TWI733378B - Method of establishing blood pressure model - Google Patents

Abstract

A method of establishing blood pressure model includes obtaining a plurality of first physiologic data of a plurality of general users and a plurality of first blood pressure data of the plurality of general users, performing a deep learning algorithm to establish a general blood pressure model according to the plurality of first physiologic data and the plurality of first blood pressure data, wherein the general blood pressure model has a parameter set and a loss function, obtaining a second physiologic data of a specified user and a second blood data of the specified user, generating a blood pressure estimation according to the second physiologic data and the parameter set, calculating an error according to the blood pressure estimation, the second physiologic data and the loss function, and adjusting the parameter set to establish a specified blood pressure model according to the error.

Description

Method of establishing blood pressure model

The present invention relates to blood pressure measurement, particularly a blood pressure measurement system and method based on electrocardiography (ECG), photoplethysmography (PPG) and pulse transit time (Pulse Transit Time, PTT) .

Cardiovascular-related diseases have been shown to be highly correlated with heart rate and blood pressure. Uncontrolled high blood pressure can lead to heart attack, stroke, heart failure or other serious life threats. Therefore, accurate blood pressure measurement is necessary to prevent accidents. According to the standards recognized by the American National Standards Institute (ANSI), the Association for the Advancement of Medical Instrumentation (AAMI) and the International Organization for Standardization (ISO) in 2018, blood pressure The acceptable error of the measurement is 10 millimeters of mercury (mm Hg) or less, and the estimated probability of this error is at least 85%.

At present, blood pressure measurement methods can be divided into cuff-based measurement and cuffless measurement. The cuff blood pressure monitor is an invasive single measurement. Because the user's single arm must be fastened with a cuff for a period of time to obtain an accurate blood pressure value, it is not suitable for long-term (such as all-day) blood pressure measurement. However, the cuff sphygmomanometer can accurately measure the user's blood pressure. On the other hand, a cuffless sphygmomanometer is equipped with a sensor on the user's body. The sensor is used to obtain the user's ECG, PPG, and PTT sensing data, and then convert the sensing data into Blood pressure value. Since the volume of the sensor is smaller than that of the cuff, the cuffless sphygmomanometer will not cause interference to the user and can measure continuously for a long time. However, compared with the cuff-type sphygmomanometer, the blood pressure value measured by the cuffless blood pressure meter is less accurate. In addition, the existing cuffless sphygmomanometer needs to collect the user's sensing data in a variety of situations (for example: walking, sitting, exercise) in order to provide a relatively accurate blood pressure measurement value. For users, it takes extra physical strength and time to provide sensing data in different situations.

In view of this, the present invention proposes a method for establishing a designated blood pressure model, which improves the accuracy of blood pressure measurement and reduces interference use while retaining the advantages of the cuffless sphygmomanometer that it can be worn on the body and can be continuously measured. The status of the person.

The method for establishing a blood pressure model described in an embodiment of the present invention is suitable for a blood pressure measuring device. The method for establishing a blood pressure model includes: obtaining a plurality of first physiological data of a plurality of general users and a plurality of first blood pressure data of these general users; executing a deep learning algorithm based on the first physiological data and the first blood pressure data To establish a universal blood pressure model, this universal blood pressure model has a parameter set and a loss function; obtain the second physiological data of the specified user and the second blood pressure data of the specified user; generate estimated blood pressure data based on the second physiological data and parameter set; Estimate the blood pressure data, the second blood pressure data, and the loss function to calculate the e error; and adjust the parameter set according to the error to establish a specified blood pressure model.

The above description of the disclosure and the following description of the embodiments are used to demonstrate and explain the spirit and principle of the present invention, and to provide a further explanation of the scope of the patent application of the present invention.

The detailed features and advantages of the present invention will be described in detail in the following embodiments. The content is sufficient to enable anyone familiar with the relevant art to understand the technical content of the present invention and implement it accordingly, and according to the content disclosed in this specification, the scope of patent application and the drawings. Anyone who is familiar with relevant skills can easily understand the purpose and advantages of the present invention. The following examples further illustrate the viewpoints of the present invention in detail, but do not limit the scope of the present invention by any viewpoint.

The method for establishing a blood pressure model proposed by the present invention is suitable for wearable and non-invasive blood pressure measurement devices.

Please refer to Figure 1. Fig. 1 shows a system design diagram of a method for establishing a blood pressure model according to an embodiment of the present invention. In the present invention, a general blood pressure model is obtained by collecting multiple first physiological data and multiple first blood pressure data of the tested group and training through deep learning neural network. For each individual individual, namely designated user 1, designated user 2, and designated user 3 shown in Figure 1, after obtaining a wearable blood pressure measurement device with a general blood pressure model, they can further based on their own physiological data and blood pressure data This general blood pressure model is fine-tuned to establish a designated blood pressure model 1, a designated blood pressure model 2, and a designated blood pressure model 3 suitable for itself. After the designated blood pressure model 1 is established, designated user 1 can use the wearable blood pressure measurement device loaded into this designated blood pressure model 1 and measure its own systolic blood pressure 1 and diastolic blood pressure 1. This method is also applicable to designated user 2 and designated user User 3.

Please refer to Figure 2. Fig. 2 shows a flowchart of a method for establishing a blood pressure model according to an embodiment of the present invention.

Please refer to step S1. Obtain plural first physiological data of plural general users and plural first blood pressure data of these general users. In one embodiment, the source of the first physiological data and the first blood pressure data is a public data set. The public data set can be, for example, the University of Queensland vital signs dataset (University of Queensland vital signs dataset). However, the present invention is not limited to this, as long as it comes from the first physiological data and first blood pressure data provided by multiple subjects. Both can be used as the data source of this step S1. The first physiological data is the original electrocardiography (ECG) signal, the original photoplethysmography (PPG) signal, the synchronization signal containing the above two or the pulse wave transmission time calculated based on the above two (Pulse Transit). Time, PTT). The first blood pressure data includes a measured value of systolic blood pressure and a measured value of diastolic blood pressure. In one embodiment, the measured value of systolic blood pressure and the measured value of diastolic blood pressure are obtained by measuring a general user with a cuff sphygmomanometer. In one embodiment, the same general user can be measured multiple times to obtain the first physiological data and the first blood pressure data. For example, the number of general users is 20, and 15 ECG signals and 15 PPG signals are obtained from each person as the first physiological data, and 15 systolic blood pressure measurements and 15 diastolic blood pressure values are obtained from each person. The measured value is used as the first blood pressure data.

After step S1 and before step S2, a pre-processing procedure can be performed on the first physiological data obtained in step S1. For example, after the original signal of the first physiological data is obtained, it is divided into multiple first physiological data segments in a unit of 5 seconds and a linear filter (such as Savitzky-Golay filter) is used to retain the sharper signals in the signal. Edges or peaks in the signal. Then remove the first physiological data segment that has no obvious peak or low variability.

Please refer to step S2: execute a deep learning algorithm based on the first physiological data and the first blood pressure data to establish a general blood pressure model. In an embodiment, the first physiological data is an ECG signal, a PPG signal, or a synchronization signal including the two, and the deep learning algorithm uses a multilayer perceptron (MLP) as the convolution of the regression factor (Regressor) Neural Networks (Convolutional Neural Networks, CNN). In other words, the general blood pressure model can be trained with the ECG signal alone, with the PPG signal alone, or with the time synchronization signal of both the ECG signal and the PPG signal. In another embodiment, the first physiological data is the pulse wave transmission time, which is based on the peak value of ECG and the peak value of PPG to calculate the interval between the two, and the general blood pressure model obtained by deep learning algorithm training adopts linear regression (Linear regression).

After step S2 is completed, the general blood pressure model can be loaded into the blood pressure measurement device to measure a single designated user. In one embodiment, the general blood pressure model can be directly used to measure the specified user. In another embodiment, steps S3 to S6 in FIG. 2 can be executed continuously. Steps S3 to S6 illustrate how to adjust the general blood pressure model to the specified blood pressure model according to the measurement value of the specified user.

Please refer to step S3: Obtain the second physiological data and the second blood pressure data of the designated user. In one embodiment, a sensor on the blood pressure measuring device is used to measure the second physiological data for the designated user. The second physiological data is the original electrocardiogram signal, the original optical volume change tracing method signal, the synchronization signal including the above two, or the pulse wave transmission time calculated based on the above two. The second blood pressure data includes a measurement value of systolic blood pressure and a measurement value of diastolic blood pressure, and the two measurement values are, for example, measured by a designated user using cuff blood pressure measurement. In one embodiment, step S3 may be repeatedly executed to obtain multiple second physiological data and second blood pressure data of the designated user. The second physiological data and the second blood pressure data are used to fine-tune the general blood pressure model in the process of steps S4 to S6. Collecting more data of designated users can adjust the general blood pressure model to a designated blood pressure model that is more suitable for the designated user. In another embodiment, in addition to obtaining the second physiological data, in step S3, reference physiological data related to the designated user can be obtained. For example, the reference physiological data includes: the body temperature of the designated user when measuring the second physiological data, the tightness of the cuff or temperature when the designated user measures the second blood pressure data, the height and weight of the designated user, etc. These data can be used for Establish designated blood pressure models based on different scenarios. However, the present invention is not limited to the above examples.

Please refer to step S4: Generate the first estimated blood pressure data based on the second physiological data and the first parameter set of the general blood pressure model. The general blood pressure model has a parameter set and a loss function. When a neural network is used to establish a general blood pressure model, the first parameter set is a set of network weights. When the general blood pressure model uses linear regression, the first parameter set is a set of network weights. A parameter set is a set of various parameters of a linear function. In one embodiment, the first estimated blood pressure data is the output obtained after substituting the second physiological data into the first parameter set of the general blood pressure model. The output value can be systolic blood pressure or diastolic blood pressure, based on the first parameter used in previous training. Depends on blood pressure data. In another embodiment, if multiple pieces of second physiological data of the designated user are obtained in step S3, step S4 can be repeated to obtain multiple pieces of first estimated data.

Please refer to step S5: Calculate the first error based on the first estimated blood pressure data, the second blood pressure data, and the loss function. In one embodiment, the calculation method of the first error is as follows:

（式一）

(Formula 1)

為通用血壓模型的損失函數，

為指定用戶的第二血壓資料，其係以其他血壓計（如袖帶式血壓計）量測到的收縮壓或舒張壓之數值。

為估計血壓資料。 in

Is the loss function of the general blood pressure model,

It is the second blood pressure data of the designated user, which is the value of systolic or diastolic blood pressure measured by other blood pressure monitors (such as cuff blood pressure monitors).

To estimate blood pressure data.

Please refer to step S6: adjust the first parameter set according to the first error to establish a specified blood pressure model with the second parameter set. For example, if the general blood pressure model is a linear model, the data points drawn based on the second physiological data and the second blood pressure data may not fall on the curve corresponding to the linear model. Therefore, this step S6 describes how to adaptively modify the curve of the linear model to minimize the error between the data points of the designated user. In order to obtain a specified blood pressure model through learning, a regularization procedure can be performed, as shown in Equation 2:

（式二）

(Formula 2)

為指定血壓模型的損失函數，

為通用血壓模型的損失函數，

為調整參數。

之設定值愈大，則指定血壓模型與通用血壓模型的相似程度愈高。若

設定為0，則代表通用血壓模型對應的曲線將完全依據指定用戶的資料點進行擬合。

為正規化程序的修正函數，其計算方式如式三所示。為了保有通用血壓模型原本的特性，避免損失函數完全受指定用戶的資料點所支配，因此透過

及適當設置的

調整

。 in

To specify the loss function of the blood pressure model,

Is the loss function of the general blood pressure model,

To adjust the parameters.

The larger the setting value, the higher the similarity between the designated blood pressure model and the general blood pressure model. like

Setting it to 0 means that the curve corresponding to the general blood pressure model will be fitted completely based on the data points of the specified user.

It is the correction function of the normalization program, and its calculation method is shown in Equation 3. In order to maintain the original characteristics of the general blood pressure model, and to avoid the loss function being completely dominated by the data points of the specified user, through

And properly set up

adjust

.

（式三）

(Formula 3)

為通用血壓模型的參數集合（權重集合），

為指定血壓模型的參數集合（權重集合）。為了不讓

偏離於原本學習得到的

，本發明一實施例採用L1正規化以保留對估計血壓資料貢獻最大的權重。 in

Is the parameter set (weight set) of the general blood pressure model,

It is the parameter set (weight set) of the specified blood pressure model. In order not to let

Deviate from what was learned

In an embodiment of the present invention, L1 normalization is used to retain the weight that contributes the most to the estimated blood pressure data.

，可最佳化指定血壓模型的損失函數，進而建立適用於指定用戶的指定血壓模型。指定血壓模型具有第二參數集合，其係依據第一參數集合及步驟S5~S6所述的正規化程序調整得出。 According to the first error obtained in step S5, and select appropriate adjustment parameters

, Can optimize the loss function of the specified blood pressure model, and then establish the specified blood pressure model suitable for the specified user. The designated blood pressure model has a second parameter set, which is obtained by adjusting according to the first parameter set and the normalization procedure described in steps S5 to S6.

In another embodiment, after calculating the first error according to the first estimated blood pressure data, the second blood pressure data, and the loss function in step S5, the method may further include: according to the reference physiological data of the specified user (for example, the specified user measures the second The body temperature in the physiological data, the tightness of the cuff or the temperature when the designated user measures the second blood pressure data, the height and weight of the designated user, etc.) and adjust the parameter set according to the error to establish another designated blood pressure model. In other words, the present invention allows the establishment of multiple designated blood pressure models of designated users in different situations.

After completing the process of steps S1 to S6 in Figure 2, a designated blood pressure model suitable for designated users can be established. In order to further improve the measurement accuracy of the specified blood pressure model, the process shown in FIG. 3 can be continued. Fig. 3 is a subsequent flowchart of a method for establishing a blood pressure model according to an embodiment of the present invention.

Please refer to step S7 to obtain the third physiological data and third blood pressure data of the designated user. This step S7 is basically the same as step S3. For example, in step S3, the first designated user's data is obtained, and this step S7 is equivalent to the second designated user's data.

Please refer to step S8 to generate second estimated blood pressure data based on the third physiological data and the second parameter set. This step S8 is basically the same as step S4. The difference is that the second estimated blood pressure data is calculated using the second parameter set of the specified blood pressure model obtained in step S6.

計算第二誤差。本步驟S9基本上與步驟S5相同。 Please refer to step S9, based on the second estimated blood pressure data, the third blood pressure data and the loss function shown in Equation 1.

Calculate the second error. This step S9 is basically the same as step S5.

之計算結果。然而，本發明並不限制第三誤差的計算式。例如第三誤差可以是第一參數集合及第二參數集合之均方誤差（Mean Square Error，MSE）、平均絕對誤差（Mean Absolute Error，MAE）或交叉熵（Cross-Entropy）。 Please refer to step S10 to calculate the third error according to the first parameter set and the second parameter set. In one embodiment, the first parameter set of the general blood pressure model and the second parameter set of the specified blood pressure model are substituted into Equation 3 to obtain the third error, namely

The calculation result. However, the present invention does not limit the calculation formula of the third error. For example, the third error may be the mean square error (MSE), mean absolute error (MAE) or cross-entropy (Cross-Entropy) of the first parameter set and the second parameter set.

調整指定血壓模型之第二參數集合。本步驟S11基本上與步驟S6相同。 Please refer to step S11, according to the second error, third error and adjustment parameters

Adjust the second parameter set of the specified blood pressure model. This step S11 is basically the same as step S6.

Each time the process shown in FIG. 3 is executed, it is equivalent to using new physiological data and blood pressure data of the designated user to improve the measurement accuracy of the designated blood pressure model.

Please refer to Figure 4. Figure 4 is a bar graph showing the accuracy of multiple blood pressure models established based on multiple first physiological data. It can be seen from Fig. 4 that the selection of the PTT training general blood pressure model in step S1 can achieve the error of systolic blood pressure within 7 mmHg and the error of diastolic blood pressure within 5 mmHg. In the embodiment of using the ECG and PPG synchronization signals to train the general blood pressure model, the error of the systolic blood pressure can also be within 11 mmHg, and the error of the diastolic blood pressure can be within 9 mmHg. On the whole, using the PPG signal to train the general blood pressure model or using the PTT to train the general blood pressure model can meet the blood pressure measurement standard set by ANSI/AAMI/ISO in 2018, that is, the error is less than 10 mmHg.

In summary, the method for establishing a blood pressure model proposed by the present invention can achieve the following effects: the present invention improves the accuracy of cuffless blood pressure measurement, and can specify the blood pressure model according to personal physiological data. Institutionalized. The sphygmomanometer applying the present invention will not interfere with the user during long-term measurement. The present invention can establish a blood pressure model based on a variety of physiological signals, so the model has more flexibility in selection.

Although the present invention is disclosed in the foregoing embodiments, it is not intended to limit the present invention. All changes and modifications made without departing from the spirit and scope of the present invention fall within the scope of the patent protection of the present invention. For the scope of protection defined by the present invention, please refer to the attached scope of patent application.

S1~S6: steps

S7~S11: steps

Fig. 1 is a system design diagram for establishing a blood pressure model according to an embodiment of the present invention. Fig. 2 is a flowchart of a method for establishing a blood pressure model according to an embodiment of the present invention. Fig. 3 is a subsequent flowchart of a method for establishing a blood pressure model according to an embodiment of the present invention. Figure 4 is a bar graph showing the accuracy of multiple blood pressure models established based on multiple first physiological data.

S1~S6: steps

Claims (6)

A method for establishing a blood pressure model is suitable for a blood pressure measurement device. The method for establishing a blood pressure model includes: obtaining a plurality of first physiological data of a plurality of general users and a plurality of first blood pressure data of the general users; The first physiological data and the first blood pressure data execute a deep learning algorithm to establish a general blood pressure model, the general blood pressure model has a parameter set and a loss function; obtain a second physiological data of a designated user and the Specify second blood pressure data of one of the users; generate an estimated blood pressure data based on the second physiological data and the parameter set; calculate an error based on the estimated blood pressure data, the second blood pressure data, and the loss function; and calculate an error based on the error Adjusting the parameter set to establish a specified blood pressure model; wherein, before executing the deep learning algorithm according to the first physiological data and the first blood pressure data, it further includes: dividing the first physiological data into multiple times Fragments; and applying a linear filter to filter the time fragments according to the time fragments; wherein the time fragments include a plurality of first-type fragments and a plurality of second-type fragments, and signal variations in the first-type fragments The degree is greater than the signal variability in the second-type segments, and the linear filter retains the first-type segments. The method for establishing a blood pressure model according to claim 1, wherein the parameter set of the general blood pressure model is a first parameter set, the estimated blood pressure data is a first estimated blood pressure data, and the error is a first error , And the designated blood pressure model has a second parameter set; After adjusting the parameter set according to the error to establish the designated blood pressure model, it further includes: obtaining one of the third physiological data of the designated user and the third blood pressure data of the designated user; according to the third physiological data and the second The parameter set generates a second estimated blood pressure data; calculates a second error according to the second estimated blood pressure data, the third blood pressure data, and the loss function; calculates a second error according to the first parameter set and the second parameter set Three errors; and adjusting the second parameter set of the designated blood pressure model according to the second error, the third error, and an adjustment parameter. The method for establishing a blood pressure model according to claim 2, wherein the third error is the mean square error, average absolute error, or cross entropy of the first parameter set and the second parameter set. The method for establishing a blood pressure model according to claim 1, wherein each of the first physiological data and the second physiological data is an electrocardiogram signal, a photovolography signal or includes the electrocardiogram signal and the light volume One of the change profile signals is a synchronization signal, and the deep learning algorithm is a convolutional neural network with a multi-layer perceptron as a regression factor. The method for establishing a blood pressure model according to claim 1, wherein each of the first physiological data and the second physiological data is pulse transit time, and the general blood pressure model is linear regression. The method for establishing a blood pressure model according to claim 1, wherein after calculating the error according to the estimated blood pressure data, the second blood pressure data, and the loss function, it further includes: referring to physiological data according to one of the designated users and according to the The error adjusts this parameter set to build another specified blood pressure model.

Images