Method for processing arterial blood pressure waveform signal through residual neural network

Abstract

The invention relates to the technical field of vital sign detection, in particular to a method for processing an arterial blood pressure waveform signal through a residual neural network. Comprising the following steps: preprocessing an arterial blood pressure waveform signal and extracting a characteristic value; performing short-time Fourier transform on the arterial blood pressure waveform signals to generate amplitude and phase images of the signals in a time-frequency domain, and inputting the images and characteristic values into a residual neural network model to obtain predicted values of hemodynamic parameters including cardiac output and stroke output; and predicting the blood pressure trend in a certain time in the future by using the time sequence prediction capability of the residual neural network. According to the method, more accurate heart and displacement parameter prediction and more accurate hypotension event prediction are provided; in addition, uncertainty quantization of hypotension event prediction is also increased.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of vital sign detection, in particular to a method for processing arterial blood pressure waveform signals through a residual neural network. BACKGROUND

[0002] Timely and accurate blood pressure monitoring is an important means of preventing hypertension and its complications. The current traditional blood pressure measurement method is usually an occasional daytime measurement method, which cannot be measured during sleep at night. In order to realize continuous dynamic monitoring of blood pressure, foreign scholars have proposed a "sleeveless" blood pressure monitoring technology. However, since this type of device needs to contact the wearer's body, long-term wearing at night will cause discomfort or skin irritation to the wearer, and the comfort level is low, thereby affecting the blood pressure measurement effect at night. In addition, neither the occasional daytime blood pressure measurement nor the wearable blood pressure monitoring can obtain more hemodynamic parameters, nor can they predict the probability of future possible hypotension events. Chinese patent application 2022800559278 discloses a "low blood pressure prediction device and method based on arterial blood pressure wavelet transform, and training method of low blood pressure prediction model thereof", which receives arterial blood pressure data of a subject, and uses a low blood pressure prediction model to determine whether the subject is hypotensive according to the change of each measurement interval of the arterial blood pressure data. The low blood pressure prediction model includes: a first layer trained to extract trend data of each interval, while performing wavelet transform on the training arterial blood pressure data to compress the training arterial blood pressure data; and a sub-sequence data generation module for generating each trend data into sub-sequence data according to behavior. The low blood pressure prediction model uses the training parameters learned in the first layer to perform low-pass filtering on the training arterial blood pressure data for training. The low blood pressure prediction model further includes: a second layer trained to assign weights to training intervals required for low blood pressure prediction among multiple intervals of the training arterial blood pressure data; and a third layer trained to calculate similarity feature values between sub-sequence data and trend data of the training interval according to the assigned weights. This patent application can use compressed arterial blood pressure (ABP) data containing overall blood pressure trend information to determine whether it is hypotensive. In addition, it can be determined whether it is hypotensive according to the change of each measurement interval of the compressed arterial blood pressure (ABP) data containing overall blood pressure trend information, so as to provide appropriate treatment by comparing these changes with general forms during surgery. However, the result of such deterministic prediction needs to be combined with uncertainty estimation to evaluate whether it is reliable. Uncertainty estimation is an important concept in the field of machine learning, which refers to the measure of the confidence of the model in its prediction results.‌ Residual neural network is a kind of deep learning model, in traditional network, with the increase of network depth, the model may be difficult to train due to gradient vanishing / explosion, and the performance may be saturated or even decreased. Residual neural network effectively solves the problem of gradient vanishing in deep neural network training by introducing residual block and jump connection, so that the network can be expanded to thousands of layers, and breakthrough results have been achieved in many computer vision tasks. SUMMARY

[0003] The purpose of the present application is to provide a method for processing arterial blood pressure waveform signals by residual neural network, which is different from the prior art, based on arterial blood pressure waveform, using convolution operator and deep residual neural network to directly predict cardiac output accurately, and then predict the uncertainty of hypotension event, and can provide the range of prediction uncertainty, providing more abundant basis for clinical use.

[0004] To achieve the above purpose, the technical scheme adopted by the present application is as follows: The method for processing arterial blood pressure waveform signals by residual neural network, comprising the following steps: 1) continuously measuring and acquiring arterial blood pressure waveform signals of the subject; 2) preprocessing the arterial blood pressure waveform signals to eliminate interference and noise, extracting characteristic values from the arterial blood pressure waveform signals and calculating pulse pressure and Liljestrand-Zander parameters based on the characteristic values; 3) performing short-time Fourier transform on the preprocessed arterial blood pressure waveform signals to generate amplitude and phase images of the signals in time-frequency domain, inputting the amplitude and phase images in time-frequency domain and the pulse pressure and Liljestrand-Zander parameters into a residual neural network model to obtain predicted values of hemodynamic parameters including cardiac output and stroke volume; 4) using the time series prediction capability of residual neural network to predict the blood pressure trend in a certain future time. First, calculate the mean arterial pressure based on the real-time arterial blood pressure; and use the mean and standard deviation calculated on the mean arterial pressure of the training sample to standardize the mean arterial pressure of the new sample; the standardized mean arterial pressure=(mean arterial pressure-mean) / standard deviation; then, perform inference of Bayesian residual neural network on the standardized mean arterial pressure. Calculate the average hypotension probability using the Monte Carlo samples of Bayesian inference, and calculate the uncertainty range of the average probability.

[0005] The technical solution is based on the arterial blood pressure waveform, uses a convolution operator and a deep residual neural network to directly and accurately predict cardiac output, and is more accurate than the current cardiac output prediction algorithm based on signal processing. In addition, the technical solution combines the Bayesian method and the deep convolutional neural network to predict the possibility of future occurrence of hypotension events. Compared with the existing commercial HPI and the advanced method in the literature, the prediction of the method is more accurate, and the range of prediction uncertainty can be provided, providing more abundant basis for clinical use.

[0006] Preferably, the residual neural network model is composed of a plurality of residual blocks, each residual block comprising a convolution layer, a batch normalization layer and an activation function layer; during the training of the residual neural network model, historical data labeled with hemodynamic parameters, including cardiac output and stroke volume, are used for supervised learning of the network.

[0007] The above technical solution effectively solves the gradient vanishing problem in deep neural network training, and can improve the prediction accuracy.

[0008] Preferably, the residual neural network model is optimized by hyperparameters, and the hyperparameters are adjusted and optimized by grid search and random search to improve the generalization ability and prediction accuracy of the residual neural network model, and the optimization parameters include learning rate, batch size and network layer number.

[0009] The above technical solution can improve the generalization ability and prediction accuracy of the residual neural network model, which is beneficial to improve the prediction accuracy.

[0010] Preferably, in step 2), the characteristic numerical value includes systolic pressure, diastolic pressure, mean arterial pressure, pulse rate, maximum pressure change rate, minimum pressure change rate, The formula for calculating the pulse pressure is: pulse pressure = systolic pressure - diastolic pressure. The formula for calculating the Liljestrand-Zander parameter is: pulse pressure / (systolic pressure + diastolic pressure).

[0011] The above technical solution can obtain the input data of the residual neural network model.

[0012] Preferably, in step 3), the residual neural network model outputs a cardiac index, and the cardiac output is calculated by multiplying the cardiac index by the body surface area of the subject, and the stroke volume is calculated by dividing the cardiac output by the heart rate.

[0013] The above technical solution can obtain the hemodynamic parameters output by the residual neural network model.

[0014] Compared with the prior art, the present application has the beneficial effects of providing more accurate prediction of cardiac output parameters, more accurate prediction of hypotension events, and further increasing the quantification of uncertainty of hypotension event prediction. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 Flowchart of an embodiment of the present application.

[0016] Figure 2 Flowchart of a Bayesian inference algorithm for hypotension events. DETAILED DESCRIPTION

[0017] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0018] As Figure 1 shown, the method for processing an arterial blood pressure waveform signal through a residual neural network comprises the following steps: 1) continuously measuring and acquiring an arterial blood pressure waveform signal of a subject, the arterial blood pressure being referred to as ABP, which is the lateral pressure of blood on the unit area of the aortic wall. The arterial blood pressure waveform signal is referred to as an ABP signal, and the sampling frequency of the present embodiment is 500 Hz; 2) pre-processing the arterial blood pressure waveform signal to eliminate interference and noise, extracting characteristic values from the arterial blood pressure waveform signal, and calculating a pulse pressure and a Liljestrand-Zander parameter based on the characteristic values; In this step, the characteristic values include systolic blood pressure, diastolic blood pressure, mean arterial pressure, pulse rate, maximum pressure change rate, and minimum pressure change rate.

[0019] ‌Systolic blood pressure, referred to as SBP, is the pressure in the artery when the heart contracts, reaching the highest value in the middle of the cardiac contraction, which is the pressure of blood on the inner wall of the blood vessel. It is also commonly referred to as high pressure, which is an important part of blood pressure measurement, reflecting the systolic function of the left ventricle and the elasticity of the aorta and large blood vessels. In normal circumstances, the normal range of systolic blood pressure of adults is 90-140 mmHg.

[0020] Diastolic blood pressure, referred to as DBP, is the pressure generated when the arterial blood vessel elastically retracts when the human heart diastolic. In normal circumstances, the normal range of diastolic blood pressure of adults is 60-90 mmHg.

[0021] Mean arterial pressure, abbreviated as MBP, refers to the average value of arterial blood pressure in a cardiac cycle. It can also reflect the average tissue perfusion. The value is calculated according to the formula: Mean arterial pressure = (Systolic pressure + 2× Diastolic pressure) / 3, or Mean arterial pressure = Diastolic pressure + 1 / 3 Pulse pressure difference. The normal value of mean arterial pressure in normal adults is 70-105mmHg.

[0022] Pulse rate, abbreviated as PR, refers to the frequency of arterial pulsation, that is, the number of pulses per minute. Pulse rate is usually consistent with heart rate. The normal pulse rate of normal adults in a quiet state is 60-100 times / minute.

[0023] Maximum pressure change rate refers to the maximum change rate of arterial blood pressure in a certain period of time. Maximum pressure change rate is related to blood pressure fluctuations, heart function, vascular elasticity and other factors.

[0024] Minimum pressure change rate refers to the minimum change rate of arterial blood pressure in a certain period of time. The calculation formula of the pulse pressure based on the above characteristic values is: Pulse pressure = Systolic pressure - Diastolic pressure. Liljestrand-Zander parameter refers to a formula or method for calculating cardiac output. This parameter or method can be used to evaluate the heart pumping function. In the calculation of cardiac output, the Liljestrand-Zander formula takes into account physiological parameters such as arterial blood pressure waveform, so it can more accurately estimate the cardiac output.

[0025] The calculation formula of the Liljestrand-Zander parameter in this embodiment is: Pulse pressure / (Systolic pressure + Diastolic pressure).

[0026] The above systolic pressure, diastolic pressure, mean arterial pressure, pulse rate, maximum pressure change rate, minimum pressure change rate, pulse pressure and Liljestrand-Zander parameter are jointly spliced into an 8-dimensional feature vector.

[0027] The pre-processed arterial blood pressure waveform signal is intercepted for a period of time for short-time Fourier transform to generate the amplitude and phase images of the signal in the time-frequency domain. In this embodiment, the ABP signal is intercepted for 50 seconds for short-time Fourier transform. The amplitude and phase images in the time-frequency domain and the above-mentioned 8-dimensional feature vector are input into the residual neural network model, so that the residual neural network model obtains the output of the cardiac index based on the input information. The amplitude and phase images in the time-frequency domain are input through the convolutional neural network based on ResNet to extract high-level features, and the characteristic values and the pulse pressure and Liljestrand-Zander parameters calculated based on the characteristic values are input through the LSTM network to capture the time sequence characteristics. The cardiac index refers to the value obtained by dividing the volume of blood pumped by the heart by the body surface area. The significance of this index is that it takes into account the body size difference of different individuals, so that the working efficiency of the heart can be more accurately evaluated. The cardiac index is determined by two factors, namely the heart rate and the amount of blood pumped by the heart per beat. The cardiac output can be obtained by multiplying the cardiac index output by the residual neural network model and the body surface area of the subject. Further, the stroke volume can be obtained by simply calculating the cardiac output and the pulse rate. The body surface area of the subject can be calculated using the Du Bois formula: BSA (m2) = 0.007184 x weight (kg) 0.425 x height (cm) 0.725 .

[0028] The residual neural network model is composed of a plurality of residual blocks, each of which includes a convolutional layer, a batch normalization layer and an activation function layer. The architecture of the specific residual neural network model is a known technology and will not be described in detail here. Figure 1 As shown in FIG. 8, during the training process of the residual neural network model, historical data labeled with hemodynamic parameters are used for supervised learning of the network. The hemodynamic parameters include cardiac output and stroke volume.

[0029] The residual neural network model is optimized by hyperparameters, which are adjusted and optimized by grid search and random search to improve the generalization ability and prediction accuracy of the residual neural network model. The optimization parameters include learning rate, batch size and network layer number.

[0030] 4) Finally, the time series prediction ability of the residual neural network is used to predict the blood pressure trend in the future for a certain period of time.

[0031] This step aims to predict the hypotension event within 5 minutes / 15 minutes in the future using mean arterial blood pressure, and to provide model prediction uncertainty estimation for the predicted probability, providing more abundant basis for clinical use.

[0032] First, a new sample is prepared, and mean arterial pressure is calculated based on real-time arterial blood pressure; and the mean arterial pressure of the new sample is standardized using the mean and standard deviation calculated on the mean arterial pressure of the training sample. The standardized mean arterial pressure = (mean arterial pressure - mean) / standard deviation.

[0033] Subsequently, inference of the Bayesian residual neural network is performed on the standardized mean arterial pressure. Using the Monte Carlo samples of Bayesian inference, the average hypotension probability is calculated, and the uncertainty range of the average probability is calculated. The Bayesian residual neural network refers to adding a Bayesian method for inference in the residual block of the residual neural network model.

[0034] The so-called Bayesian method is a probability inference framework based on Bayes' theorem, which is not described in detail in the prior art.

[0035] Finally, based on the final prediction value of the model, the corresponding hypotension classification result is generated according to the optimized classification rule. If a higher uncertainty is evaluated, it is marked as a low confidence result.

[0036] The contents not described in detail in the specification belong to the prior art known to those skilled in the art. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions recorded in the foregoing embodiments or make equivalent replacements for part of the technical features, and any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for processing arterial blood pressure waveform signals using a residual neural network, characterized in that... Includes the following steps: 1) Continuously measure and acquire the subject's arterial blood pressure waveform signal; 2) Preprocess the arterial blood pressure waveform signal to eliminate interference and noise, extract feature values ​​from the arterial blood pressure waveform signal, and calculate pulse pressure and Liljestrand-Zander parameters based on the feature values; 3) Perform a short-time Fourier transform on the preprocessed arterial blood pressure waveform signal to generate amplitude and phase images of the signal in the time-frequency domain. Input the amplitude and phase images in the time-frequency domain, as well as the pulse pressure and Liljestrand-Zander parameters, into the residual neural network model to obtain predicted values ​​of hemodynamic parameters, including cardiac output and stroke volume. 4) Utilize the time series prediction capabilities of residual neural networks to predict blood pressure trends over a certain period of time in the future.

2. The method for processing arterial blood pressure waveform signals using a residual neural network according to claim 1, characterized in that: The residual neural network model consists of multiple residual blocks, each containing a convolutional layer, a batch normalization layer, and an activation function layer. During the training of the residual neural network model, historical data labeled with hemodynamic parameters are used to supervise the learning of the network. The hemodynamic parameters include cardiac output and stroke volume.

3. The method for processing arterial blood pressure waveform signals using a residual neural network according to claim 2, characterized in that: The residual neural network model undergoes hyperparameter optimization, which involves adjusting the optimization parameters through grid search and random search to improve the generalization ability and prediction accuracy of the residual neural network model. The optimization parameters include learning rate, batch size, and number of network layers.

4. The method for processing arterial blood pressure waveform signals using a residual neural network according to claim 1 or 2, characterized in that: In step 2), the characteristic values ​​include systolic blood pressure, diastolic blood pressure, mean arterial pressure, pulse rate, maximum pressure change rate, and minimum pressure change rate. The formula for calculating pulse pressure is: Pulse pressure = Systolic pressure - Diastolic pressure; The formula for calculating the Liljestrand-Zander parameter is: pulse pressure / (systolic blood pressure + diastolic blood pressure).

5. The method for processing arterial blood pressure waveform signals using a residual neural network according to claim 1 or 2, characterized in that: In step 3), the residual neural network model outputs a cardiac index, which is multiplied by the subject's body surface area to calculate cardiac output, and cardiac output is divided by heart rate to calculate stroke volume.

6. The method for processing arterial blood pressure waveform signals using a residual neural network according to claim 1 or 2, characterized in that: In step 4), firstly, the mean arterial pressure is calculated based on real-time arterial blood pressure; then, the mean arterial pressure of the new sample is standardized using the mean and standard deviation calculated on the mean arterial pressure of the training sample; the standardized mean arterial pressure = (mean arterial pressure - mean) / standard deviation; Subsequently, a Bayesian residual neural network is used to infer the standardized mean arterial pressure. Using the Monte Carlo samples obtained from the Bayesian inference, the average probability of hypotension is calculated, along with the uncertainty range of this average probability.

Images