CN118844968A

CN118844968A - A blood pressure measurement system based on conformal prediction

Info

Publication number: CN118844968A
Application number: CN202410988759.9A
Authority: CN
Inventors: 丁晓蓉; 申展
Original assignee: University of Electronic Science and Technology of China
Current assignee: University of Electronic Science and Technology of China
Priority date: 2024-07-23
Filing date: 2024-07-23
Publication date: 2024-10-29

Abstract

The invention discloses a blood pressure measurement system based on conformal prediction, which relates to the field of non-invasive blood pressure measurement. The system specifically includes: a signal acquisition module, a feature extraction module, a blood pressure estimation module, a reference blood pressure measurement module and a conformal prediction module. The present invention introduces a conditional conformal prediction method to provide a corresponding confidence interval for the user according to the user's heart rate variability level, so as to facilitate doctors to accurately assess the blood pressure status; especially for patients with atrial fibrillation, the present invention can provide more reliable blood pressure measurement and assist in the monitoring and treatment of atrial fibrillation. The present invention uses conformal prediction to generate a confidence interval for blood pressure prediction, and provides a measurement interval that can cover the reference value on the basis of a single measurement value; the confidence interval for blood pressure prediction generated by conformal prediction can include the reference blood pressure value with a specific coverage rate; the confidence interval for blood pressure measurement is provided for patients with atrial fibrillation, so as to facilitate clinicians to accurately assess the patient's condition.

Description

Conformal prediction-based blood pressure measurement system

Technical Field

The invention relates to the field of noninvasive blood pressure measurement, in particular to a conformal prediction-based blood pressure measurement system.

Background

For arterial blood pressure measurement, invasive direct measurement methods (such as arterial puncture catheterization type blood pressure measurement) are invasive and require specialized medical personnel to perform operations, and are only suitable for clinical situations. Oscillometric blood pressure measurement based on inflatable cuffs is widely applied to clinical and home monitoring due to the noninvasive and easy-to-use characteristics of the oscillometric blood pressure measurement. Oscillometric blood pressure measurements record the pulse pressure profile and its envelope by detecting the oscillatory wave transmitted from the brachial or radial artery to the cuff, and calculate therefrom the systolic (Systolic Blood Pressure, SBP) and diastolic (Diastolic Blood Pressure, DBP) pressures. However, in atrial fibrillation, oscillometric blood pressure measurements tend to have large deviations due to the high variability of heart rate and stroke volume.

Atrial fibrillation (Atrial Fibrillation, AF), also known as atrial fibrillation, is a common arrhythmic disorder whose incidence increases with age. Patients with atrial fibrillation often have cardiovascular diseases such as hypertension and the like, and have higher thromboembolic risks; its abnormal arterial blood pressure will also exacerbate the onset of atrial fibrillation and even affect the rate of success of atrial fibrillation. Therefore, effective control of arterial blood pressure has an important role in monitoring and controlling atrial fibrillation. Abnormal ventricular rates (Ventricular Rate, VR) can lead to fluctuations in arterial pressure due to severe atrial electrical activity disturbances in patients with atrial fibrillation, producing irregularly fluctuating pulse pressure waveforms. When oscillometric blood pressure measurements are used, the difference between the multiple measurements tends to be large. Xie et al studied the effect of different ventricular rates of patients with atrial fibrillation on the accuracy of oscillometric blood pressure measurements with reference to invasive arterial pressure. The results show that the error in oscillometric measurement of blood pressure increases with increasing ventricular rate when the ventricular rate of an atrial fibrillation patient is greater than 80 beats per minute (beats per minute, bpm) as compared to a sinus rhythm patient. Thus, the accuracy and reliability of oscillometric blood pressure measurements in patients with atrial fibrillation still need to be further improved.

The existing researches are all based on single-point measurement of arterial blood pressure, the single-point measurement is difficult to accurately capture blood pressure values contained in abnormal arterial pressure waves, and the single-point measurement also often has large variability during atrial fibrillation. Unreliable blood pressure measurements will provide the clinician with erroneous diagnostic information and lower accuracy will also affect the clinician's assessment of the condition of the atrial fibrillation patient and the formulation of further treatment options. In 2005, vovk et al proposed a method of conformal prediction (Conformal Prediction, CP) to provide a reliable confidence reference for model predictions. Conformal prediction relies on a calibration set to estimate the confidence of the model for the current prediction, generating a prediction set with a limited error rate, ensuring that the prediction set covers the reference value with a specified probability. The conformal prediction establishes a statistically strict confidence interval, is applied to medical diagnosis, and can reduce the serious risk caused by model prediction failure.

Disclosure of Invention

The invention aims to provide a conformal prediction-based blood pressure measurement system which is applied to the field of noninvasive blood pressure measurement and can generate a confidence interval for blood pressure measurement so as to enhance the reliability of a prediction result. The invention introduces a condition conformal prediction method, and provides a corresponding confidence interval for the heart rate variability (HEART RATE Variability, HRV) according to the heart rate variability (HEART RATE Variability, HRV) level of the user, so that a doctor can accurately evaluate the blood pressure condition; the invention can provide more reliable blood pressure measurement and assist in monitoring and treating atrial fibrillation, especially for atrial fibrillation patients. Furthermore, the method is applicable to, but not limited to, blood pressure measurement, and can be applied to measurement of vital signs based on different types of methods.

The technical scheme of the invention is a conformal prediction-based blood pressure measurement system, which comprises: the device comprises a signal acquisition module, a blood pressure prediction module and a confidence interval generation module; the signal acquisition module is arranged on the surface of the skin and is used for acquiring heart rate signals and pressure shock wave curves; the blood pressure prediction module predicts the current blood pressure p by adopting a pressure shock wave curve; the calculation method in the confidence interval generation module is as follows:

step 1: calculating standard deviation sigma of heart rate according to the heart rate signal, and calculating average mu of heart rate;

step 2: calculating heart rate variability HRV;

step 3: dividing heart rate variability into a plurality of intervals in advance, and setting a corresponding confidence coefficient for each interval; determining the interval of the heart rate variability HRV obtained in the step 2, wherein the confidence coefficient corresponding to the interval is The confidence interval C (p) of the current blood pressure measurement is output as follows:

Further, the confidence degree determining method of each interval in the confidence interval generating module comprises the following steps:

S1: the signal acquisition module and the blood pressure prediction module are adopted to predict blood pressure to obtain a blood pressure predicted value Simultaneously, a non-invasive blood pressure measurement by an auscultation method or an invasive blood pressure measurement by an arterial catheterization method is adopted to measure a reference blood pressure value Y _i of a blood sample, i represents the number of a sample, a sample set is established (X _test,Y_test);X_test is a blood pressure predicted valueThe set of components, Y _test, is the set of reference blood pressure values Y _i;

s2: computing conformal score I=1, …, n, n represents the total number of samples;

s3: ascending order of conformal scores: s ₁<…<s_n, select The confidence corresponding to the interval is the confidence, and alpha is the set error rate.

Further, the population is divided into 3 groups according to heart rate variability levels, including but not limited to: g1= { HRV normal } = (0.03,0.15 ], g2= { HRV medium } = (0.15,0.27), g3= { HRV high } = (0.27,0.35 ].

The invention has the beneficial effects that:

1. The invention utilizes conformal prediction to generate a confidence interval of blood pressure prediction, and provides a measurement interval capable of covering a reference value on the basis of a single measurement value;

2. The confidence interval of the blood pressure prediction is generated by utilizing conformal prediction, and the reference blood pressure value can be contained with specific coverage rate;

3. The invention provides a confidence interval for blood pressure measurement for patients with atrial fibrillation, which is convenient for clinicians to accurately evaluate the conditions of the patients;

4. The invention has strong expansibility, and the blood pressure estimation module can be other sleeveless blood pressure estimation models capable of quantifying uncertainty; the commonality prediction module may generate reliable prediction intervals using the model uncertainty as a confidence interval.

Drawings

Fig. 1 is a system block diagram of the present invention.

Fig. 2 is a data flow diagram of the present invention.

FIG. 3 is a block diagram of a class conditional conformal prediction flow.

FIG. 4 is a graph of reference coverage and estimation error versus each group.

FIG. 5 is an exemplary graph of oscillometric estimates of blood pressure and prediction intervals.

Detailed Description

In order to achieve the purpose, the invention adopts oscillometric blood pressure measuring equipment to measure noninvasive blood pressure, and adopts an arterial catheterization method to measure invasive blood pressure as a reference blood pressure value; the calibration data set is then used to perform conditional conformal prediction, calculate confidence, and generate confidence intervals with a certain coverage for the blood pressure prediction.

The noninvasive blood pressure measuring device can be an automatic oscillometric arm type sphygmomanometer or a wrist type sphygmomanometer;

The invasive blood pressure measurement selects radial artery blood pressure;

the conformal prediction-based blood pressure confidence interval estimation method specifically comprises the following steps:

Step one, synchronously measuring the intra-aortic blood pressure and oscillometric blood pressure of a patient, taking an average value of 3 intra-aortic blood pressure records as an invasive blood pressure value, and taking an average value of 3 measured oscillometric blood pressures as a noninvasive blood pressure value;

step two, extracting peak values of the pulse pressure curve to calculate average HRV;

Thirdly, constructing a conformal prediction calibration set by the invasive blood pressure value, the noninvasive blood pressure value and the average HRV acquired in the first step and the second step; the calibration set data were divided into 3 groups according to HRV levels, respectively: g ₁ = { HRV normal }, G ₂ = { HRV medium }, G ₃ = { HRV high }; different tested person groups have different heart rate variability levels, and specific thresholds can be adjusted according to actual application conditions.

Step four, respectively implementing condition conformal prediction on each group, and calculating the corresponding confidence coefficient;

Step five, generating a prediction interval with a determined coverage rate according to HRV level self-adaptive matching confidence coefficient of newly acquired sample data;

further, the HRV calculation method in the second step is as follows:

where σ is the standard deviation of heart rate and μ is the mean of heart rate.

Further, the implementation of the conditional conformal prediction in the fourth step includes the following steps:

(1) For each group of calibration set data, a conformal score is calculated: n represents the number of samples of the calibration set;

(2) Setting an error rate alpha, and ensuring that the prediction set statistically contains correct labels with the probability of 1-alpha;

(3) Definition of the definition Is of conformal score s ₁,…,s_n Empirical quantile, whereIs an upward rounding function; the specific implementation process is that the conformal scores are sorted in ascending order: s ₁<…<s_n, selectIs a confidence level;

further, step five generates a prediction interval for the new data samples, comprising the steps of:

(1) Measuring noninvasive blood pressure, identifying peak values of pulse pressure waveforms, and calculating average HRV;

(2) According to the HRV level of the patient, self-adapting the confidence level;

(3) Conformal prediction can guarantee statistically:

(4) Generating a confidence interval:

the invention will be further described with reference to the drawings and specific examples, which are not intended to limit the invention, so that those skilled in the art will better understand the invention.

In order to verify the method provided by the invention, 674 healthy subjects and 726 atrial fibrillation subjects are collected through experiments, 1-3 times of oscillometric measurement data are acquired for each subject by using an intelligent wearable watch, and a mercury sphygmomanometer is synchronously used for measuring blood pressure as a reference value. The specific implementation steps are as follows:

step one, synchronously measuring oscillometric blood pressure and mercury cuff blood pressure of a patient, taking a measured oscillometric blood pressure average value as a measured value and taking a measured mercury blood pressure average value as a reference value;

Step two, dividing the crowd into two groups: g ₁ = { healthy subjects }, G ₂ = { atrial fibrillation subjects }, and selecting 80% of each group of data to construct a blood pressure calibration set;

setting error rate alpha=0.05, performing condition conformal prediction on each group, and calculating corresponding confidence coefficient;

Step four, taking subject data of 20% of each group as a test set, and generating a prediction interval according to the confidence coefficient obtained by calculation in the step three;

And fifthly, evaluating the coverage rate of the prediction interval to the reference value of the test set, and evaluating the estimation error of the oscillography in the test set.

Fig. 4 shows coverage of the prediction interval with reference values and oscillometric blood pressure estimation errors for atrial fibrillation patients and healthy subjects in the test set. For the population of two groups, the coverage rate of the prediction interval generated by conformal prediction on the reference value reaches more than 95%, and the coverage rate of the healthy subject group is higher and the estimation error is smaller.

Fig. 5 shows an example of prediction intervals of a part of samples in the test set, wherein gray solid lines represent reference blood pressure, black dotted lines represent estimated blood pressure of the wave method, and shaded areas of different colors represent prediction intervals generated at different error rate settings. The sample estimation points in the black dashed box can prove that when the blood pressure estimation is inaccurate (i.e. the difference between the estimated value and the reference value is large), the prediction interval provided by the model can still cover the reference blood pressure value, so that more reliable blood pressure estimation is provided, and further support is provided for clinical decision.

Claims

1. A blood pressure measurement system based on conformal prediction, the system comprising: a signal acquisition module, a blood pressure prediction module, and a confidence interval generation module; the signal acquisition module is arranged on the skin surface to collect heart rate signals and pressure oscillation wave curves; the blood pressure prediction module calculates the current blood pressure p using the pressure oscillation wave curve; the calculation method in the confidence interval generation module is:

Step 1: Calculate the standard deviation σ and mean μ of the heart rate based on the heart rate signal;

Step 2: Calculate heart rate variability (HRV);

Step 3: Divide the heart rate variability into multiple intervals in advance, and set a corresponding confidence level for each interval; determine the interval where the heart rate variability HRV obtained in step 2 is located, and the confidence level corresponding to this interval is Then the confidence interval C(p) of this blood pressure measurement is output as:

2. A blood pressure measurement system based on conformal prediction according to claim 1, characterized in that the confidence level of each interval in the confidence interval generation module is determined by:

S1: Using the signal acquisition module and the blood pressure prediction module to predict the blood pressure and obtain a blood pressure prediction value At the same time, the reference blood pressure value _Yi of the blood sample is measured by non-invasive blood pressure measurement based on auscultation or invasive blood pressure measurement based on arterial catheterization, where i represents the sample number, and a sample set (X _test , Y _test ) is established; X _test is the blood pressure prediction value Y _test is a set composed of reference blood pressure values _Yi ;

S2: Calculate the conformal score n represents the total number of samples;

S3: Sort the conformal scores in ascending order: s ₁ <…<s _n , select That is, the confidence level corresponding to the interval, and α is the set error rate.

3. A blood pressure measurement system based on conformal prediction as described in claim 1, characterized in that the population is divided into three groups according to the level of heart rate variability, including but not limited to: G_1 = {HRV normal} = (0.03, 0.15], G_2 = {HRV medium} = (0.15, 0.27], G_3 = {HRV high} = (0.27, 0.35]