Non-invasive blood pressure estimation method based on multiple measurement pulse wave fusion

Abstract

According to the non-invasive blood pressure estimation method based on multiple measurement pulse wave fusion, multiple cuff measurements are conducted on the same object, quality evaluation is conducted on the pulse wave oscillation sequence obtained through multiple measurements, all irregular oscillation waveforms are deleted, it is ensured that the reserved waveforms are regular waveforms, and the accuracy of blood pressure estimation is improved. All amplitude data are mapped to the same coordinate system through pressure values in the cuff pressure curve; due to the fact that all irregular waveforms are deleted, stable and high pulse wave amplitude data can be obtained after all reserved pulse wave segments are spliced, a unified oscillation envelope is formed by means of the overlapping performance of multiple pulse waves on a pressure axis, and then the blood pressure is obtained through the unified envelope.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, specifically to a non-invasive blood pressure estimation method based on the fusion of multiple pulse wave measurements. Background Technology

[0002] Non-invasive blood pressure measurement technology is widely used in home health monitoring and medical devices, with the oscillometric method being the most common. The oscillometric method works by inflating a cuff to block arterial blood flow. During deflation, it detects pressure fluctuations caused by the vibration of the blood vessel wall. By collecting the periodic oscillation amplitude changes of the pressure from the cuff, an oscillation envelope is constructed. The mean arterial pressure (MAP), systolic blood pressure (SBP), and diastolic blood pressure (DBP) are then calculated from the peak value and proportional relationship of the envelope.

[0003] However, in arrhythmias such as atrial fibrillation (AF), the traditional oscillation method has significant limitations: 1. Highly unstable pulse amplitude: Atrial fibrillation causes dramatic fluctuations in the diastolic length, ventricular filling volume, and ejection volume between pulses, resulting in extremely irregular pulse wave oscillation amplitude, which no longer conforms to the smooth single-peak envelope assumed by the oscillation method.

[0004] 2. Single measurement envelope is often distorted or difficult to fit: In the same cuff measurement, there may be alternation between strong and weak pulsations, missing waveforms, double peaks, etc., which makes it impossible to stably fit the envelope curve.

[0005] 3. Traditional algorithms ignore the complementarity of multiple measurements: Most existing technologies rely on a single inflation / deflation process to estimate blood pressure. However, in atrial fibrillation, each measurement may contain different noise, abnormal pulsations, and amplitude escape. A single measurement is often unreliable, but multiple measurements can provide complementary information.

[0006] To address the above issues, the current common technical solution is to perform multiple blood pressure measurements and then use either the average or the best result as the final measurement. However, the accuracy of either the average or the best result is inconsistent. Summary of the Invention

[0007] To address the issue that existing blood pressure measurement methods based on cuff oscillation cannot guarantee the accuracy of blood pressure measurements in cases of atrial fibrillation or other arrhythmias, this invention provides a non-invasive blood pressure estimation method based on the fusion of multiple pulse wave measurements. This method can construct a stable and reliable oscillation envelope, thereby improving the accuracy of blood pressure estimation under atrial fibrillation and other irregular rhythms.

[0008] The technical solution of this invention is as follows: a non-invasive blood pressure estimation method based on the fusion of multiple pulse wave measurements, characterized by comprising the following steps: S1: Perform M consecutive cuff measurements on the same object, where M≥2; collect data to be processed after each measurement; The data to be processed includes: cuff pressure curve and pulse wave oscillation sequence synchronized with pressure changes; The horizontal axis of the cuff pressure curve represents time, and the vertical axis represents the cuff pressure value. The horizontal axis of the pulse wave oscillation sequence is time, and the vertical axis is the amplitude of the pulse wave. S2: The pulse wave oscillation sequence in the data to be processed is denoted as: the pulse wave to be processed; Each pulse wave to be processed is divided and segmented based on the heartbeat to obtain the pulse wave segment to be processed; S3: Evaluate the quality of each pulse wave segment to be processed, find all segments with irregular pulse wave oscillation amplitudes, and delete them; the remaining pulse wave segments are denoted as: effective pulsation segments; S4: To address the irregularity of the heartbeat, the amplitude of each effective pulsation segment is compensated to obtain the compensated amplitude; S5: For the same data to be processed, the cuff pressure curve and the effective pulsation segment after amplitude compensation are aligned according to the horizontal axis and fused to obtain the pressure amplitude curve. The horizontal axis of the pressure amplitude curve represents pressure, and the vertical axis represents the amplitude of the pulse wave. S6: Repeat steps S2 to S5; merge the pressure amplitude curves corresponding to all the data to be processed into the same coordinate system to obtain a unified pressure amplitude curve, denoted as: merged pressure amplitude curve; S7: Determine the quality of the amplitude data corresponding to the fused pressure amplitude curve; If the quality results are sufficient to support subsequent calculations, proceed to step S8; Otherwise, continue to increase the number of measurements, and execute steps S2 to S6 for each measurement cycle; S8: Convert the amplitude data in the fused pressure amplitude curve to the coordinate system of the pulse wave oscillation sequence to obtain the fused pulse wave oscillation sequence; Based on the fused pulse wave oscillation sequence, the oscillation envelope is extracted and denoted as: fused oscillation envelope; S9: The blood pressure value is obtained by subsequent calculation based on the fused oscillation envelope.

[0009] Its further features are: In step S4, the method for calculating the compensated amplitude is as follows: , In the formula, To compensate for the amplitude, A i C represents the original pulse wave amplitude corresponding to the i-th effective pulsation segment. i As a compensation factor; ; In the formula, clip() is a clipping function that clips data exceeding the lower boundary value C. min and the upper boundary value C max Replace the elements with the corresponding boundary values; ε is the average amplitude in the pressure neighborhood; ε is a minimal constant to prevent division by zero. λ is the jump suppression coefficient, which controls the strength of suppression against sudden changes in pulsation amplitude; ΔA i This represents the change in amplitude between adjacent beats; △A i = A i -A i-1 ; The method for calculating the average amplitude of the pressure neighborhood is as follows: ; In the formula, ΔP is the pressure neighborhood width, defined as the threshold range of similar pressures that divide a pressure interval; N is the total number of pressure intervals included in the i-th effective pulsation segment; A j The original pulsation amplitude corresponding to the j-th pressure zone; P j : The representative pressure value corresponding to the current pulsation in the j-th pressure zone; Step S5 specifically includes the following operations: a1: Construct the pressure axis; The pressure values ​​are evenly divided into segments according to a preset range to obtain pressure intervals; a2: Using the time axis as a reference, align the amplitude values ​​of the effective pulsation segments after amplitude compensation to the pressure axis; a3: For each pressure range, collect all compensated amplitude points that fall within this pressure range from all measurements, and calculate the weighted fused amplitude A. fusion The fused amplitude point sequence is obtained; ; In the formula, P j : The representative pressure value corresponding to the current pulsation in the j-th pressure zone; k is the measurement number, representing the kth independent blood pressure measurement process; i is the pulsation number, representing the i-th valid pulsation detected in the k-th measurement; Let be the amplitude value of the i-th pulse after irregularity compensation in the k-th measurement; w i k To integrate weights, w i k = q i k ×m k ×s i k ; q i k For waveform quality weights, m k To measure the mass weight, s i k Weights for stable segments; The waveform quality weight q i k The calculation method is as follows: b1: Segment the i-th valid pulse into a waveform, and mark the segmented waveform with waveform time data; The waveform time data includes: t 10 t 90 and t p ; Among them, t 10 This indicates the moment when the amplitude rises to 10% of its peak. t 90 The moment when the amplitude rises to 90% of its peak. t p Peak time; b2: The rise time is calculated as follows: T rise =t p -t 10 The rising slope segment T 10→90 =t 90 -t 10 ; b3: Calculate the morphological rationality score q shape,i (k) : ; In the formula, q shape,i (k) The waveform rationality score corresponding to the i-th valid pulse wave in the k-th measurement; μ T This is the mean value obtained from the statistical analysis of effective pulsations of the test subjects; σ T The variance is obtained from the effective pulsation statistics of the test subjects; b4: Define the waveform sequence of the i-th effective pulse as x[n], where L is the length of the i-th effective pulse wave on the horizontal axis; Define second-order difference: Δ 2 x[n] = x[n] - 2x[n-1] + x[n-2]; b5: Calculate the noise ratio r i (k) : ; In the formula, r i (k) The noise ratio corresponding to the i-th effective pulse wave in the k-th measurement; n is the n-th value of the horizontal axis; The average value of x[0] to x[L]; b6: Calculate the noise fraction q corresponding to the i-th valid pulse wave in the k-th measurement. noise,i (k) : q noise,i (k) =exp(-γr i (k) ); In the formula, γ is the noise penalty intensity; b7: Calculate the quality weights of the fused waveform; q i k =clip(q shape,i (k) ×q noise,i (k) (0, 1); The measured quality weight m k The calculation method is as follows: c1: Calculate the effective pulsation ratio p (k) ; p (k) =N (k) valid / (N (k) + ε); In the formula, N (k) N represents the total number of pulses detected in the k-th measurement. (k) valid The effective number of pulses obtained after passing quality screening; c2: Map each stroke count in this measurement to a pressure interval to obtain the coarse envelope point (P). j A j (k) ); A j (k)This represents the pulsation amplitude corresponding to the j-th pressure interval in the pulsation data obtained from the k-th measurement; P j This represents the representative pressure value corresponding to the current pulsation in the j-th pressure interval; For A j (k) Perform first-order difference: d j = A j (k) - A j-1 (k) ; c3: The number of sign changes of the statistical derivative Z (k) This describes the extreme points included in the amplitude curve of the k-th measurement. ; In the formula, Z (k) Let J be the number of times the derivative sign changes corresponding to the k-th measurement, and J be the total number of pressure intervals corresponding to the k-th measurement data. c4: Calculate the unimodal fraction u (k) ; u (k) = exp(-ηZ (k) ); In the formula, η represents the intensity of the sign change penalty; c5: Calculate the measurement quality weight m k ; m k = clip(p (k) ×u (k) ,0,1); The stable segment weights s i k The calculation method is as follows: d1: Evaluate the local mean and dispersion of the i-th effective pulsation corresponding to the k-th measurement within the neighborhood of the pressure interval; Define the neighborhood set of the i-th valid pulse corresponding to the k-th measurement. : ; Calculate the local mean μ i (k) ; ; Calculate the local standard deviation σ i (k) : ; Calculate the dimensionless coefficient of variation cv i (k) ; ; d2: Calculate the weight s of the stable segment. i k ; s i k =exp(-βcv) i (k) ); In the formula, β is the stability penalty intensity; In step S3, the quality of each pulse wave segment to be processed is evaluated, and all segments with irregular pulse wave oscillation amplitudes are identified and deleted; specifically, the following operations are performed: e1: Evaluate the quality of the rising slope; Calculate the rising slope of the pulse wave segment to be processed and compare it with a preset slope range; When the value is outside the range, the rising slope quality RiseQ is recorded as 0; otherwise, it is recorded as 1. e2: Evaluate the symmetry of the waveform of the pulse wave segment to be processed; If the waveform is symmetrical about the R-peak, then the symmetry quality symmetryQ is marked as 1; otherwise, it is marked as 0. e3: Evaluate the peak quality of the pulse wave segment to be processed; Detect whether the waveform segment to be processed has double peaks or missing peaks; If double peaks or missing peaks exist, the waveform quality peakV is marked as 0; otherwise, it is marked as 1. e4: Evaluate the noise and distortion indices of the pulse wave segment to be processed; Evaluation of noise performance based on the signal-to-noise ratio (SNR) method; If the noise is significant, mark the noise and distortion index ND as 0 and stop the calculation. Otherwise, determine whether there is distortion in the shape of the pulse wave segment to be processed. If there is, mark the noise and distortion index as 0; otherwise, mark the noise and distortion index as 1. e5: For each pulse wave segment to be processed, evaluate the quality by performing the above calculations. If any index is marked as 0, it is determined that the pulse wave oscillation amplitude is irregular and is deleted.

[0010] This application provides a non-invasive blood pressure estimation method based on multiple pulse wave measurements. It involves performing multiple cuff measurements on the same subject, evaluating the quality of the pulse wave oscillation sequences obtained from these measurements, removing all irregular oscillation waveforms to ensure only regular waveforms remain, and then mapping all amplitude data to the same coordinate system using the pressure values ​​from the cuff pressure curve. Because all irregular waveforms have been removed, the remaining pulse wave segments are spliced ​​together to obtain stable pulse wave amplitude data. The overlap of multiple pulse waves on the pressure axis forms a unified oscillation envelope, from which blood pressure is calculated. This method allows for "multiple inflation and deflation processes" and fuses the effective pulses from each measurement across measurements, thus overcoming the problem of single-measurement envelope distortion caused by atrial fibrillation. This method does not rely on the transient waveform of a single measurement but instead utilizes the statistical stability of multiple measurements to construct a highly robust oscillation envelope, improving accuracy and repeatability under atrial fibrillation. This method relies solely on pulse waves and pressure curves to achieve atrial fibrillation-friendly measurements in ECG-free devices and is suitable for various portable devices. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of the non-invasive blood pressure estimation method based on the fusion of multiple pulse wave measurements in this paper. Figure 2 This is an example of an estimation method. Detailed Implementation

[0012] In atrial fibrillation scenarios, relying solely on the single-oscillation method is insufficient to obtain stable and accurate blood pressure. This invention proposes a non-invasive blood pressure estimation method based on the fusion of multiple pulse wave measurements. This method can construct a stable and reliable oscillation envelope across multiple fused measurements, thereby enabling blood pressure estimation in atrial fibrillation and other irregular rhythms. Figure 1 As shown, this method specifically includes the following steps.

[0013] S1: Perform M consecutive cuff measurements on the same object, where M≥2; collect data to be processed after each measurement; The data to be processed includes: cuff pressure curves and pulse wave oscillation sequences with synchronized pressure changes; because the pressure values ​​of the two change synchronously, the pressure values ​​of the cuff pressure curves and the amplitude values ​​of the pulse wave oscillation sequences collected in the same time can be aligned and merged based on the time coordinate to obtain the pressure amplitude curve, and then the data collected in different times can be mapped and fused based on the pressure value.

[0014] The horizontal axis of the cuff pressure curve represents time, and the vertical axis represents the cuff pressure value. The horizontal axis of the pulse wave oscillation sequence represents time, and the vertical axis represents the amplitude of the pulse wave.

[0015] Each pulse wave measurement is used as an input sample for subsequent calculations.

[0016] S2: The pulse wave oscillation sequence in the data to be processed is denoted as: the pulse wave to be processed; Each pulse wave to be processed is divided and segmented based on the heartbeat to obtain the pulse wave segment to be processed.

[0017] like Figure 2 In the diagram, the curves shown for k=1~3 represent the pulse waves to be processed. Each peak represents a heartbeat. A normal heartbeat duration is between 0.5 and 1 second. For specific segmentation, a specific value can be selected within the normal heartbeat range. Alternatively, in this embodiment, high-quality heartbeat data from the test subject is analyzed to obtain the duration of the subject's normal heartbeat. Using the parallax of the subject's normal heartbeat, the pulse wave to be processed is segmented, resulting in pulse wave fragments. This ensures the segmentation results are more targeted, thereby ensuring the accuracy of subsequent results.

[0018] S3: Evaluate the quality of each pulse wave segment to be processed, find all segments with irregular pulse wave oscillation amplitudes, and delete them; the remaining pulse wave segments are denoted as: effective pulsation segments.

[0019] The method for finding and deleting segments with irregular pulse wave amplitudes includes the following operations.

[0020] e1: Evaluate the quality of the rising slope; Calculate the rising slope of the pulse wave segment to be processed and compare it with a preset slope range; When the value is outside the range, the rising edge slope quality RiseQ is recorded as 0; otherwise, it is recorded as 1.

[0021] The rising edge slope intuitively reflects how quickly a signal transitions from a low level to a high level. The specific calculation method can be implemented using existing technology, such as determining the slope by measuring the ratio of the vertical pressure change to the corresponding horizontal time change during the rising edge.

[0022] e2: Evaluate the symmetry of the waveform of the pulse wave segment to be processed; If the waveform is symmetrical about the R-peak, then the symmetry quality symmetryQ is marked as 1; otherwise, it is marked as 0.

[0023] Symmetry of a pulse wave segment refers to the geometric symmetry of the waveform relative to the peak on the time axis. Normal pulse waveforms typically exhibit a certain degree of symmetry and consistency, while waveforms caused by atrial fibrillation or other reasons may show asymmetry. Existing techniques assess symmetry by comparing the consistency of the shape, slope, time, or amplitude of the rising limb (from trough to peak) and the falling limb (from peak to trough). Specific calculation methods can be implemented based on existing techniques. In this embodiment, an allowable range of difference is set. For example, by fitting the rising limb as a sloping line, reversing it around the peak, and comparing its slope with that of the falling limb, the slope difference is calculated. This difference is then compared to a preset threshold; if the difference is less than the threshold, the waveform is considered symmetrical.

[0024] e3: Evaluate the peak quality of the pulse wave segment to be processed; Detect whether the waveform segment to be processed has double peaks or missing peaks; If double peaks or missing peaks exist, the waveform quality peakV is marked as 0; otherwise, it is marked as 1.

[0025] If a heartbeat lasts between 0.5s and 1s, and there are two or more peaks on the pulse wave segment within a heartbeat time, it is determined to be a double peak; if there is no peak, it is determined to be a missing peak.

[0026] e4: Evaluate the noise and distortion indices of the pulse wave segment to be processed; Evaluation of noise performance based on the signal-to-noise ratio (SNR) method; If the noise is significant, mark the noise and distortion index ND as 0 and stop the calculation. Otherwise, determine whether there is distortion in the shape of the pulse wave segment to be processed. If there is, mark the noise and distortion index as 0; otherwise, mark the noise and distortion index as 1.

[0027] Besides the biphasic or missing peaks mentioned above, pulse wave distortion can also manifest in other ways. For example, atrial fibrillation-related characteristics mainly include delayed onset of the rising wave or absolutely irregular intervals. Specific distortion features are based on historical data collection. Pulse wave distortion can be determined using existing computational methods or by constructing an atrial fibrillation distortion waveform detection and classification model based on neural networks to detect and classify the pulse wave data being processed, thus achieving distortion determination.

[0028] e5: For each pulse wave segment to be processed, evaluate the quality by performing the above calculations. If any index is marked as 0, it is determined that the pulse wave oscillation amplitude is irregular and is deleted.

[0029] After the calculation in step S3, each measurement independently generates a set of effective pulsation amplitudes and corresponding pressure points.

[0030] S4: To address the irregularity of the heartbeat, the amplitude of each effective pulsation segment is compensated to obtain the compensated amplitude; To suppress the abrupt change in pulsation amplitude caused by AF, this invention constructs a compensation factor C. i The original amplitude Ai is corrected to obtain the compensated amplitude. The calculation method for the compensated amplitude is as follows: , In the formula, To compensate for the amplitude, A i The amplitude of the original pulse wave corresponds to the i-th effective pulsation segment.

[0031] C i The compensation factor is calculated by comprehensively considering factors such as waveform differences, adjacent pulsation characteristics, and local pressure stability. ; In the formula, clip() is a clipping function that clips data exceeding the lower boundary value C. min and the upper boundary value C max The elements are replaced with the corresponding boundary values, and the compensation intensity is limited by clip() to avoid abnormal pulsations having an excessive impact on the oscillation envelope. In this embodiment, C min =0.6, C max =1.6; that is, when the input value is greater than C max When the input value is less than C, replace it with 1.6. min When the input value is 0.6, replace it with 0.6.

[0032] ε is the average amplitude in the pressure neighborhood; ε is a minimal constant to prevent division by zero, and ε takes the value of 0.00001 or a smaller positive number.

[0033] △A i This represents the change in amplitude between adjacent beats; △A i =A i -A i-1 .

[0034] In atrial fibrillation, the amplitude often increases or decreases abruptly between adjacent beats, hence the use of ΔA. i Depict AF transitions.

[0035] The method for calculating the average amplitude of the pressure neighborhood is as follows: ; In the formula, ΔP is the pressure neighborhood width, which is defined as the threshold range of similar pressures that divide a pressure interval, for example: 1 to 3 mmHg; when specifically selecting a value, under the condition that it is close to the current pulsating pressure, the amplitude should be "roughly at the same level", and this value is used as the reference amplitude, and ΔP is selected.

[0036] N is the total number of pressure zones included in the i-th effective pulsation segment; A j The original pulsation amplitude corresponding to the j-th pressure zone; P j : The representative pressure value corresponding to the current pulsation in the j-th pressure zone, and the cuff pressure corresponding to the i-th effective pulsation segment in the j-th pressure zone.

[0037] Compensation factor C i The calculation formula includes a jump suppression term. This method uses the jump suppression term to prevent overcompensation. The jump suppression term is as follows: , When the change in adjacent pulses is very small (waveform stable): |△A i If |≈0, then the jump suppression term≈1; When a sudden increase or decrease in strength of the pulsation occurs due to AF, |△A i As the value increases, the calculated value of the jump suppression term decreases.

[0038] λ is the jump suppression coefficient, which controls the strength of suppression against sudden changes in pulsation amplitude. When |ΔA i When the value of λ is large, reduce the variation range of the compensation factor to prevent overcorrection. The value range of λ is 0.3 to 0.8.

[0039] This application proposes a compensation factor C that can effectively suppress random jumps in beat amplitude caused by atrial fibrillation. i The original amplitude is compensated by introducing an average amplitude under similar pressure conditions. As a reference, the compensation intensity is adjusted by combining the changes in the amplitude of adjacent pulsations, thereby improving the stability of the oscillation envelope construction and ensuring the accuracy of the compensation results.

[0040] S5: For the same data to be processed, align the cuff pressure curve and the effective pulsation segment after amplitude compensation according to the horizontal axis, and perform a fusion operation to obtain a pressure amplitude curve; the horizontal axis of the pressure amplitude curve is pressure, and the vertical axis is the amplitude of the pulse wave. Specifically, this includes the following operations.

[0041] a1: Construct the pressure axis; The pressure values ​​are evenly divided into segments according to a preset range to obtain pressure intervals.

[0042] This method maps amplitude data from different measurements to multiple intervals on the pressure axis, also known as pressure buckets; in this embodiment, pressure intervals are constructed in segments of 1–2 mmHg. Subsequent calculations are performed for each pressure bucket.

[0043] a2: Using the time axis as a reference, align the amplitude values ​​of the effective pulsation segments after amplitude compensation to the pressure axis.

[0044] a3: For each pressure range, collect all compensated amplitude points that fall within this pressure range in the measurements, and calculate the weighted fusion amplitude A based on waveform quality, measurement quality, and stable segment weight. fusion The fused amplitude point sequence is obtained; ; In the formula, P j : The representative pressure value corresponding to the current pulsation in the j-th pressure zone; k is the measurement number, representing the kth independent blood pressure measurement process; i is the pulsation number, representing the i-th valid pulsation detected in the k-th measurement; The amplitude of the compensated pulsation is represented by the amplitude value of the i-th pulsation after irregularity compensation in the k-th measurement. w i k To integrate weights, ; q i k For waveform quality weights, m k To measure the mass weight, s i k Weights for stable segments; The summation symbol represents all measurements numbered k, pulses numbered i, and whose corresponding pressures fall within the pressure range P. j The pulsations accumulate.

[0045] This method allows multiple malformed envelopes to be merged into a smooth, statistically robust oscillating envelope.

[0046] The waveform quality weight q i k The waveform quality assessment for the pulsation level is calculated as follows.

[0047] b1: Segment the i-th valid pulse into a waveform, and mark the segmented waveform with waveform time data; The waveform time data includes: t 10 t 90 and t p ; Among them, t 10 This indicates the moment when the amplitude rises to 10% of its peak. t 90 The moment when the amplitude rises to 90% of its peak. t p Peak time; T 10→90 This represents the slope segment of the rising edge from the moment when the peak amplitude is 10% to the moment when the peak amplitude is 90%.

[0048] b2: The rise time is calculated as follows: T rise =t p -t 10 The rising slope segment T 10→90 =t 90 -t 10 .

[0049] b3: Calculate the morphological rationality score q shape,i (k) : ; In the formula, q shape,i (k) The waveform rationality score corresponding to the i-th valid pulse wave in the k-th measurement; μ T This is the mean value obtained from the statistical analysis of effective pulsations of the test subjects; σ T The variance is obtained from the effective pulsation statistics of the test subjects; The effective pulsation of a specific test subject can be calculated using data from this measurement or by collecting effective pulsation data from the subject's historical data.

[0050] b4: Evaluate noise / unsmoothness using second-order difference energy ratio; Define the waveform sequence of the i-th effective pulse as x[n], and L as the length of the i-th effective pulse wave on the horizontal axis; Define second-order difference: .

[0051] b5: Calculate the noise ratio r i (k) : ; In the formula, r i (k) The noise ratio corresponding to the i-th effective pulse wave in the k-th measurement; n is the n-th value of the horizontal axis; Let x[0], x[1], ..., x[L] be the average value of x[0], x[1], ..., x[L].

[0052] b6: Calculate the noise fraction q corresponding to the i-th valid pulse wave in the k-th measurement. noise,i (k) : ; In the formula, γ is the noise penalty intensity. In this embodiment, the value of γ is 5~20.

[0053] The greater the noise, the higher q noise,i (k) The smaller the value.

[0054] b7: Calculate the quality weight q of the fused waveform i k ; .

[0055] The closer the shape is to the "typical rising edge of this measurement" and the smoother the waveform (lower noise), the better q i k The closer it is to 1.

[0056] The measured quality weight m k The calculation method includes the following steps.

[0057] This method measures the mass weight m. k This indicates the effective pulsation ratio and envelope interpretability, where envelope interpretability includes: unimodality and smoothness.

[0058] c1: Calculate the effective pulsation ratio p (k) ; ; In the formula, N (k) N represents the total number of pulses detected in the k-th measurement. (k) valid The effective number of pulses obtained after passing quality screening; c2: Map each stroke count in this measurement to a pressure interval to obtain the coarse envelope point (P). j A j (k) ); A j (k) This represents the pulsation amplitude corresponding to the j-th pressure interval in the pulsation data obtained from the k-th measurement; P j This represents the representative pressure value corresponding to the current pulsation in the j-th pressure interval; For A j (k) Perform first-order difference: ; c3: The number of sign changes of the statistical derivative Z(k) This describes the extreme points included in the amplitude curve of the k-th measurement. ; In the formula, Z (k) Let J be the number of times the derivative sign changes corresponding to the k-th measurement, and J be the total number of pressure intervals corresponding to the k-th measurement data. The peak values ​​included in the pulse wave curve are the extreme points in the waveform. The number of times the derivative changes sign equals the total number of these extreme points, Z. (k) A larger value indicates a less unimodal nature; c4: Calculate the unimodal fraction u (k) Through single-peak fraction u (k) A simple evaluation of envelope unimodality and smoothness; ; In the formula, η is the sign change penalty intensity. In this embodiment, the value of η ranges from 0.2 to 1.0. c5: Calculate the measurement quality weight m k ; ; Unimodality refers to whether the amplitude curve exhibits only one major amplitude peak within a cardiac cycle. Smoothness describes the continuity of the amplitude curve; the more effective beats and the closer the envelope is to unimodal smoothness, the more reliable the overall measurement. (k) The closer it is to 1.

[0059] In this method, the stable segment weight s is used. i k The waveform quality is evaluated at the local pressure segment level to determine whether the amplitude fluctuation of the evaluated pulsation is stable within a similar pressure neighborhood. The stable segment weight s... i k The calculation method is shown below.

[0060] First, determine the local mean and dispersion of the pulsation within the pressure neighborhood.

[0061] d1: Evaluate the local mean and dispersion of the i-th effective pulsation corresponding to the k-th measurement within the neighborhood of the pressure interval; Define the neighborhood set of the i-th valid pulse corresponding to the k-th measurement. : ; Calculate the local mean μ i (k) ; ; Calculate the local standard deviation σ i (k) : ; Calculate the dimensionless coefficient of variation cv i (k) ; .

[0062] d2: Calculate the weight s of the stable segment. i k ; ; In the formula, β is the stability penalty intensity, and in this embodiment, β takes the value of 2 to 8.

[0063] The more consistent the amplitude (smaller cv) within the same pressure neighborhood, the more stable the condition. i k The closer it is to 1, the lower it becomes.

[0064] S6: Repeat steps S2 to S5; merge the pressure amplitude curves corresponding to all the data to be processed into the same coordinate system to obtain a unified pressure amplitude curve, denoted as: merged pressure amplitude curve.

[0065] S7: Determine the quality of the amplitude data corresponding to the fused pressure amplitude curve; If the quality results are sufficient to support subsequent calculations, proceed to step S8; Otherwise, continue to increase the number of measurements, and execute steps S2 to S6 for each measurement cycle.

[0066] Because the data for each measurement subject is different, this method pre-sets an initial number of measurements, such as three times, where k ranges from 1 to 3. After three measurements, irregular pulse wave segments are removed, resulting in enough effective pulse segments to construct a complete and smooth fused pressure amplitude curve. The quality of this fused pressure amplitude curve is then evaluated. If the quality is sufficient to support subsequent blood pressure calculations, the calculations can proceed directly. If the data quality is poor and the obtained effective pulse segments are insufficient to construct a complete and smooth curve, measurements are continued to extract more high-quality effective pulse segments. This method allows for flexible and adaptive adjustment of the number of measurements for different measurement subjects, making the measurement process more practical.

[0067] like Figure 2 As shown in this embodiment, in the three pulse wave oscillation sequences collected in the first three times, it is obvious that each curve in the original data of k=1~3 has a pulsating segment with abnormal waveform. After deleting the irregular segments of the three original pulse wave sequences and compensating for the effective pulse wave segments, a relatively smooth and complete fourth curve can be obtained: the fused pulse wave oscillation sequence.

[0068] The evaluation of the fused pressure amplitude curve is conducted from the perspectives of curve smoothness and symmetry. This can be done using existing pulse wave sequence evaluation methods, or by using high-quality pulse wave data from historical data as sample data to train a neural network model to evaluate the quality of the fused pressure amplitude curve.

[0069] S8: Transform the amplitude data in the fused pressure amplitude curve back into the coordinate system of the pulse wave oscillation sequence to obtain the fused pulse wave oscillation sequence (P). j A j fusion ); For the fused pulse wave oscillation sequence (P j A j fusion After smoothing and spline curve fitting, the oscillation envelope is extracted using existing techniques: connecting all peaks of the fused pulse wave oscillation sequence yields the upper envelope, and connecting all troughs yields the lower envelope. The two envelopes are denoted as: Fusion Oscillation Envelope E. target (P).

[0070] S9: The blood pressure value is obtained by subsequent calculation based on the fused oscillation envelope.

[0071] The specific method for calculating blood pressure values ​​is implemented based on existing technologies, such as: the envelope peak value corresponds to the MAP, and the SBP and DBP are obtained through the peak ratio relationship.

[0072] This method is the first to propose allowing "multiple inflation and deflation processes" and to fuse the effective pulses from each measurement across multiple measurements, thereby overcoming the problem of single-embedded envelope distortion caused by atrial fibrillation. This method does not require a single measurement to be completely accurate, but allows multiple cuff measurements to be performed on the same subject. By evaluating the quality, aligning the pressure, compensating for the amplitude, and fusing the pulse wave data obtained from multiple measurements, a unified "target oscillation envelope" is formed, and then the blood pressure value is obtained from this unified envelope.

[0073] After using the technical solution of this invention, the compensated amplitudes from different measurements are mapped to a unified pressure coordinate system through a pressure bucketing mechanism and pressure axis alignment. This application designs a cross-measurement amplitude fusion method, utilizing robust statistics or weighted fusion to construct a unified amplitude sequence. This method designs a compensation factor C based on waveform differences, pulsation quality, and local stability. iThis method ensures that anomalous jumps in the amplitude sequence are removed to the greatest extent possible before fusion. Instead of relying on the transient waveform of a single measurement, this method utilizes the statistical stability of multiple measurements to construct a robust oscillation envelope, improving accuracy and repeatability in atrial fibrillation. Furthermore, this method supports ECG-free conditions, relying solely on pulse waves and pressure curves to achieve ECG-free, atrial fibrillation-friendly measurements in devices without ECG capabilities. It is applicable to various portable devices and is more versatile than the atrial fibrillation BP market.

Claims

1. A non-invasive blood pressure estimation method based on the fusion of multiple pulse wave measurements, characterized in that, It includes the following steps: S1: Perform M consecutive cuff measurements on the same object, where M≥2; Collect data to be processed after each measurement; The data to be processed includes: cuff pressure curve and pulse wave oscillation sequence synchronized with pressure changes; The horizontal axis of the cuff pressure curve represents time, and the vertical axis represents the cuff pressure value. The horizontal axis of the pulse wave oscillation sequence represents time, and the vertical axis represents the amplitude of the pulse wave. S2: The pulse wave oscillation sequence in the data to be processed is denoted as: the pulse wave to be processed; Each pulse wave to be processed is divided and segmented based on the heartbeat to obtain the pulse wave segment to be processed; S3: Evaluate the quality of each pulse wave segment to be processed, find all segments with irregular pulse wave oscillation amplitudes, and delete them; the remaining pulse wave segments are denoted as: effective pulsation segments; S4: To address the irregularity of the heartbeat, the amplitude of each effective pulsation segment is compensated to obtain the compensated amplitude; S5: For the same data to be processed, the cuff pressure curve and the effective pulsation segment after amplitude compensation are aligned according to the horizontal axis and fused to obtain the pressure amplitude curve. The horizontal axis of the pressure amplitude curve represents pressure, and the vertical axis represents the amplitude of the pulse wave. S6: Repeat steps S2 to S5; merge the pressure amplitude curves corresponding to all the data to be processed into the same coordinate system to obtain a unified pressure amplitude curve, denoted as: merged pressure amplitude curve; S7: Determine the quality of the amplitude data corresponding to the fused pressure amplitude curve; If the quality results are sufficient to support subsequent calculations, proceed to step S8; Otherwise, continue to increase the number of measurements, and execute steps S2 to S6 for each measurement cycle; S8: Convert the amplitude data in the fused pressure amplitude curve to the coordinate system of the pulse wave oscillation sequence to obtain the fused pulse wave oscillation sequence; Based on the fused pulse wave oscillation sequence, the oscillation envelope is extracted and denoted as: fused oscillation envelope; S9: The blood pressure value is obtained by subsequent calculation based on the fused oscillation envelope.

2. The non-invasive blood pressure estimation method based on the fusion of multiple pulse wave measurements according to claim 1, characterized in that: In step S4, the method for calculating the compensated amplitude is as follows: ， In the formula, To compensate for the amplitude, A i C represents the original pulse wave amplitude corresponding to the i-th effective pulsation segment. i As a compensation factor; ； In the formula, clip() is a clipping function that limits the data exceeding the lower boundary value C. min and the upper boundary value C max Replace the elements with the corresponding boundary values; ε is the average amplitude in the pressure neighborhood; ε is a minimal constant to prevent division by zero. λ is the jump suppression coefficient, which controls the strength of suppression against sudden changes in pulsation amplitude; ΔA i This represents the change in amplitude between adjacent beats; △A i = A i -A i-1 。 3. The non-invasive blood pressure estimation method based on the fusion of multiple pulse wave measurements according to claim 2, characterized in that: The method for calculating the average amplitude of the pressure neighborhood is as follows: ； In the formula, ΔP is the pressure neighborhood width, defined as the threshold range of similar pressures that divide a pressure interval; N is the total number of pressure intervals included in the i-th effective pulsation segment; A j The original pulsation amplitude corresponding to the j-th pressure zone; P j : The representative pressure value corresponding to the current pulsation in the j-th pressure zone.

4. The non-invasive blood pressure estimation method based on the fusion of multiple pulse wave measurements according to claim 1, characterized in that: Step S5 specifically includes the following operations: a1: Construct the pressure axis; The pressure values ​​are evenly divided into segments according to a preset range to obtain pressure intervals; a2: Using the time axis as a reference, align the amplitude values ​​of the effective pulsation segments after amplitude compensation to the pressure axis; a3: For each pressure range, collect all compensated amplitude points that fall within this pressure range from all measurements, and calculate the weighted fused amplitude A. fusion The fused amplitude point sequence is obtained; ； In the formula, P j : The representative pressure value corresponding to the current pulsation in the j-th pressure zone; k is the measurement number, indicating the kth independent blood pressure measurement process; i is the pulsation number, representing the i-th valid pulsation detected in the k-th measurement; Let be the amplitude value of the i-th pulse after irregularity compensation in the k-th measurement; w i k To integrate weights, w i k = q i k ×m k ×s i k ; q i k For waveform quality weights, m k To measure the mass weight, s i k Weights for stable segments.

5. The non-invasive blood pressure estimation method based on the fusion of multiple pulse wave measurements according to claim 4, characterized in that: The waveform quality weight q i k The calculation method is as follows: b1: Segment the i-th valid pulse into a waveform, and mark the segmented waveform with waveform time data; The waveform time data includes: t 10 t 90 and t p ; Among them, t 10 This indicates the moment when the amplitude rises to 10% of its peak. t 90 The moment when the amplitude rises to 90% of its peak. t p Peak time; b2: The rise time is calculated as follows: T rise =t p -t 10 The rising slope segment T 10→90 =t 90 -t 10 ; b3: Calculate the morphological rationality score q shape,i (k) : ; In the formula, q shape,i (k) The waveform rationality score corresponding to the i-th valid pulse wave in the k-th measurement; μ T This is the mean value obtained from the statistical analysis of effective pulsations of the test subjects; σ T The variance is obtained from the effective pulsation statistics of the test subjects; b4: Define the waveform sequence of the i-th effective pulse as x[n], where L is the length of the i-th effective pulse wave on the horizontal axis; Define second-order difference: Δ 2 x[n] = x[n] - 2x[n-1] + x[n-2]; b5: Calculate the noise ratio r i (k) : ； In the formula, r i (k) The noise ratio corresponding to the i-th effective pulse wave in the k-th measurement; n is the n-th value of the horizontal axis; The average value of x[0] to x[L]; b6: Calculate the noise fraction q corresponding to the i-th valid pulse wave in the k-th measurement. noise,i (k) : q noise,i (k) =exp(-γr i (k) ); In the formula, γ is the noise penalty intensity; b7: Calculate the quality weights of the fused waveform; q i k =clip(q shape,i (k) ×q noise,i (k) ,0,1) 。 6. The non-invasive blood pressure estimation method based on the fusion of multiple pulse wave measurements according to claim 4, characterized in that: The measured quality weight m k The calculation method is as follows: c1: Calculate the effective pulsation ratio p (k) ; p (k) =N (k) valid / (N (k) + e); In the formula, N (k) N represents the total number of pulses detected in the k-th measurement. (k) valid The effective number of pulses obtained after passing quality screening; c2: Map each stroke count in this measurement to a pressure interval to obtain the coarse envelope point (P). j A j (k) ); A j (k) This represents the pulsation amplitude corresponding to the j-th pressure interval in the pulsation data obtained from the k-th measurement; P j This represents the representative pressure value corresponding to the current pulsation in the j-th pressure interval; For A j (k) Perform first-order difference: d j = A j (k) - A j-1 (k) ; c3: The number of sign changes of the statistical derivative Z (k) This describes the extreme points included in the amplitude curve of the k-th measurement. ; In the formula, Z (k) Let J be the number of times the derivative sign changes corresponding to the k-th measurement, and J be the total number of pressure intervals corresponding to the k-th measurement data. c4: Calculate the unimodal fraction u (k) ; u (k) = exp(-ηZ (k) ); In the formula, η represents the intensity of the sign change penalty; c5: Calculate the measurement quality weight m k ; m k = clip(p (k) ×u (k) ,0,1) 。 7. The non-invasive blood pressure estimation method based on the fusion of multiple pulse wave measurements according to claim 4, characterized in that: The stable segment weights s i k The calculation method is as follows: d1: Evaluate the local mean and dispersion of the i-th effective pulsation corresponding to the k-th measurement within the neighborhood of the pressure interval; Define the neighborhood set of the i-th valid pulse corresponding to the k-th measurement. : ； Calculate the local mean μ i (k) ; ; Calculate the local standard deviation σ i (k) : ; Calculate the dimensionless coefficient of variation cv i (k) ; ; d2: Calculate the weight s of the stable segment. i k ; s i k =exp(- βcv i (k) ); In the formula, β is the stability penalty intensity.

8. The non-invasive blood pressure estimation method based on the fusion of multiple pulse wave measurements according to claim 1, characterized in that: In step S3, the quality of each pulse wave segment to be processed is evaluated, and all segments with irregular pulse wave oscillation amplitudes are identified and deleted; specifically, the following operations are performed: e1: Evaluate the quality of the rising edge slope; Calculate the rising slope of the pulse wave segment to be processed and compare it with a preset slope range; When the value is outside the range, the rising slope quality RiseQ is recorded as 0; otherwise, it is recorded as 1. e2: Evaluate the symmetry of the waveform of the pulse wave segment to be processed; If the waveform is symmetrical about the R-peak, then the symmetry quality symmetryQ is marked as 1; otherwise, it is marked as 0. e3: Evaluate the peak quality of the pulse wave segment to be processed; Detect whether the waveform segment to be processed has double peaks or missing peaks; If double peaks or missing peaks exist, the waveform quality peakV is marked as 0; otherwise, it is marked as 1. e4: Evaluate the noise and distortion indices of the pulse wave segment to be processed; Evaluation of noise performance based on the signal-to-noise ratio (SNR) method; If the noise is significant, mark the noise and distortion index ND as 0 and stop the calculation. Otherwise, determine whether there is distortion in the shape of the pulse wave segment to be processed. If there is, mark the noise and distortion index as 0; otherwise, mark the noise and distortion index as 1. e5: For each pulse wave segment to be processed, evaluate the quality by performing the above calculations. If any index is marked as 0, it is determined that the pulse wave oscillation amplitude is irregular and is deleted.

Images