference blood pressure and the feature such that a covariance between the residual reference blood pressure and the feature is maximized. According to claim 1,
the processor is
A residual reference blood pressure is generated by removing a component associated with the first predictive variable value from the reference blood pressure, and a predictive variable is obtained using a singular value decomposition (SVD) decomposition technique based on the residual reference blood pressure and the characteristic. generating a model, and obtaining the second predictor variable value using the predictor variable acquisition model;
A blood pressure estimation model generating apparatus for converting the residual reference blood pressure into a reference blood pressure and repeating the above process until a predetermined number of second predictive variable values are obtained. receiving a first signal and a second signal;
obtaining a PTT as a first predictor value based on the first signal and the second signal;
extracting a feature from the second signal;
obtaining a second predictive variable value based on the first predictive variable value and the extracted feature; and
and generating a blood pressure estimation model based on the first predictive variable value and the second predictive variable value. 16. The method of claim 15,
The step of obtaining the value of the second predictor variable is
A method for generating a blood pressure estimation model, comprising normalizing a reference blood pressure, a first predictive variable value, and the characteristic. 16. The method of claim 15,
The step of obtaining the value of the second predictor variable is
generating a predictive variable acquisition model based on the reference blood pressure, the first predictive variable value, and the characteristic; and
and obtaining the second predictive variable value by using the predictive variable obtaining model. 18. The method of claim 17,
The step of generating the predictor acquisition model is
and generating a residual reference blood pressure by removing a component associated with a first predictive variable value from the reference blood pressure. 19. The method of claim 18,
The step of generating the residual reference blood pressure includes:
A method of generating a blood pressure estimation model for removing a component associated with the first predictive variable value by subtracting a dot product of the reference blood pressure and the first predictive variable value from the reference blood pressure. 19. The method of claim 18,
The step of generating the predictor acquisition model is
A method of generating a blood pressure estimation model for generating a predictive variable acquisition model using a singular value decomposition (SVD) decomposition technique based on the residual reference blood pressure and the characteristic. 21. The method of claim 20,
The step of generating the predictor acquisition model is
A method for generating a blood pressure estimation model, wherein a singular value decomposition is performed by decomposing a component obtained by dot product of the residual reference blood pressure and the feature such that a covariance between the residual reference blood pressure and the feature is maximized. 16. The method of claim 15,
The step of obtaining the value of the second predictor variable is
generating a residual reference blood pressure by removing a component associated with the first predictive variable value from the reference blood pressure;
generating a predictor variable acquisition model using a singular value decomposition (SVD) decomposition technique based on the residual reference blood pressure and the characteristics;
obtaining a second predictor variable value using the predictor variable acquisition model; and
and repeating the steps of generating the residual reference blood pressure by converting the residual reference blood pressure into a reference blood pressure until a predetermined number of second predictive variable values are obtained. a first sensor for measuring a first signal from a user;
a second sensor for measuring a second signal from the user; and
PTT is obtained as a first predictor variable value based on the first signal and the second signal, a feature is extracted from the second signal, and a second predictor variable value based on the extracted feature using a predictor variable acquisition model and estimating the blood pressure based on the first predictive variable value and the second predictive variable value using the blood pressure estimation model.

Images