A blood pressure monitoring algorithm independent of dicrotic notch

Abstract

The present invention provides a blood pressure monitoring algorithm that is independent of the dicrotic notch, characterized in that it includes four parts: diastolic pressure modeling; pulse pressure modeling; morphology-independent blood pressure feature extraction and a blood pressure monitoring algorithm; compared with the traditional PTT method or oscillometric method, the design method of the present invention can perform continuous blood pressure monitoring during the day and at night with high comfort; compared with the latest deep learning method, the present invention can better adapt to the problem of missing dicrotic notch; the estimation errors of systolic pressure and diastolic pressure are 0.01±6.74 mmHg and 0.02±6.27 mmHg respectively, which meet the AAMI standard (the international standard for evaluating electronic sphygmomanometers promulgated by the American Association for the Advancement of Medical Instrumentation: ≤5±8 mmHg), while other methods do not meet the AAMI standard for the estimation errors of diastolic pressure and systolic pressure after discarding the dicrotic notch-related features. The continuous blood pressure monitoring of the present invention has high accuracy and is suitable for popularization and application.

Description

Technical Field

[0001] The present invention belongs to the technical field of non-invasive arterial blood pressure monitoring, and in particular to a blood pressure monitoring algorithm that is independent of the dicrotic notch. Background Art

[0002] Blood pressure is an important physiological indicator and is associated with a variety of health complications such as stroke, heart attack, heart failure and kidney damage. The World Health Organization reports that about one-third of adults worldwide suffer from hypertension. However, due to the hidden nature of hypertension, 80% of patients fail to receive appropriate treatment. Therefore, a large amount of research is devoted to blood pressure monitoring methods based on wearable devices to achieve continuous, comfortable and convenient blood pressure monitoring.

[0003] PPG sensors, as a commonly used sensor in wearable devices, can detect blood volume by illuminating the skin / tissue and measuring light absorption. Changes in blood volume are closely related to changes in blood pressure, so in recent years there have been many research works exploring the use of PPG sensors for blood pressure monitoring.

[0004] At present, most studies have constructed deep learning models to model the correlation between blood pressure and PPG data waveform and manual features of PPG data, and realized non-pressurized continuous blood pressure monitoring on wrist-worn devices; however, these models are heavily dependent on features related to the dicrotic notch, such as the reflected wave transmission time.

[0005] The dicrotic notch is crucial as it represents the moment when the aortic valve closes, marking the end of the systolic phase of a cardiac cycle. However, analysis of a large amount of PPG data shows that due to more than 20 personal factors (such as cholesterol content, vascular hardness, etc.) and data noise, a large proportion of individual PPG data lack the dicrotic notch.

[0006] There are also some studies in the existing technology that do not rely on the dicrotic notch, that is, measuring blood pressure by the pulse wave transmission time, but they require additional sensors and active cooperation from users, and cannot achieve continuous, convenient and non-contact blood pressure monitoring. Summary of the invention

[0007] In order to solve the above technical problems, the present invention provides a blood pressure monitoring algorithm that is independent of the dicrotic notch. Through scientific architecture design, the problem of missing dicrotic notch in PPG data obtained from the wrist, palm and fingertips is solved. The present invention can be applied to wearable devices with PPG to achieve continuous, comfortable and universal blood pressure monitoring.

[0008] A blood pressure monitoring algorithm independent of the dicrotic notch, including four parts: diastolic pressure modeling; pulse pressure modeling; morphology-independent blood pressure feature extraction and blood pressure monitoring algorithm;

[0009] Step 1: Diastolic blood pressure modeling;

[0010] The Frank-Starling law states that cardiac output increases with increasing diastolic blood pressure; this means that the PPG DC component, which reflects the end-diastolic arterial blood volume, will also increase with increasing diastolic blood pressure; in addition, by analyzing continuous blood pressure data and simultaneously collected PPG data, we also found that diastolic blood pressure (DBP) is correlated with systolic blood pressure (SBP).

[0011] Therefore, diastolic pressure is modeled according to the PPG DC component; based on this theory, a linear equation (Formula 1) can be established to express the relationship between diastolic pressure, PPG DC component and systolic pressure:

[0012]

[0013] in: , Personalized blood pressure coefficients for DC and SBP, is the personalized blood pressure deviation coefficient of DBP;

[0014] Step 2: Pulse pressure modeling;

[0015] The greater the pressure difference between diastolic pressure and systolic pressure, the greater the pulse pressure, the more blood volume in the artery increases during contraction, and the longer it takes to recover to the previous blood volume, that is, the longer the diastolic time; at the same time, blood flow velocity is related to blood pressure, that is, the higher the systolic pressure, the faster the increased blood volume in the artery recovers to the previous blood volume; based on this theory, a linear equation formula 2 is established to express the relationship between pulse pressure, diastolic time T and systolic pressure:

[0016]

[0017] in: , Personalized blood pressure coefficients for T and SBP, is the personalized blood pressure deviation coefficient of PP;

[0018] Step 3: Extraction of morphology-independent blood pressure features;

[0019] ① Extraction of PPG DC component;

[0020] First, the PPG signal is segmented into individual cardiac cycles based on the peak points detected by the peak detection (AMPD) algorithm;

[0021] As an example, the AMPD can adapt to different heart rates without the need for specific threshold settings for different individuals.

[0022] Secondly, in each cardiac cycle, the minimum point of the PPG signal is regarded as the DC component of the current PPG segment;

[0023] ②Diastolic time;

[0024] If the dicrotic notch is absent, it is difficult to determine the time of aortic valve closure based on the PPG morphology, which is necessary to calculate the diastolic time; according to the Frank-Sterling law, under normal circumstances, the systolic ejection volume is equal to the diastolic filling volume. If it is not equal, the heart will quickly adjust to restore the balance; this inspired us to estimate the time of aortic valve closure by analyzing the area under the AC component of the PPG, that is, the volume increased during contraction should be equal to the volume decreased during diastole;

[0025] Since the start and end positions of the PPG segments within a cardiac cycle fluctuate, these fluctuations cannot guarantee that the cardiac output in the current cardiac cycle is equal to the cardiac input; to accurately determine the diastolic time, the following steps need to be taken:

[0026] i) determine a PPG segment with a stable baseline within a cardiac cycle; to do this, it is necessary to ensure that the difference between the start and end positions of the PPG segment is less than a certain threshold of the PPG amplitude (e.g., 10%);

[0027] ii) If Figure 1 As shown, find a The vertical line represented by divides the PPG AC component into two parts of equal area;

[0028] iii) Use the end time of the current cardiac cycle Subtract vertical lines The position of , calculate the diastolic time ;

[0029] Step 4: Blood pressure monitoring algorithm;

[0030] Since both Formula 1 and Formula 2 involve the unknown variable SBP, according to The relationship between formula 1 and formula 2 is converted to obtain formula 3:

[0031]

[0032] Further, formula 3 is simplified to formula 4 as follows:

[0033]

[0034] Since vascular elasticity and diameter vary among individuals, it is necessary to calibrate the coefficients personalized for each individual;

[0035] By taking at least three blood pressure measurements, both diastolic and pulse pressure measurements were obtained, allowing six formulas to be established; these formulas can be used to fit six personalized blood pressure coefficients: , , , , and ;Once the six personalized blood pressure coefficients are determined, the diastolic blood pressure and pulse pressure can be estimated using Equation 4;

[0036] As an illustration, systolic pressure can also be estimated by adding diastolic pressure and pulse pressure.

[0037] Beneficial effects of the present invention:

[0038] ① Compared with the traditional PTT (pulse wave transmission time) method or oscillometric method, the design method of the present invention can perform continuous blood pressure monitoring during the day and night with high comfort;

[0039] ② Compared with the latest deep learning method, the present invention can better adapt to the problem of missing dicrotic notch; the experimental results of 85 people show that the estimation errors of systolic and diastolic blood pressure by the present application method are 0.01±6.74 mmHg and 0.02±6.27 mmHg, respectively, which meet the AAMI standard (the international standard for evaluating electronic blood pressure monitors issued by the American Association for the Advancement of Medical Instrumentation: ≤5±8 mmHg), while other methods do not meet the AAMI standard for the estimation errors of diastolic and systolic blood pressure after discarding the relevant features of the dicrotic notch.

[0040] ③In addition, compared to existing methods that were tested on no more than 35 subjects, we evaluated it on 85 subjects and achieved the best performance in continuous blood pressure monitoring accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 Schematic diagram of diastolic time extraction of a blood pressure monitoring algorithm independent of the dicrotic notch of the present invention.

[0042] Figure 2 The present invention is a flow chart of a blood pressure monitoring algorithm that is independent of the dicrotic notch.

[0043] Figure 3 This is an overall consistency evaluation diagram and a correlation diagram of the estimated systolic and diastolic blood pressures of a blood pressure monitoring algorithm that is independent of the dicrotic notch of the present invention. DETAILED DESCRIPTION

[0044] Below, reference Figures 1 to 3 As shown, a blood pressure monitoring algorithm independent of the dicrotic notch includes four parts: diastolic pressure modeling; pulse pressure modeling; morphology-independent blood pressure feature extraction and blood pressure monitoring algorithm;

[0045] Step 1: Diastolic blood pressure modeling;

[0046] The Frank-Starling law states that cardiac output increases with increasing diastolic blood pressure; this means that the PPG DC component, which reflects the end-diastolic arterial blood volume, will also increase with increasing diastolic blood pressure; in addition, by analyzing continuous blood pressure data and simultaneously collected PPG data, we also found that diastolic blood pressure is correlated with systolic blood pressure.

[0047] Therefore, diastolic pressure is modeled according to the PPG DC component; based on this theory, a linear equation (Formula 1) can be established to express the relationship between diastolic pressure, PPG DC component and systolic pressure:

[0048]

[0049] in: , Personalized blood pressure coefficients for DC and SBP, is the personalized blood pressure deviation coefficient of DBP;

[0050] Step 2: Pulse pressure modeling;

[0051] The greater the pressure difference between diastolic pressure and systolic pressure, the greater the pulse pressure, the more blood volume in the artery increases during contraction, and the longer it takes to recover to the previous blood volume, that is, the longer the diastolic time; at the same time, blood flow velocity is related to blood pressure, that is, the higher the systolic pressure, the faster the increased blood volume in the artery recovers to the previous blood volume; based on this theory, a linear equation formula 2 is established to express the relationship between pulse pressure, diastolic time T and systolic pressure:

[0052]

[0053] in: , Personalized blood pressure coefficients for T and SBP, is the personalized blood pressure deviation coefficient of PP;

[0054] Step 3: Extraction of morphology-independent blood pressure features;

[0055] ① Extraction of PPG DC component;

[0056] First, the PPG signal is segmented into individual cardiac cycles based on the peak points detected by the peak detection (AMPD) algorithm;

[0057] As an example, the AMPD can adapt to different heart rates without the need for specific threshold settings for different individuals.

[0058] Secondly, in each cardiac cycle, the minimum point of the PPG signal is regarded as the DC component of the current PPG segment;

[0059] ②Diastolic time;

[0060] If the dicrotic notch is absent, it is difficult to determine the time of aortic valve closure based on the PPG morphology, which is necessary to calculate the diastolic time; according to the Frank-Sterling law, under normal circumstances, the systolic ejection volume is equal to the diastolic filling volume. If it is not equal, the heart will quickly adjust to restore the balance; this inspired us to estimate the time of aortic valve closure by analyzing the area under the AC component of the PPG, that is, the volume increased during contraction should be equal to the volume decreased during diastole;

[0061] Since the start and end positions of the PPG segments within a cardiac cycle fluctuate, these fluctuations cannot guarantee that the cardiac output in the current cardiac cycle is equal to the cardiac input; to accurately determine the diastolic time, the following steps need to be taken:

[0062] i) determine a PPG segment with a stable baseline within a cardiac cycle; to do this, it is necessary to ensure that the difference between the start and end positions of the PPG segment is less than a certain threshold of the PPG amplitude (e.g., 10%);

[0063] ii) If Figure 1 As shown, find a The vertical line represented by divides the PPG AC component into two parts of equal area;

[0064] iii) Use the end time of the current cardiac cycle Subtract vertical lines The position of , calculate the diastolic time ;

[0065] Step 4: Blood pressure monitoring algorithm;

[0066] Since both Formula 1 and Formula 2 involve the unknown variable SBP, according to The relationship between formula 1 and formula 2 is converted to obtain formula 3:

[0067]

[0068] Further, formula 3 is simplified to formula 4 as follows:

[0069]

[0070] Since vascular elasticity and diameter vary among individuals, it is necessary to calibrate the coefficients personalized for each individual;

[0071] By taking at least three blood pressure measurements, both diastolic and pulse pressure measurements were obtained, allowing six formulas to be established; these formulas can be used to fit six personalized blood pressure coefficients: , , , , and ;Once the six personalized blood pressure coefficients are determined, the diastolic blood pressure and pulse pressure can be estimated using Equation 4;

[0072] As an illustration, systolic pressure can also be estimated by adding diastolic pressure and pulse pressure.

[0073] In order to better illustrate the design principle of the present invention, now combined with the attached Figure 2 Briefly introduce the operating steps of the present invention when it is applied:

[0074] Embodiment 1:

[0075] Figure 2 Shown is a flow chart of the blood pressure monitoring algorithm proposed in this application;

[0076] First, after wearing the ring, the user will be asked to perform a personalized blood pressure coefficient initialization operation, which involves three PPG data collections and blood pressure measurements;

[0077] Specifically, users are asked to keep the finger wearing the smart ring aligned with their heart for 30 seconds and measure their blood pressure during this period; this step will be repeated three times to solve the model of four proportional coefficients and two deviation coefficients required for the blood pressure monitoring algorithm.

[0078] Secondly, after the initialization operation is completed, the blood pressure monitoring method proposed in this application can realize continuous monitoring of blood pressure; specifically, the PPG sensor of the smart ring will continuously collect PPG data at the fingertips and input it into a morphology-independent blood pressure feature extraction module to extract blood pressure-related features, namely, the PPG DC component and diastolic time.

[0079] Finally, we input the above features into the blood pressure monitoring algorithm to estimate the user's diastolic and systolic blood pressure at the current moment.

[0080] Embodiment 2:

[0081] Compared with the latest deep learning methods, the present invention can better adapt to the problem of missing dicrotic notch; the experimental results of 85 people show that the estimation errors of systolic and diastolic blood pressure by the present application method are 0.01±6.74 mmHg and 0.02±6.27 mmHg respectively, which meet the AAMI standard (the international standard for evaluating electronic blood pressure monitors issued by the American Association for the Advancement of Medical Instrumentation: ≤5±8 mmHg), while other methods do not meet the AAMI standard for the estimation errors of diastolic and systolic blood pressure after discarding the relevant features of dicrotic notch; in addition, compared with the existing methods that were tested on no more than 35 subjects, we evaluated on 85 subjects, and our method had the best performance in continuous blood pressure monitoring accuracy.

[0082] Below, through the specific statistics in Table 1, a detailed comparison is shown as follows:

[0083] Table 1: Comparison of blood pressure measurement performance of different products

[0084] Compare Products Seismo Watches Blood pressure measuring phone case EBP blood pressure measurement earphones Crisp-BP Blood Pressure Wristband This application Number of testers 13 30 35 35 85 Usage scenarios daytime Single daytime Day + Night Day + Night Comfort middle middle Low high high Versatility high high high Low high Systolic blood pressure ME / STD 4.8 / - 3.3 / 8.8 1.8 / 7.2 1.67 / 7.31 0.01 / 6.74 Diastolic blood pressure ME / STD 2.9 / - 5.6 / 7.7 3.1 / 7.9 0.86 / 6.55 0.02 / 6.27

[0085] Where: ME is the mean error, see formula 5, STD is the standard deviation, see formula 6; For the algorithm to estimate blood pressure, To truly measure blood pressure, is the sample size.

[0086]

[0087]

[0088] It can be found from Table 1 that compared with other conventional methods in the prior art, the present application was tested on multiple groups of subjects, and the beneficial effects of the present application were the best in five aspects: usage scenario (continuity), comfort and versatility, systolic blood pressure estimation error, and diastolic blood pressure estimation error.

[0089] Embodiment 3:

[0090] Refer to the instruction manual Figure 3 As shown, in the consistency evaluation graph, more than 95% of the data points were within the consistency range of systolic and diastolic blood pressure (𝑀±1.96𝑆𝑇𝐷);

[0091] In addition, Figure 3 As shown, ME (SBP: 0.01, DBP: 0.02) and 𝑆𝑇𝐷 (SBP: 6.74, DBP: 6.27) meet the error bounds defined by AAMI, i.e., 𝑀≤5 and 𝑆𝑇𝐷≤8.

[0092] This shows that the application has high accuracy in practical use.

[0093] Furthermore, Figure 3 The correlation plot shown again demonstrates the significant correlation between the estimated ABP and the reference ABP, calculated using Equation 7. The value is 0.86, diastolic blood pressure The value is 0.81;

[0094]

[0095] In order to better illustrate the architecture of the present invention, the ideas of the present invention are briefly described as follows through the introduction of relevant principles and knowledge:

[0096] The present invention creates a morphology-independent blood pressure monitoring algorithm based on PPG data, according to the Frank-Starling law, which states that changes in cardiac output can be used to estimate blood pressure; that is, during cardiac contraction, blood volume in the arteries increases, causing arterial wall pressure to rise, i.e., systolic pressure; as the aortic valve closes, blood volume gradually decreases during cardiac diastole, causing arterial wall pressure to return to diastolic pressure.

[0097] According to the Frank-Starling law:

[0098] (i) The increase / decrease in the final residual blood volume in the artery is related to diastolic blood pressure; this change can be captured by the morphology-independent PPG direct current (DC) component of the PPG data;

[0099] (ii) The rate of blood volume loss during diastole is related to pulse pressure (PP = SBP-DBP), which can be captured by the alternating current (AC) component of the PPG; therefore, blood pressure can be estimated based on features related to blood volume changes without relying on the presence of the dicrotic notch.

[0100] The present invention utilizes the law of cardiac dynamics (Frank-Starling Law) to address the problem of missing "dicrotic notch", an important morphological feature of blood pressure, in photoplethysmography (PPG) data, so as to achieve continuous, comfortable and universal blood pressure monitoring; compared with the prior art, the present invention focuses on utilizing hemodynamics to explore a blood pressure monitoring algorithm that is independent of the dicrotic notch, so as to achieve more universal continuous and comfortable blood pressure monitoring.

[0101] The above are only preferred embodiments of the present invention. It should be understood that the description of the above embodiments is only used to help understand the method and core ideas of the present invention, and is not used to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, etc. made within the ideas and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A blood pressure monitoring algorithm independent of the dicrotic notch, characterized in that: include: Diastolic pressure modeling;Pulse pressure modeling; The four parts are morphology-independent blood pressure feature extraction and blood pressure monitoring algorithm; Step 1: Diastolic blood pressure modeling; The diastolic pressure is modeled based on the PPG DC component; a linear equation, Formula 1, is established to represent the relationship between the diastolic pressure and the PPG DC component and the systolic pressure: DBP=k1DC+k2SBP+b1, (1) Among them: k1 and k2 are the personalized blood pressure coefficients of DC and SBP, and b1 is the personalized blood pressure deviation coefficient of DBP; Step 2: Pulse pressure modeling; A linear equation formula 2 is established to express the relationship between pulse pressure, diastolic time T and systolic pressure: PP=k3T+k4SBP+b2, (2) Among them: k3 and k4 are the personalized blood pressure coefficients of T and SBP, and b2 is the personalized blood pressure deviation coefficient of PP; Step 3: Extraction of morphology-independent blood pressure features; ① Extraction of PPG DC component; First, the PPG signal is segmented into individual cardiac cycles based on the peak points detected by the peak detection algorithm; Secondly, in each cardiac cycle, the minimum point of the PPG signal is regarded as the DC component of the current PPG segment; ② Accurately determine the diastolic time; Step 4: Blood pressure monitoring algorithm; Since both Formula 1 and Formula 2 involve the unknown variable SBP, according to the relationship SBP=DBP+PP, Formula 1 and Formula 2 are converted to obtain Formula 3: Formula 3 is simplified to Formula 4 as follows: Calibrate the personalized coefficients for each individual; by performing at least three blood pressure measurements, the measured values ​​of diastolic pressure and pulse pressure can be obtained simultaneously, thereby establishing six formulas for fitting six personalized blood pressure coefficients: k′1, k′2, k′3, k′4, b′1 and b′2; after determining the six personalized blood pressure coefficients, use formula 4 to estimate the diastolic pressure and pulse pressure.

2. A dicrotic notch-independent blood pressure monitoring algorithm according to claim 1, characterized in that: To accurately determine the diastolic time, the following steps are required: i) Determine a PPG segment with a stable baseline within a cardiac cycle; to do this, it is necessary to ensure that the difference between the start and end positions of the PPG segment is less than a certain threshold of the PPG amplitude; ii) Find a path with x=t v The vertical line represented by divides the PPG AC component into two parts of equal area; iii) Use the end time t of the current cardiac cycle c Subtract the vertical line x = t v The position of c -t v , calculate the diastolic time T.

3. A dicrotic notch-independent blood pressure monitoring algorithm according to claim 2, characterized in that: A certain threshold value of the PPG amplitude is 10%.

4. The dicrotic notch-independent blood pressure monitoring algorithm according to claim 1, characterized in that: The peak detection can adapt to different heart rates without the need for specific threshold settings for different individuals.

5. The dicrotic notch-independent blood pressure monitoring algorithm according to claim 1, characterized in that: The systolic blood pressure can also be estimated by adding the diastolic blood pressure and pulse pressure.

Images