KR102171640B1 - System and method for estimating oscillometric blood pressure based on deep learning - Google Patents

Abstract

A method and system for measuring blood pressure using a two-stage adaptive ensemble deep neural network are disclosed. In the blood pressure measurement method using an ensemble deep neural network, the step of extracting feature data from a blood pressure-related oscillometric signal measured by a blood pressure measuring device worn by a test subject, and a deep neural network (DNN) based on the extracted feature data. ) Setting input parameters of a model, setting auscultation blood pressure data measured when measuring the blood pressure of the subject in the blood pressure measurement device as target data, and setting the input parameters, deep neural network model, and target data It may include learning a nonlinear relationship between the blood pressure vibration signal measured by the blood pressure measuring device and the auscultation blood pressure data measured based on a stethoscope.

Description

Blood pressure measurement method and system using a two-stage adaptive ensemble deep neural network {SYSTEM AND METHOD FOR ESTIMATING OSCILLOMETRIC BLOOD PRESSURE BASED ON DEEP LEARNING}

Embodiments of the present invention relate to a technique for measuring blood pressure indicating a health state of a person using a learning model.

Blood pressure is used as one of the most important indicators of human health, and a method used as a standard technique for measuring blood pressure is an auscultatory method in which a trained doctor or nurse uses a stethoscope to measure blood pressure. In the case of a blood pressure measurement method using a stethoscope, due to the inconvenience of measuring by an expert, an oscillometric-based blood pressure measurement technology has recently been used so that it can be easily used at home, work, and hospital.

In the oscillometric-based blood pressure measurement technology, a blood pressure vibration signal is measured at the wrist or upper forearm using a pressure cuff, and blood pressure is measured using a predetermined systolic and diastolic ratio. For example, the oscillometric-based blood pressure measurement technology measures blood pressure based on the ratio of systolic blood pressure (SBP) and diastolic blood pressure (DBP) to maximum amplitude algorithm (MAA). . As such, the pre-defined systolic and diastolic ratio is a problem that the accuracy is lowered as the blood pressure that is constantly changing, such as blood pressure, is measured at a pre-defined empirical ratio based on the ratio of blood pressure testers before algorithm development. Exists. In other words, oscillometric-based blood pressure measurement is relatively inferior to blood pressure measurement using a stethoscope.

Accordingly, there is a need for a technology capable of measuring blood pressure based on an oscillometric signal and improving measurement accuracy without using a ratio defined for assessment.

Korean Patent Registration No. 10-1778533 relates to a method for estimating oscillometric blood pressure based on machine learning, and acquiring an envelope of an oscillometric waveform for a measured value based on a sample blood pressure measurement value, and using a non-parametric bootstrap to obtain an oscilloscope for the measured value. Acquire the pseudo-maximum amplitude based on the maximum amplitude of the envelope of the metric waveform, obtain a pseudo-envelope based on the envelope of the oscillometric waveform for the measured value using nonparametric bootstrap, and connect the pseudo-maximum amplitude with the pseudo-envelope, A technique for estimating the systolic blood pressure characteristic ratio and the diastolic blood pressure characteristic ratio for individual subjects using machine learning is described.

[1] S.Lee, and J.-H. Chang, Oscillometric Blood Pressure Estimation Based on Deep Learning, IEEE Transactions on Industrial Informatics, vol. 13, no. 2, pp. 461-472, Apr. 2017. [2] S. Lee, and J.-H. Chang, Deep Belief Networks Ensemble for Blood Pressure Estimation, IEEE ACCESS, vol. 5, pp. 9962-9972, Dec. 2017.

An embodiment of the present invention is to estimate the blood pressure of a test subject based on an oscillometric signal using a deep neural network (DNN) without prior definition of the systolic and diastolic ratios.

In addition, by applying the auscultation blood pressure data measured using the oscillometric signal and a stethoscope to the DNN, it is to relatively increase the accuracy of blood pressure measurement compared to when blood pressure is measured based on the oscillometric signal.

In the blood pressure measurement method using an ensemble deep neural network, the step of extracting feature data from a blood pressure-related oscillometric signal measured by a blood pressure measuring device worn by a test subject, and a deep neural network (DNN) based on the extracted feature data. ) Setting input parameters of a model, setting auscultation blood pressure data measured when measuring the blood pressure of the subject in the blood pressure measurement device as target data, and setting the input parameters, deep neural network model, and target data It may include learning a nonlinear relationship between the blood pressure vibration signal measured by the blood pressure measuring device and the auscultation blood pressure data measured based on a stethoscope.

According to one aspect, the extracting of the feature data includes generating an envelope signal connecting each peak of the signal, and the oscillometric signal and the envelope It may include the step of extracting the feature data from the signal signal.

According to another aspect, the feature data includes at least one of mean arterial pressure (MAP), maximum amplitude (MA), and area under the envelope (AE). and,

The target data may include a reference systolic blood pressure (RSBP) and a reference diastolic blood pressure (RDBP).

According to another aspect, the setting of the input parameter comprises duplicating the extracted feature data in a plurality based on a bootstrap technique, a bootstrap sample generated through the cloning, and the feature data It may include generating virtual feature data including, and setting the virtual feature data as the input parameter.

According to another aspect, in the learning step, primary learning may be performed on the input data based on a Restricted Boltzmann Machine (RBM) probability model.

According to another aspect, in the learning step, a weight for secondary learning may be set based on a result of performing the primary learning.

According to another aspect, the bootstrap sample may be updated based on virtual feature data on which secondary learning is performed based on the set weight.

In a blood pressure measurement system using an ensemble deep neural network, a feature data extracting unit for extracting feature data for a blood pressure-related oscillometric signal measured by a blood pressure measuring device worn by a test subject, and a feature data extracting unit for extracting feature data; A setting unit for setting input parameters of a neural network (DNN) model, and setting auscultation blood pressure data measured when measuring the blood pressure of the test subject in the blood pressure measuring device as target data, and the set input parameters, a deep neural network model, It may include a learning control unit that learns a nonlinear relationship between the blood pressure vibration signal measured by the blood pressure measuring device and the auscultation blood pressure data measured based on the stethoscope based on target data.

According to one aspect, the feature data extracting unit generates an envelope signal connecting each peak of the signal, for the oscillometric signal, and feature data from the oscillometric signal and the envelope signal. Can be extracted.

According to another aspect, the feature data includes at least one of mean arterial pressure (MAP), maximum amplitude (MA), and area under the envelope (AE). And, the target data may include a reference systolic blood pressure (RSBP) and a reference diastolic blood pressure (RDBP).

According to another aspect, the setting unit duplicates a plurality of extracted feature data based on a bootstrap technique, and a bootstrap sample generated through the cloning and virtual feature data including the feature data And set the virtual feature data as the input parameter.

According to another aspect, the learning control unit may perform primary learning based on a Restricted Boltzmann Machine (RBM) probability model for the input data.

According to another aspect, the learning controller may set a weight for secondary learning based on a result value of performing the primary learning.

According to another aspect, the bootstrap sample may be updated based on virtual feature data on which secondary learning is performed based on the set weight.

According to the embodiments of the present invention, one embodiment of the present invention estimates the blood pressure of the test subject based on the oscillometric signal using a deep neural network (DNN), without prior definition of the systolic and diastolic ratios. Blood pressure can be measured.

In addition, by applying the auscultation blood pressure data measured using an oscillometric signal and a stethoscope to the DNN, accuracy can be improved compared to when blood pressure is measured based on the oscillometric signal. That is, when measuring based on an oscillometric signal, it is possible to improve the accuracy of the blood pressure measurement, which is relatively lower than when measured with a stethoscope.

1 is a diagram showing waveforms of an oscillometric signal and an envelope signal according to an embodiment of the present invention.
2 is a flow chart showing a blood pressure measurement method using a two-stage ensemble deep neural network according to an embodiment of the present invention.
3 is a block diagram showing the internal configuration of a blood pressure measuring system according to an embodiment of the present invention.
4 is a graph showing a cumulative function distribution (CDF) of virtual feature data according to an embodiment of the present invention.
5 is a diagram showing the structure of a two-stage ensemble deep neural network (DNN) according to an embodiment of the present invention.
6 is a diagram showing the performance of target data and a two-stage adaptive ensemble deep neural network model according to an embodiment of the present invention.
7 is a graph showing a standard error error (SDE) according to an increase in a virtual blood pressure sample according to an embodiment of the present invention.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

The present invention relates to a technology for measuring blood pressure using a two-stage adaptive ensemble deep neural network (DNN), in which an oscillometric signal, which is a vibration signal measured by a blood pressure measuring device, and a stethoscope are used by a nurse or a doctor The present invention relates to a technology for estimating blood pressure by applying auscultation blood pressure data, which is blood pressure measurement data, to a deep neural network. In particular, it relates to a technique for learning a complex nonlinear relationship between a blood pressure vibration signal (ie, an oscillometric signal) and auscultation blood pressure data measured by a blood pressure measuring device by using a deep neural network model based on bootstrap technology. In other words, it solves the weaknesses based on advanced neural network technology due to insufficient amount of input feature data, and solves the problem of random initialization assignment, which still has many training data sets, weights, and biases, which can cause unstable estimation in the pre-training stage. It relates to a technology to solve through an adaptive ensemble deep neural network using bootstrap technology.

In the present embodiments, the'blood pressure measuring device' is a device that measures blood pressure of a test subject by being worn on the wrist, forearm of the test subject, for example, manufactured in the form of a wearable watch, or using a wearable smart pressure sensor. It can be implemented in the form of measuring the blood pressure of the test subject by extracting blood pressure and heart rate signals based on it.

The full name of terms used in the two-stage ensemble deep neural network-based learning algorithm presented in the present embodiments may be as shown in Table 1 below.

1 is a diagram showing waveforms of an oscillometric signal and an envelope signal according to an embodiment of the present invention.

Referring to 110 of FIG. 1, as the blood pressure of the test subject is measured, the cuff pressure received from the blood pressure meter may be indicated.

In 120 of FIG. 1, an oscillometric envelop signal 121 may be generated by connecting each peak of an oscillometric wave signal. Here, the oscillometric signal may be obtained by filtering and detrending the cuff pressure 110 as a target.

2 is a flow chart showing a blood pressure measurement method using a two-stage ensemble deep neural network according to an embodiment of the present invention, and FIG. 3 is a block showing the internal configuration of a blood pressure measurement system according to an embodiment of the present invention. Is also.

The blood pressure measurement system 300 of FIG. 3 may include a feature data extraction unit 310, a setting unit 320, and a learning control unit 330, and the blood pressure measurement system 300 is a blood pressure measurement device worn by the test subject. The blood pressure vibration signal being measured from (not shown) may be received through a wired or wireless network. For example, it is possible to receive a blood pressure vibration signal through short-range wireless communication such as Bluetooth and Zigbee. In addition, each of the steps (210 to 240) of FIG. 2 is performed by the feature data extraction unit 310, the setting unit 320, and the learning control unit 330, which are components of the blood pressure measurement system 300 of FIG. Can be done.

In FIGS. 2 and 3, a case in which a wearable device in the form of a wearable watch is used as a blood pressure measurement device is described as an example, but this corresponds to an example, and is mounted on the forearm of the test subject in addition to the form of the wearable watch. A device including a wearable smart pressure sensor may be used to measure the value.

In step 210, the feature data extractor 310 may extract feature data from the blood pressure related oscillometric signal measured by the wearable device worn by the test subject.

For example, the feature data extractor 310 may receive a blood pressure vibration signal of a test subject, which is measured for a predetermined time (eg, 1 minute, 5 minutes, etc.), from the wearable device. Here, the blood pressure vibration signal may include a cuff pressure. Then, the feature data extraction unit 310 may generate an oscillometric wave signal by performing signal processing such as filtering and trend analysis on the blood pressure vibration signal.

In step 211, the feature data extractor 310 may generate an envelope signal by connecting each peak of the signal to the oscillometric signal.

In step 212, the feature data extracting unit 310 may extract feature data for blood pressure estimation from the oscillometric signal and the envelope signal.

As an example, the feature data extraction unit 310 includes an estimated mean arterial pressure (MAP), a maximum amplitude (MA), an area under the envelope (AE), etc. It can be extracted as feature data. In addition to these three types of feature data, the feature data extraction unit 310 includes the above non-patent document [1] S.Lee, and J.-H. Chang, Oscillometric Blood Pressure Estimation Based on Deep Learning, IEEE Transactions on Industrial Informatics, vol . 13, no. 2, pp. 461-472, Apr. 2017. and [2] S. Lee, and J.-H. Chang, Deep Belief Networks Ensemble for Blood Pressure Estimation, IEEE ACCESS, vol. 5, pp. 9962-9972, Dec. Based on the feature data extraction method presented in 2017. 11 types of feature data can be extracted from the oscillometric signal and the envelope signal.

In step 220, the setting unit 320 may set input parameters of a deep neural network (DNN) model based on the extracted feature data. In this case, a large amount of data is required for learning the deep neural network, and the number of data may be insufficient for learning the deep neural network only with data obtained through blood pressure measurement of a subject. Accordingly, the setting unit 320 may additionally generate data sufficient to perform deep neural network learning using a bootstrap technique.

In step 221, the setting unit 320 may generate a bootstrap sample by replicating the extracted feature data in a plurality based on the bootstrap technique.

In step 222, the setting unit 320 may generate virtual feature data including the generated bootstrap sample and feature data.

As an example, blood pressure data can be obtained from 425 blood pressure measurement signals and auscultation blood pressure data used as labels five times each from 85 test subjects based on the international protocol, and the above non-patent document [1] S.Lee, and J .-H. Chang, Oscillometric Blood Pressure Estimation Based on Deep Learning, IEEE Transactions on Industrial Informatics, vol. 13, no. 2, pp. 461-472, Apr. 2017. and [2] S. Lee, and J.-H. Chang, Deep Belief Networks Ensemble for Blood Pressure Estimation, IEEE ACCESS, vol. 5, pp. 9962-9972, Dec. Based on the virtual feature data creation method presented in 2017. As a large amount of data is needed to train the deep neural network model, a parameter-based bootstrap using statistical feature values (ie, mean and standard deviation) of the selected feature data By applying the technique, it is possible to generate 8500 virtual feature data for 1000 times for 85 people. For example, as shown in FIG. 4, it can be seen that the cumulative distribution (CDF) of the virtual feature data (ie, the virtual feature vector) has a pattern similar to the Gaussian distribution.

In step 223, the setting unit 320 may set the virtual feature data as an input parameter of the deep neural network model.

In step 230, the setting unit 320 may set auscultation blood pressure data as target data.

For example, when measuring a subject's blood pressure in a blood pressure measuring device, at the same time or immediately as measured in a blood pressure measuring device, a certain number of blood pressure may be measured for a predetermined period of time specified in advance using a stethoscope by an expert such as a nurse or doctor. The measured blood pressure information may be input through a user terminal (eg, smartphone, PC, etc.) possessed by an expert, and auscultation blood pressure data may be generated as an average of the input data. For example, when measured over five times per minute, auscultation blood pressure data may be generated as an average of blood pressure measurement data using a stethoscope for five times. In addition, the setting unit 320 may set the auscultation blood pressure data as target data of a deep neural network (DNN) model. For example, RSBP and RDBP shown in Table 2 above can be set as target data.

In step 240, the learning control unit 330 may learn a nonlinear relationship between a blood pressure vibration signal (ie, an oscillometric signal) measured in a blood pressure measurement value and auscultation blood pressure data based on the set input parameter, the deep neural network model, and target data. .

Table 2 below shows the results of validating the original feature data through statistical analysis of virtual feature data (ie, virtual feature vector), and can indicate the degree of convergence between the actual feature data and the virtual feature data. Here, the actual feature data may represent data measured by the wearable measurement device worn by the test subject, and the virtual feature data may represent data generated through duplication based on the actual feature data.

For example, Table 2 shows a result of verifying the validity using the Kolmogorov-Smirnov test technique, and may indicate a prediction error between actual feature data and virtual feature data. The degree of convergence between the actual feature data and the virtual feature data shown in Table 2 can be expressed as Equation 1 below.

[Equation 1]

는 실제 특징데이터(즉, 특징벡터) X를 기반으로 부트스트랩 기법을 적용한 조건부 확률을 나타내고, 가상특징데이터의 평균값과 실제 특징데이터의 차이는 실제 특징데이터의 평균값과 실제 특징데이터 값의 차이와 동일하며, 좌측

과 우측

의 차이가 거의 0에 수렴하는 것을 알 수 있다. 위의 수학식 1에서

는 진실(true)을 나타내는 파라미터에 해당할 수 있다.In Equation 1 above,

Represents the conditional probability of applying the bootstrap technique based on the actual feature data (i.e. feature vector) X, and the difference between the average value of the virtual feature data and the actual feature data is the same as the difference between the average value of the actual feature data and the actual feature data value. And left

And right

It can be seen that the difference of converges to almost zero. In Equation 1 above

May correspond to a parameter indicating true.

가 계산될 수 있다.Based on Equation 1 above, as shown in Equation 2 below

Can be calculated.

[Equation 2]

Referring to Equation 2 and Table 2 above, a difference between the average value of the actual feature data (MAP, AE) and the average value of the virtual feature data may be calculated as a prediction error. In Table 2, in the case of RSBP and RDBP, which are target data, the error between the actual average value and the virtual average value may converge to almost zero.

When using the feature data shown in Table 2 above, an error of the standard deviation of the virtual feature data may be calculated based on Equation 3 below.

[Equation 3]

Referring to Table 2 above, the error (MAP: 0.016) of the standard deviation of the virtual feature data calculated based on Equation 3 above is relatively more compared with the error (MAP: 0.036) of the standard deviation of the actual feature data. It can be seen that it has a small value.

5 is a diagram showing the structure of a two-stage ensemble deep neural network (DNN) according to an embodiment of the present invention.

In 510 of FIG. 5, an arrow may indicate a flow direction of information through a Restricted Boltzmann Machine (RBM).

)과 출력층(Y) 상이에 복수개의 유닛, 즉, 은닉층(H)이 계층적으로 존재할 수 있다. 최상층의 층간은 무향 에지(undirected edge)로 연결되고, 나머지 층간은 아래방향(directed edge)로 연결될 수 있다. 그리고, 정보는 위에서 아래 방향으로 전달되며, 최하위층인 입력층

의 상태구조가 결정될 수 있다.Referring to 520 of FIG. 5, the input layer (

) And the output layer Y, a plurality of units, that is, the hidden layer H, may be hierarchically present. Interlayers of the uppermost layer may be connected by an undirected edge, and the remaining interlayers may be connected by a directed edge. And, the information is transferred from top to bottom, and the input layer is the lowest level.

The state structure of can be determined.

로 표현되고, 심화신경망의 모든 유닛(즉, 은닉층)의 결합 확률분포는 아래의 수학식 4와 같이 표현될 수 있다.In FIG. 5, the second-stage ensemble deep neural network model may classify and estimate a state of the highest layer determined when virtual feature data is set in the input layer as a representation of the data. Units of each layer (i.e., hidden layer) upward from the lowest layer, the input layer

It is expressed as, and the combined probability distribution of all units (ie, hidden layers) of the deep neural network can be expressed as Equation 4 below.

[Equation 4]

부터

까지의 층(layer)은 이웃하는 윗층과 아래 방향 유향 에지(undirected edge)로 연결되고, 연결된 층 사이의 조건부 분포는 위의 수학식 4에서

에 해당할 수 있다. 그리고, 최상위층은 RBM(Restricted Boltzmann Machine)의 분포함수

로 표현될 수 있다.Input layer, the lowest level

from

The layers up to are connected by adjacent upper and lower undirected edges, and the conditional distribution between the connected layers is shown in Equation 4 above.

May correspond to. And, the top layer is the distribution function of RBM (Restricted Boltzmann Machine)

It can be expressed as

Table 3 below can show an algorithm of a two-stage adaptive deep neural network regression model for effectively finding a nonlinear relationship between input data and target data.

In Table 3, input data is virtual feature data generated for oscillometric-based blood pressure estimation, and target data may correspond to systolic and diastolic blood pressure data, which is auscultation blood pressure data. Here, the virtual feature data set as input data may be generated using a bootstrap technique. For example, during pre-training of the deep neural network, unstable blood pressure estimation results may appear due to lack of training data, and instability of the standard deviation of learning, fit and estimation error may occur due to the initial random assignment of variables such as weights and biases during pre-training. I can. Accordingly, by training the deep neural network model in two steps using the bootstrap technique, the error and the standard deviation of the mean absolute error can be relatively lower than when estimating blood pressure using the existing single deep neural network regression model. Referring to Table 3 above, in steps 1 to 2 (step 1-2), feature data (i.e., actual feature data) is extracted from the oscillometric signal, and the feature data is input parameters (i.e., deep neural network model). Input data). In addition, auscultation blood pressure data may be set as target data.

)를 계산할 수 있다.Subsequently, in step 3, the average and standard deviation of specific data and target data (

) Can be calculated.

In step 4, a predetermined initial weight value is assigned, and virtual feature data may be calculated, that is, generated, by applying the bootstrap technique in steps 7 to 9 (step 7-9). For example, a bootstrap sample may be generated by duplicating actual feature data, and an average value of the generated bootstrap samples may be calculated. In addition, virtual feature data may be generated to include the generated bootstrap sample and actual feature data. Then, in step 10, pre-training (ie, primary learning) may be performed based on the virtual feature data and the RBM.

In step 12, the result of pre-training, that is, the average value of each output value (eg, ensemble 50) that has undergone primary learning may be calculated and set as outputs of weights and biases. .

In this way, after pre-training, that is, primary learning, is performed through steps 1 to 12, secondary learning may be performed.

In steps 16 to 18 (step 16-18), initial virtual feature data generated in the pre-training step may be allocated for secondary learning. In steps 19 to 25 (step 19-25), weights for secondary learning may be updated based on initial virtual feature data generated through the primary learning. And, by applying a two-step technique based on the error, the distribution of the actual training data set can be adaptively adjusted before each new bootstrap sample is acquired. Here, in the estimator using each deep neural network, the bootstrap training sample may be updated according to the weighted training sample as expressed in step 19. That is, the bootstrap sample can be obtained from a distribution that is repeatedly updated by a relative error function for a subsequent estimator. In addition, the instance and weight of each estimator may be adjusted according to the complexity. That is, the probability that an instance with a large previous error appears in the next bootstrap sample may increase. Accordingly, the two-stage adaptive ensemble deep neural network model can reduce the variance of the mean error and increase the reliability of the result of the deep neural network regression estimator. The error function of the deep neural network can be expressed as Equation 5 below.

[Equation 5]

과 목표혈압값(즉, 청진 혈압데이터)

간의 차이를 기반으로 최소평균오차가 계산될 수 있다.Referring to Equation 5 above, the blood pressure value estimated based on the deep neural network model

And target blood pressure value (ie, auscultation blood pressure data)

The minimum average error can be calculated based on the difference between.

[Equation 6]

Equation 6 above can be used to calculate the updated weight. Based on the ensemble weights and biases of the first stage pre-training, that is, the weights and biases calculated through the first-order learning, a backpropagation technique to reduce an estimation error and a weight update for the error may be performed. In this case, based on Equation 6 above, the weight of the error and the bias variable may be updated.

In Equation 6, ε denotes a learning rate, η denotes a momentum variable, and L denotes the number of hidden layers.

6 is a diagram showing the performance of target data and a two-stage adaptive ensemble deep neural network model according to an embodiment of the present invention.

That is, FIG. 6 may show the standard error result based on the data shown in Table 5 below and the 2-step adaptive ensemble deep neural network model (DBN-DNN) compared to target data (RSBP and RDBP), which are target blood pressure values.

In FIG. 6, it is based on oscillometric blood pressure measurement data from Biosign Technology Co., Ltd. located in Toronto, Canada used in an experiment for performance evaluation. Here, blood pressure data was approved by a regional research committee in Ontario, Canada, and all participants submitted consent for blood pressure measurement based on the criteria of the research ethics committee. Blood pressure measurement samples required for the experiment were extracted from 85 subjects aged 12 to 80 years without a history of cardiovascular disease, and 85 subjects could include 48 men and 37 women. Oscillometric blood pressure measurement is performed on wrist-worn blood pressure for all subjects according to the standard recommendations of the American national standard manual, electronic or automated sphygmonanometers/association for the advancement of medical instrumentation (ANSI/AAMI). It is assumed that blood pressure measurements were made 5 times using a measuring device, and (5 measurements, 85 subjects = 425 blood pressure samples) were obtained. In addition, it is assumed that two nurses at the same time measure systolic and diastolic blood pressure using an auscultation sphygmomanometer (ie, a stethoscope) and then use the calculated average value as the target blood pressure value. As a result of the experiment, the systolic blood pressure of all the subjects was 78-147 mmHg, and the diastolic blood pressure was 42-99 mmHg. In this case, the blood pressure measurement process may be performed by measuring five times for one experimenter, and taking a one-minute break after measuring once. Table 4 below may show the configuration of variables of a two-stage adaptive deep neural network model.

Referring to Table 4 above, the variables used in the two-stage adaptive deep neural network are the number of input feature vectors 11, the vomiting blood pressure value vector (diastolic and systolic) 2, the number of samples of each virtual feature vector 100, and the actual number of samples. 5, hidden layer 3, etc. may be included. At this time, for the performance evaluation of the two-stage adaptive deep neural network, the performance can be verified based on the mean error (ME), the standard deviation of the mean error (SDE), and the mean absolute error (MAE) shown in Table 5 below. .

In the AAMI protocol, the blood pressure measuring device is compared with the auscultation blood pressure value and recognized as a standard error (SDE) of 8 mmHg within the mean error (ME) of 5 mmHg. Referring to the performance evaluation of the blood pressure measurement methods in Table 5, it can be seen that the mean error (ME) is within 5 mmHg in all methods, and the standard error (SDE) maximum amplitude algorithm (MAA) 9.03mmHg and 7.40 Diastolic and systolic standard errors of mmHg can be checked. And, feed-forward neural network (FFNN) 7.61mmHg and 6.79mmHg, support vector regression (SVR) 7.11mmHg and 6.14mmHg, Gaussian mixture regression (GMR) 6.12 mmHg and 5.29mmHg, single deep belief network-deep neural network (DBN-DNN) 6.20mmHg and 5.26mmHg, and two-step adaptive ensemble deep neural network model (DBN-DNN ensemble) of 5.74mmHg and 4.68mmHg You can check the error performance. According to Table 5, it can be seen that the two-stage adaptive ensemble deep neural network model (DBN-DNN ensemble) shows a reduction in standard error of about 11% and about 18% compared to the single deep neural network regression model. In addition, referring to the root mean square error (RMSE) in Table 5, it can be seen that the 2-step adaptive ensemble deep neural network model (DBN-DNN ensemble) exhibits a lower error than other methods in error measurement.

7 is a graph showing a standard error error (SDE) according to an increase in a virtual blood pressure sample according to an embodiment of the present invention.

Using the British hypertension society (BHS) evaluation method shown in Table 6 below, a grade can be assigned for each mean absolute error (MAE) calculated based on 425 data of blood pressure measurement samples of all subjects. For example, the measurement methods are three groups, and MAE 5mmHg or less, 10mmHg or less, and 15mmHg or less are calculated for each method, and as a result of the calculation, a grade from A to D may be given. At this time, if the blood pressure sample measured based on the two-stage adaptive ensemble deep neural network (DBN-DNN) model indicates 60% of MAE 5mmHg or less, 85% of 10mmHg or less, and 95% of 15mmHg or less, A grade can be given. And, based on the BHS standard, both DBN-DNN and other blood pressure measurement methods are grade A, but the performance measurement results of DBN-DNN are 81.18%, 96.24%, and 99.29% in the systolic (SBP) of 71.06%, 90.82%, and 95.53%. , It can be seen that it has relatively superior performance than other methods.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, and the like alone or in combination. The program instructions recorded on the medium may be specially designed and configured for the embodiment, or may be known and usable to those skilled in computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. -A hardware device specially configured to store and execute program instructions such as magneto-optical media, and ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine language codes such as those produced by a compiler but also high-level language codes that can be executed by a computer using an interpreter or the like. The hardware device described above may be configured to operate as one or more software modules to perform the operation of the embodiment, and vice versa.

As described above, although the embodiments have been described by the limited embodiments and drawings, various modifications and variations are possible from the above description by those of ordinary skill in the art. For example, the described techniques are performed in a different order from the described method, and/or components such as a system, structure, device, circuit, etc. described are combined or combined in a form different from the described method, or other components Alternatively, even if substituted or substituted by an equivalent, an appropriate result can be achieved.

Therefore, other implementations, other embodiments, and claims and equivalents fall within the scope of the claims to be described later.

Claims (14)

In the blood pressure measuring method using an ensemble deep neural network of a blood pressure measuring device,
Extracting feature data from the blood pressure related oscillometric signal measured by the blood pressure measuring device worn by the test subject;
Setting input parameters of a deep neural network (DNN) model based on the extracted feature data;
Setting auscultation blood pressure data measured when the blood pressure measurement device measures the blood pressure of the test subject as target data; And
Learning a nonlinear relationship between the blood pressure vibration signal measured by the blood pressure measuring device and the auscultation blood pressure data measured based on a stethoscope based on the set input parameter, the deep neural network model, and target data
Including,
The learning step,
Performing primary learning based on a predetermined initial weight;
Setting a weight based on a result of performing the primary learning; And
Performing secondary learning based on the set weight
Including,
The step of performing the primary learning,
Generating a bootstrap sample by duplicating the extracted feature data;
Generating virtual feature data including the generated bootstrap sample and the extracted feature data; And
Performing the primary learning on the generated virtual feature data
Including,
The step of performing the secondary learning,
Updating the generated virtual feature data based on the set weight; And
Performing the secondary learning on the updated virtual feature data
Blood pressure measurement method comprising a. The method of claim 1,
The step of extracting the feature data,
Generating an envelope signal in which each peak of the signal is connected to the oscillometric signal; And
Extracting feature data from the oscillometric signal and the envelope signal
Blood pressure measuring method comprising a. The method of claim 1,
The feature data includes at least one of a mean arterial pressure (MAP), a maximum amplitude (MA), and an area under the envelope (AE),
The target data includes RSBP (reference systolic blood pressure) and RDBP (reference diastolic blood pressure)
Blood pressure measurement method, characterized in that. delete The method of claim 1,
The step of performing the primary learning,
A blood pressure measurement method for performing the primary learning based on a Restricted Boltzmann Machine (RBM) probability model for the generated virtual feature data. delete delete In the blood pressure measurement system using the ensemble deep neural network,
A feature data extracting unit for extracting feature data from the blood pressure-related oscillometric signal measured by the blood pressure measuring device worn by the test subject;
A setting unit for setting input parameters of a deep neural network (DNN) model based on the extracted feature data, and setting auscultation blood pressure data measured when the blood pressure measurement device measures the blood pressure of the test subject as target data; And
Learning control unit for learning a nonlinear relationship between the blood pressure vibration signal measured by the blood pressure measuring device and the auscultation blood pressure data measured based on a stethoscope based on the set input parameter, the deep neural network model, and target data
Including,
The learning control unit,
Based on the pre-specified initial weight, first-order learning is performed,
Based on the result of performing the primary learning, a weight is set,
Based on the set weight, secondary learning is performed,
The learning control unit,
Create a bootstrap sample by duplicating the extracted feature data,
Generate virtual feature data including the generated bootstrap sample and the extracted feature data,
Performing the primary learning on the generated virtual feature data,
The learning control unit,
Update the generated virtual feature data based on the set weight,
A blood pressure measurement system that performs the secondary learning on the updated virtual feature data. The method of claim 8,
The feature data extraction unit,
Generating an envelope signal connecting each peak of the signal as a target of the oscillometric signal, and extracting feature data from the oscillometric signal and the envelope signal
Blood pressure measurement system, characterized in that. The method of claim 8,
The feature data includes at least one of a mean arterial pressure (MAP), a maximum amplitude (MA), and an area under the envelope (AE),
The target data includes RSBP (reference systolic blood pressure) and RDBP (reference diastolic blood pressure)
Blood pressure measurement system, characterized in that. delete The method of claim 8,
The learning control unit,
Performing the first-order learning based on the RBM (Restricted Boltzmann Machine) probability model for the generated virtual feature data
Blood pressure measurement system, characterized in that. delete delete

Images