A method for pulse wave quality classification and stable waveform extraction

Abstract

The present invention relates to a method for pulse wave waveform quality grade classification and stable waveform extraction, comprising: obtaining a pulse wave signal from a body surface acquisition site, preprocessing the pulse wave signal; performing single-cycle pulse wave segmentation; judging whether P_T satisfies a constraint, if so, proceeding to step (4), otherwise proceeding to step (6); judging whether P_TP satisfies a constraint, if so, proceeding to step (5), otherwise, proceeding to step (6); judging whether P_TPD satisfies a constraint, if so, marking the waveform quality grade of P_TPD as the first grade, proceeding to step (6), otherwise, marking the waveform quality grade of P_TPD as the second grade, proceeding to step (6); stable waveform extraction and quality grade marking. The present invention can preferentially eliminate abnormal single-cycle pulse waves, avoid the influence of abnormal data, is more accurate than the result of using the average value method, and provides a guarantee for the reliability of subsequent analysis results.

Description

Technical Field

[0001] The invention relates to the technical field of biological signal processing, in particular to a method for dividing pulse wave waveform quality levels and extracting stable waveforms. Background Art

[0002] Pulse wave is an important signal source for human health detection. With the development of sensor technology, people can quickly and non-invasively obtain pulse wave signals from multiple parts of the human body surface. Since pulse wave contains rich human hemodynamic information, pulse wave waveform analysis methods are often used clinically to extract relevant physiological parameters, such as the Augmentation Index (AI) and heart rate (HR), to perform non-invasive cardiovascular health assessment.

[0003] However, the computer analysis software of non-invasive detection methods does not pay enough attention to the pulse signal quality assessment link. If the pulse waveform within one or several cardiac cycles used in parameter calculation has serious quality problems, the calculation results will be unreliable. To address this problem, the commonly used method is to combine manual waveform judgment and average calculation to evaluate whether the results are reliable. On the one hand, the method of obtaining a "stable" pulse wave signal through manual visual observation has artificial subjective influence, and it may require multiple operations in actual applications to collect a pulse wave signal that meets the requirements; on the other hand, the average value is used by computer software to reduce the influence of abnormal values, but it cannot fundamentally eliminate the influence. Existing studies have used the Signal Quality Index (SQI) to evaluate the quality of a continuous signal. Commonly used signal quality indices include kurtosis and skewness. However, the calculation methods of these indicators are complicated, and they are for a continuous pulse wave waveform. The evaluation results are generally only good and bad. It is a rough evaluation method and cannot be further evaluated for each single-cycle pulse wave. Generally speaking, the contour of the pulse wave will change to a certain extent due to aging and disease progression. Obviously, as people age, the dicrotic wave will continue to move forward, making it impossible to distinguish it with the naked eye, or the noise during detection will cause a significant change in the shape of a single-cycle pulse wave. If this "abnormal" single-cycle pulse wave is used as the analysis object, the analysis result will be biased. Summary of the invention

[0004] In order to solve the defect of insufficient quality assessment of existing pulse wave signals, the purpose of the present invention is to provide a pulse wave waveform quality grading and stable waveform extraction method which can further refine the grading of pulse wave quality and facilitate the extraction of a stable pulse wave waveform.

[0005] To achieve the above object, the present invention adopts the following technical solution: a method for pulse wave waveform quality classification and stable waveform extraction, the method comprising the following steps in order:

[0006] (1) obtaining a pulse wave signal from a body surface collection site, preprocessing the pulse wave signal, and obtaining a pulse wave waveform S1 to be determined;

[0007] (2) performing single-cycle pulse wave segmentation on the pulse wave waveform S1 to obtain a single-cycle pulse wave set P;

[0008] (3) Identify the qualified pulsation period of the single-cycle pulse wave set P to obtain the single-cycle pulse wave set P_T with qualified pulsation period, and judge whether the single-cycle pulse wave set P_T with qualified pulsation period meets the constraint. If the single-cycle pulse wave set P_T with qualified pulsation period meets the constraint, go to step (4); otherwise, mark the waveform quality level of the single-cycle pulse wave set P as level 4, and go to step (6);

[0009] (4) Identify the qualified main wave of the single-cycle pulse wave set P_T with qualified pulsation period, obtain the single-cycle pulse wave set P_TP with qualified main wave, and judge whether the single-cycle pulse wave set P_TP with qualified main wave meets the constraint. If the single-cycle pulse wave set P_TP with qualified main wave meets the constraint, go to step (5); otherwise, mark the waveform quality level of the single-cycle pulse wave set P_T with qualified pulsation period as level 3, and go to step (6);

[0010] (5) Identify the qualified dicrotic waves of the main wave qualified single-cycle pulse wave set P_TP, obtain the qualified dicrotic wave single-cycle pulse wave set P_TPD, and judge whether the qualified dicrotic wave single-cycle pulse wave set P_TPD meets the constraints. If the qualified dicrotic wave single-cycle pulse wave set P_TPD meets the constraints, mark the waveform quality level of the qualified dicrotic wave single-cycle pulse wave set P_TPD as the first level, and go to step (6); otherwise, mark the waveform quality level of the main wave qualified single-cycle pulse wave set P_TPD as the second level, and go to step (6);

[0011] (6) Stable waveform extraction and quality level marking: The single-cycle pulse waves are combined into a stable pulse waveform S in chronological order, and the quality level corresponding to the stable pulse waveform S is marked.

[0012] The step (1) specifically refers to: collecting pulse waves from any collection site on the body, the collection site includes major arterial sites and body surface sites where pulse waves are easily obtained, recording the collected pulse wave signal as an original pulse wave signal, performing a preprocessing operation on the original pulse wave signal to remove high-frequency interference and baseline drift, and recording the obtained waveform as a pending pulse wave waveform S1, wherein the pending pulse wave waveform S1 contains multiple continuous pulsation period signals and / or chaotic signals.

[0013] The step (2) specifically includes the following steps: using the differential threshold method to segment the pulse waveform S1 to obtain N single-cycle pulse waves, the set of N single-cycle pulse waves is denoted as P, that is, P = {Pi|i = 1, 2, ..., N}, specifically:

[0014] (2a) performing a first-order difference operation on the pulse wave waveform S1 to obtain a first-order difference waveform S2;

[0015] (2b) The first-order differential waveform S2 is evenly divided into m segments, and the maximum value of each segment of the first-order differential waveform is extracted. There are a total of m segment maximum values, namely M1, M2, ... Mm. Then the threshold θ of the differential signal is:

[0016]

[0017] Among them, k is a weight coefficient ranging from 0 to 1, and its value is 0.6. is the average value of the maximum values ​​of m segments;

[0018] (2c) Traverse the first-order difference waveform S2 and extract the positions of the local maximum points greater than the threshold θ. These points are recorded as the differential maximum point set Pos, which are marked as Pos1, Pos2, ..., PosN, Pos(N+1) in sequence, that is, Pos = {Posi|i = 1, 2, ..., N, N+1};

[0019] (2d) The differential threshold method is: for each differential maximum value to be determined point Posi, a differential waveform of a window is traversed in reverse order, and the length of the window L is:

[0020] L=t×Fs

[0021] Among them, Fs is the sampling frequency of the signal, t is the weighting coefficient of the pulse cycle, and its value is 0.3;

[0022] In the i-th window, find the position of the first point where the first-order differential signal is less than 0, recorded as Ai. If not found, set the position of the starting point Ai as the starting point of the current window, Ai = Posi, indicating that Ai is a virtual position point, as a mark for subsequent elimination, these points are recorded as the starting point set A, marked as A1, A2, ..., AN, A (N + 1) in sequence, that is, A = {Ai|i = 1, 2, ..., N, N + 1};

[0023] (2e) The pending pulse wave waveform S1 is divided into independent single-cycle pulse waves according to the starting pending points A1, A2, ..., AN, A(N+1) in sequence, wherein the single-cycle pulse wave between point A1 and point A2 is denoted as P1, the single-cycle pulse wave between point A2 and point A3 is denoted as P2, ..., the single-cycle pulse wave between point AN and point A(N+1) is denoted as PN, and these single-cycle pulse wave sets are denoted as set P, i.e., P = {Pi|i = 1, 2, ..., N}.

[0024] The step (3) specifically comprises the following steps:

[0025] (3a) For each single-cycle pulse wave, calculate the time interval set T of its pulse cycle, which are recorded as T1, T2, ..., T(N-1), TN in sequence:

[0026] T i =A i+1 -A i

[0027] Wherein, i is an integer from 1 to N, and Ai is the starting point to be determined;

[0028] (3b) In the time interval set T, for the i-th time interval Ti, if the starting point Ai is a virtual position point, that is, Ai coincides with the position of the maximum difference point Posi, Ai == Posi, then the time interval Ti is removed from the time interval set T, otherwise the time interval Ti is retained, and the set of all retained time intervals Ti is recorded as the clustering set Set1;

[0029] (3c) Perform cluster analysis on cluster set Set1: Use K-means algorithm to perform 3 cluster clustering on cluster set Set1, select the cluster with the largest number of middle ones as the clustering result, obtain the clustering result of time interval T, record the corresponding single-cycle pulse wave waveform set as P_T={Pi|i=1,2,…,N_T and N_T<=N}, and count the number N_T; the constraint refers to: whether N_T satisfies the threshold c×N, if N_T≥c×N, c is the clustering coefficient, c∈(0.5,1), then go to step (4), otherwise mark the single-cycle pulse wave set P as the fourth-level waveform, and subsequent analysis cannot be performed, go to step (6).

[0030] The step (4) specifically comprises the following steps:

[0031] (4a) For each single-cycle pulse wave P_Ti, the differential threshold method is used to identify the main wave point position in the waveform, and the time interval TPi from the starting point A_T to the main wave point is calculated; the differential threshold method for main wave point position identification means that within a time window after the starting point A_T, the first point where the first-order differential waveform S2 is less than 0 is the main wave point;

[0032] (4b) The time intervals TPi of all main wave points are merged into a cluster set Set2, cluster analysis is performed on the cluster set Set2, and the corresponding single-cycle pulse wave waveform set is recorded as P_TP={Pi|i=1,2,…,N_TP and N_TP<=N_T}, and the number N_TP is counted; the constraint refers to: whether N_TP satisfies the threshold c×N_T, c is the clustering coefficient, c∈(0.5,1): if N_TP≥c×N_T, go to step (5); otherwise, mark the single-cycle pulse wave set P_T as the third level waveform, and subsequent analysis cannot be performed, and go to step (6).

[0033] The step (5) specifically comprises the following steps:

[0034] (5a) For each single-cycle pulse wave P_TPi, the dicrotic wave point position in the waveform is identified using the differential threshold method, and the time interval TPNi from the starting point A_TP to the dicrotic wave point is calculated; the differential threshold method for identifying the dicrotic wave point position means that within a time window after the main wave point, the first point where the first-order differential waveform S1 is greater than 0 is the dicrotic wave point. If there is no such point, the local maximum value point of the first-order differential waveform S1 is taken as the dicrotic wave point;

[0035] (5b) The time intervals TPNi of all dicrotic wave points are merged into a cluster set Set3, and cluster analysis is performed on the cluster set Set3. The corresponding single-cycle pulse wave waveform set is recorded as P_TPD = {Pi|i = 1, 2, ..., N_TPD and N_TPD < = N_TP}, and the number N_TPD is counted. The constraint refers to whether N_TPD satisfies the threshold c × N_TP, where c is the clustering coefficient, c∈(0.5, 1): If N_TPD ≥ c × N_TP, then the single-cycle pulse wave set P_TPD is marked as the first-level waveform, and the dicrotic wave feature is stable, and go to step (6); otherwise, the single-cycle pulse wave set P_TP is marked as the second-level waveform, and the main wave feature is stable, and go to step (6).

[0036] The step (6) specifically means: if the single-cycle pulse wave is obtained from step (3), the single-cycle pulse waves in the single-cycle pulse wave set P are combined in time sequence into a stable pulse wave waveform, which is recorded as S, and the waveform quality level is level 4; if the single-cycle pulse wave is obtained from step (4), the single-cycle pulse waves in the single-cycle pulse wave set P_T with qualified pulse cycles are combined in time sequence into a stable pulse wave waveform, which is recorded as S, and the waveform quality level is level 3; if the single-cycle pulse wave is obtained from step (5) and the result is If the main wave feature is stable, the single-cycle pulse waves in the single-cycle pulse wave set P_TP with qualified main waves are combined into a stable pulse wave waveform in chronological order, denoted as S, and the waveform quality level is the second level; if the analysis result is obtained from step (5) and the result is that the dicrotic wave feature is stable, the single-cycle pulse waves in the single-cycle pulse wave set P_TPD with qualified dicrotic waves are combined into a stable pulse wave waveform in chronological order, denoted as S, and the waveform quality level is the first level; finally, a stable pulse wave waveform S and its corresponding quality level are obtained.

[0037] It can be seen from the above technical scheme that the beneficial effects of the present invention are as follows: first, the present invention first uses the differential threshold method to perform single-cycle pulse wave segmentation on the pulse wave waveform S1 to be determined, and then uses the quantitative index to quantitatively describe the waveform of each single-cycle pulse wave. According to the result of the cluster analysis of the quantitative index, a stable single-cycle pulse wave is extracted from all single-cycle pulse waves, and they are merged into a stable pulse wave waveform S in chronological order, and the quality level of the waveform is marked, so that the division of the pulse wave quality level is more refined; second, the present invention can preferentially eliminate abnormal single-cycle pulse waves, avoid the influence of abnormal data, and is more accurate than the result of using the average value method, which provides a guarantee for the reliability of subsequent analysis results; third, the intermediate results of the present invention, such as the reference point position, can also be used for subsequent pulse wave waveform analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 is a flow chart of the method of the present invention;

[0039] Figure 2 Screening diagram for time intervals;

[0040] Figure 3 It is a schematic diagram of the reference points of a typical single-cycle pulse wave;

[0041] Figure 4 It is the triangle characteristic diagram of single cycle pulse wave waveform;

[0042] Figure 5 This is a schematic diagram of the intuitive effect of the pulse wave quality level being level 1. DETAILED DESCRIPTION

[0043] like Figure 1 As shown, a method for pulse wave waveform quality classification and stable waveform extraction includes the following steps in order:

[0044] (1) obtaining a pulse wave signal from a body surface collection site, preprocessing the pulse wave signal, and obtaining a pulse wave waveform S1 to be determined;

[0045] (2) performing single-cycle pulse wave segmentation on the pulse wave waveform S1 to obtain a single-cycle pulse wave set P;

[0046] (3) Identify the qualified pulsation period of the single-cycle pulse wave set P to obtain the single-cycle pulse wave set P_T with qualified pulsation period, and judge whether the single-cycle pulse wave set P_T with qualified pulsation period meets the constraint. If the single-cycle pulse wave set P_T with qualified pulsation period meets the constraint, go to step (4); otherwise, mark the waveform quality level of the single-cycle pulse wave set P as level 4, and go to step (6);

[0047] (4) Identify the qualified main wave of the single-cycle pulse wave set P_T with qualified pulsation period, obtain the single-cycle pulse wave set P_TP with qualified main wave, and judge whether the single-cycle pulse wave set P_TP with qualified main wave meets the constraint. If the single-cycle pulse wave set P_TP with qualified main wave meets the constraint, go to step (5); otherwise, mark the waveform quality level of the single-cycle pulse wave set P_T with qualified pulsation period as level 3, and go to step (6);

[0048] (5) Identify the qualified dicrotic waves of the main wave qualified single-cycle pulse wave set P_TP, obtain the qualified dicrotic wave single-cycle pulse wave set P_TPD, and judge whether the qualified dicrotic wave single-cycle pulse wave set P_TPD meets the constraints. If the qualified dicrotic wave single-cycle pulse wave set P_TPD meets the constraints, mark the waveform quality level of the qualified dicrotic wave single-cycle pulse wave set P_TPD as the first level, and go to step (6); otherwise, mark the waveform quality level of the main wave qualified single-cycle pulse wave set P_TPD as the second level, and go to step (6);

[0049] (6) Stable waveform extraction and quality level marking: The single-cycle pulse waves are combined into a stable pulse waveform S in chronological order, and the quality level corresponding to the stable pulse waveform S is marked.

[0050] The step (1) specifically refers to: collecting pulse waves from any collection site on the body, the collection site includes major arterial sites and body surface sites where pulse waves are easily obtained, recording the collected pulse wave signal as an original pulse wave signal, performing a preprocessing operation on the original pulse wave signal to remove high-frequency interference and baseline drift, and recording the obtained waveform as a pending pulse wave waveform S1, wherein the pending pulse wave waveform S1 contains multiple continuous pulsation period signals and / or chaotic signals.

[0051] The step (2) specifically includes the following steps: using the differential threshold method to segment the pulse waveform S1 to obtain N single-cycle pulse waves, the set of N single-cycle pulse waves is denoted as P, that is, P = {Pi|i = 1, 2, ..., N}, specifically:

[0052] (2a) performing a first-order difference operation on the pulse wave waveform S1 to obtain a first-order difference waveform S2;

[0053] (2b) The first-order differential waveform S2 is evenly divided into m segments, and the maximum value of each segment of the first-order differential waveform is extracted. There are a total of m segment maximum values, namely M1, M2, ... Mm. Then the threshold θ of the differential signal is:

[0054]

[0055] Among them, k is a weight coefficient ranging from 0 to 1, and its value is 0.6. is the average value of the maximum values ​​of m segments;

[0056] (2c) Traverse the first-order difference waveform S2 and extract the positions of the local maximum points greater than the threshold θ. These points are recorded as the differential maximum point set Pos, which are marked as Pos1, Pos2, ..., PosN, Pos(N+1) in sequence, that is, Pos = {Posi|i = 1, 2, ..., N, N+1};

[0057] (2d) The differential threshold method is: for each differential maximum value to be determined point Posi, a differential waveform of a window is traversed in reverse order, and the length of the window L is:

[0058] L=t×Fs

[0059] Among them, Fs is the sampling frequency of the signal, t is the weighting coefficient of the pulse cycle, and its value is 0.3;

[0060] In the i-th window, find the position of the first point where the first-order differential signal is less than 0, recorded as Ai. If not found, set the position of the starting point Ai as the starting point of the current window, Ai = Posi, indicating that Ai is a virtual position point, as a mark for subsequent elimination, these points are recorded as the starting point set A, marked as A1, A2, ..., AN, A (N + 1) in sequence, that is, A = {Ai|i = 1, 2, ..., N, N + 1};

[0061] (2e) The pending pulse wave waveform S1 is divided into independent single-cycle pulse waves according to the starting pending points A1, A2, ..., AN, A(N+1) in sequence, wherein the single-cycle pulse wave between point A1 and point A2 is denoted as P1, the single-cycle pulse wave between point A2 and point A3 is denoted as P2, ..., the single-cycle pulse wave between point AN and point A(N+1) is denoted as PN, and these single-cycle pulse wave sets are denoted as set P, i.e., P = {Pi|i = 1, 2, ..., N}.

[0062] The step (3) specifically comprises the following steps:

[0063] (3a) For each single-cycle pulse wave, calculate the time interval set T of its pulse cycle, which are recorded as T1, T2, ..., T(N-1), TN in sequence:

[0064] T i =A i+1 -A i

[0065] Wherein, i is an integer from 1 to N, and Ai is the starting point to be determined;

[0066] (3b) In the time interval set T, for the i-th time interval Ti, if the starting point Ai is a virtual position point, that is, Ai coincides with the position of the maximum difference point Posi, Ai == Posi, then the time interval Ti is removed from the time interval set T, otherwise the time interval Ti is retained, and the set of all retained time intervals Ti is recorded as the clustering set Set1, as shown in Figure 2 As shown;

[0067] (3c) Perform cluster analysis on cluster set Set1: Use K-means algorithm to perform 3 cluster clustering on cluster set Set1, select the cluster with the largest number of middle ones as the clustering result, obtain the clustering result of time interval T, record the corresponding single-cycle pulse wave waveform set as P_T={Pi|i=1,2,…,N_T and N_T<=N}, and count the number N_T; the constraint refers to: whether N_T satisfies the threshold c×N, if N_T≥c×N, c is the clustering coefficient, c∈(0.5,1), then go to step (4), otherwise mark the single-cycle pulse wave set P as the fourth-level waveform, and subsequent analysis cannot be performed, go to step (6).

[0068] The step (4) specifically comprises the following steps:

[0069] (4a) For each single-cycle pulse wave P_Ti, the differential threshold method is used to identify the main wave point position in the waveform, and the time interval TPi from the starting point A_T to the main wave point is calculated; the differential threshold method for main wave point position identification means that within a time window after the starting point A_T, the first point where the first-order differential waveform S2 is less than 0 is the main wave point;

[0070] (4b) The time intervals TPi of all main wave points are merged into a cluster set Set2, cluster analysis is performed on the cluster set Set2, and the corresponding single-cycle pulse wave waveform set is recorded as P_TP={Pi|i=1,2,…,N_TP and N_TP<=N_T}, and the number N_TP is counted; the constraint refers to: whether N_TP satisfies the threshold c×N_T, c is the clustering coefficient, c∈(0.5,1): if N_TP≥c×N_T, go to step (5); otherwise, mark the single-cycle pulse wave set P_T as the third level waveform, and subsequent analysis cannot be performed, and go to step (6).

[0071] The step (5) specifically comprises the following steps:

[0072] (5a) For each single-cycle pulse wave P_TPi, the dicrotic wave point position in the waveform is identified using the differential threshold method, and the time interval TPNi from the starting point A_TP to the dicrotic wave point is calculated; the differential threshold method for identifying the dicrotic wave point position means that within a time window after the main wave point, the first point where the first-order differential waveform S1 is greater than 0 is the dicrotic wave point. If there is no such point, the local maximum value point of the first-order differential waveform S1 is taken as the dicrotic wave point;

[0073] (5b) The time intervals TPNi of all dicrotic wave points are merged into a cluster set Set3, and cluster analysis is performed on the cluster set Set3. The corresponding single-cycle pulse wave waveform set is recorded as P_TPD = {Pi|i = 1, 2, ..., N_TPD and N_TPD < = N_TP}, and the number N_TPD is counted. The constraint refers to whether N_TPD satisfies the threshold c × N_TP, where c is the clustering coefficient, c∈(0.5, 1): If N_TPD ≥ c × N_TP, then the single-cycle pulse wave set P_TPD is marked as the first-level waveform, and the dicrotic wave feature is stable, and go to step (6); otherwise, the single-cycle pulse wave set P_TP is marked as the second-level waveform, and the main wave feature is stable, and go to step (6).

[0074] The step (6) specifically means: if the single-cycle pulse wave is obtained from step (3), the single-cycle pulse waves in the single-cycle pulse wave set P are combined in time sequence into a stable pulse wave waveform, which is recorded as S, and the waveform quality level is level 4; if the single-cycle pulse wave is obtained from step (4), the single-cycle pulse waves in the single-cycle pulse wave set P_T with qualified pulse cycles are combined in time sequence into a stable pulse wave waveform, which is recorded as S, and the waveform quality level is level 3; if the single-cycle pulse wave is obtained from step (5) and the result is If the main wave feature is stable, the single-cycle pulse waves in the single-cycle pulse wave set P_TP with qualified main waves are combined into a stable pulse wave waveform in chronological order, denoted as S, and the waveform quality level is the second level; if the analysis result is obtained from step (5) and the result is that the dicrotic wave feature is stable, the single-cycle pulse waves in the single-cycle pulse wave set P_TPD with qualified dicrotic waves are combined into a stable pulse wave waveform in chronological order, denoted as S, and the waveform quality level is the first level; finally, a stable pulse wave waveform S and its corresponding quality level are obtained.

[0075] like Figure 3 As shown, for each single-cycle pulse wave, the main components of the waveform include the main wave, the dicrotic wave and the reflected wave, and the present invention uses reference points to describe these waveform components. Generally, the main reference points of a single-cycle pulse wave are the starting point A, the main wave point P, the dicrotic wave point D, the slope point K, the reflection point R and the descending gorge point N. The waveform characteristics are described in the form of coordinate axis description, and the coordinate axis form of the reference point is: (time parameter, amplitude parameter). In the technical solution of the present invention, the starting point A, the main wave point P and the dicrotic wave point D are used as reference points, and the pulse wave contour is simply described using time parameters. Similarly, the amplitude parameters and other reference points (such as the descending gorge point N, etc.) can also be included as the description of the waveform characteristics, which is also within the protection scope of the present invention.

[0076] In addition, the description of waveform characteristics is not limited to the description in the form of coordinate axes, such as Figure 4As shown, it is a method for describing the waveform characteristics of a triangle. There are three triangles in total: Triangle 1 (starting point A, main wave point P, descending mid-isthmus point N), Triangle 2 (starting point A, main wave point P, re-isthmus point D) and Triangle 3 (starting point A, main wave point P, starting point A').

[0077] K-means clustering method, to obtain the median cluster among the three clusters, can adaptively select the mode data in the index, remove the larger or smaller outliers, and combined with the designed metric threshold, it can determine whether the clustering effect of the median cluster is stable. If it is unstable, the current pulse wave waveform quality level does not meet the requirements of the target characteristics, and the quality is marked as the closest target characteristic level. If it is stable, the current pulse wave waveform quality level meets the requirements of the target characteristics, and the judgment of the next target characteristic is continued.

[0078] The judgment of the target characteristics should be in order. Generally, the characteristics should be judged in the order of periodicity, main wave, dicrotic wave, and reflected wave, and the judgment of each characteristic should be based on the previous characteristic judgment. First, the entire pending pulse wave waveform S1 should be judged for periodicity, that is, the pulsation period can be identified in the waveform; then, the main wave characteristics of each single-cycle pulse wave should be identified, and it should be judged whether the overall main wave characteristics are stable, that is, the stable main wave can be identified in the waveform; secondly, the dicrotic wave characteristics of each single-cycle pulse wave should be identified, and it should be judged whether the overall dicrotic wave characteristics are stable, that is, the stable dicrotic wave can be identified in the waveform; finally, the reflected wave characteristics of each single-cycle pulse wave should be identified, and it should be judged whether the overall reflected wave characteristics are stable, that is, the stable reflected wave can be identified in the waveform.

[0079] The determination of target characteristics in the present invention can be disassembled and assembled. In practical applications, if you are only concerned with identifying whether the heart rate is abnormal, you only need to determine the characteristics of the pulsation cycle; if you are concerned with the evaluation of cardiac ejection function, you need to first determine the characteristics of the arterial cycle, and then determine the characteristics of the main wave; and the same goes for more detailed concerns. Therefore, other description methods based on the characteristics of some waveform components (such as based on the main wave, based on the dicrotic wave, etc.) should also be within the scope of protection of the present invention.

[0080] An intuitive effect is Figure 5As shown, the undetermined pulse wave waveform S1 is divided into N single-cycle pulse waves, which are recorded as P1, P2, ..., PN in chronological order. Through the method of the present invention, the single-cycle pulse wave P5 is first eliminated due to the "abnormality" of the main wave characteristics, and then the single-cycle pulse wave P3 is eliminated due to the "abnormality" of the dicrotic wave characteristics, and the other single-cycle pulse waves are deduced in the same way. Finally, all the single-cycle pulse waves P1, P2, P4, P6, ..., PN_TPN that pass the waveform characteristic inspection are extracted separately and merged into a new pulse wave waveform, which is recorded as the stable pulse wave waveform S. It can be seen intuitively that in the stable pulse wave waveform S, the pulsation period is stable, the main wave is stable, and the dicrotic wave is stable, so the quality level is level 1.

[0081] In summary, in the research based on pulse wave waveform analysis, the present invention can preferentially eliminate abnormal single-cycle pulse waves, avoid the influence of abnormal data, and is more accurate than the result of using the average value method, which provides a guarantee for the reliability of subsequent analysis results. In addition, the intermediate results (such as the reference point position) of the method proposed by the present invention can also be used for subsequent pulse wave waveform analysis.

Claims

1. A method for pulse wave quality classification and stable waveform extraction, characterized in that: The method comprises the following steps in order: (1) Obtaining a pulse wave signal from a body surface collection site, preprocessing the pulse wave signal, and obtaining a pulse wave waveform S1 to be determined; (2) Segment the pulse wave S1 into single-cycle pulse waves to obtain a single-cycle pulse wave set P; (3) Identify the qualified pulsation period of the single-cycle pulse wave set P to obtain the single-cycle pulse wave set P_T with qualified pulsation period, and judge whether the single-cycle pulse wave set P_T with qualified pulsation period satisfies the constraint. If the single-cycle pulse wave set P_T with qualified pulsation period satisfies the constraint, go to step (4); otherwise, mark the waveform quality level of the single-cycle pulse wave set P as level 4, and go to step (6); (4) Identify the qualified main wave of the single-cycle pulse wave set P_T with qualified pulsation period, obtain the single-cycle pulse wave set P_TP with qualified main wave, and judge whether the single-cycle pulse wave set P_TP with qualified main wave meets the constraint. If the single-cycle pulse wave set P_TP with qualified main wave meets the constraint, go to step (5); otherwise, mark the waveform quality level of the single-cycle pulse wave set P_T with qualified pulsation period as level 3, and go to step (6); (5) Identify the qualified dicrotic waves of the single-cycle pulse wave set P_TP with qualified main waves to obtain the single-cycle pulse wave set P_TPD with qualified dicrotic waves, and judge whether the single-cycle pulse wave set P_TPD with qualified dicrotic waves satisfies the constraints. If the single-cycle pulse wave set P_TPD with qualified dicrotic waves satisfies the constraints, mark the waveform quality level of the single-cycle pulse wave set P_TPD with qualified dicrotic waves as the first level, and go to step (6); otherwise, mark the waveform quality level of the single-cycle pulse wave set P_TP with qualified main waves as the second level, and go to step (6); (6) Stable waveform extraction and quality level marking: Merge the single-cycle pulse waves into a stable pulse waveform S in chronological order, and mark the quality level corresponding to the stable pulse waveform S; The step (6) specifically means: if the single-cycle pulse wave is obtained from step (3), the single-cycle pulse waves in the single-cycle pulse wave set P are combined in time sequence into a stable pulse wave waveform, which is recorded as S, and the waveform quality level is level 4; if the single-cycle pulse wave is obtained from step (4), the single-cycle pulse waves in the single-cycle pulse wave set P_T with qualified pulse cycles are combined in time sequence into a stable pulse wave waveform, which is recorded as S, and the waveform quality level is level 3; if the single-cycle pulse wave is obtained from step (5) and the result is If the main wave feature is stable, the single-cycle pulse waves in the single-cycle pulse wave set P_TP with qualified main waves are combined in chronological order into a stable pulse wave waveform, denoted as S, and the waveform quality level is the second level; if the analysis result is obtained from step (5) and the result is that the dicrotic wave feature is stable, the single-cycle pulse waves in the single-cycle pulse wave set P_TPD with qualified dicrotic waves are combined in chronological order into a stable pulse wave waveform, denoted as S, and the waveform quality level is the first level; finally, a stable pulse wave waveform S and its corresponding quality level are obtained.

2. The method for pulse wave quality classification and stable waveform extraction according to claim 1, characterized in that: The step (1) specifically refers to: collecting pulse waves from any collection site on the body, the collection site includes major arterial sites and body surface sites where pulse waves are easily obtained, recording the collected pulse wave signal as an original pulse wave signal, performing a preprocessing operation on the original pulse wave signal to remove high-frequency interference and baseline drift, and recording the obtained waveform as a pending pulse wave waveform S1, wherein the pending pulse wave waveform S1 contains multiple continuous pulsation period signals and / or chaotic signals.

3. The method for pulse wave quality classification and stable waveform extraction according to claim 1, characterized in that: The step (2) specifically includes the following steps: using the differential threshold method to segment the pulse wave waveform S1 to obtain N single-cycle pulse waves, the set of N single-cycle pulse waves is denoted as P, that is, P = {Pi|i = 1,2,…,N}, specifically: (2a) performing a first-order difference operation on the pulse wave waveform S1 to obtain a first-order difference waveform S2; (2b) Divide the first-order differential waveform S2 into m segments on average, extract the maximum value of each segment of the first-order differential waveform, a total of m segment maximum values, namely M1, M2, …Mm, then the threshold of the differential signal is for: ； in, is a weight coefficient between 0 and 1, with a value of 0.

6. is the average value of the maximum values ​​of m segments; (2c) Traverse the first-order difference waveform S2 and extract the values ​​greater than the threshold The positions of the local maximum points of , these points are recorded as the differential maximum undetermined point set Pos, marked as Pos1, Pos2, ..., PosN, Pos(N+1) in sequence, that is, Pos = {Posi|i = 1,2,...,N,N+1}; (2d) The differential threshold method is: for each differential maximum value to be determined point Posi, a window of differential waveform is traversed in reverse order, and the length of the window L is: ； Among them, Fs is the sampling frequency of the signal, t is the weighting coefficient of the pulse cycle, and its value is 0.3; In the i-th window, find the position of the first point where the first-order difference signal is less than 0, recorded as Ai. If not found, set the position of the starting point Ai as the starting point of the current window, Ai = Posi, indicating that Ai is a virtual position point, as a mark for subsequent elimination, these points are recorded as the starting point set A, marked as A1, A2, ..., AN, A(N+1) in sequence, that is, A={Ai|i=1,2, ...,N,N+1}; (2e) The pending pulse wave waveform S1 is divided into independent single-cycle pulse waves according to the starting pending points A1, A2, …, AN, and A(N+1) in sequence, where the single-cycle pulse wave between points A1 and A2 is denoted as P1, the single-cycle pulse wave between points A2 and A3 is denoted as P2, …, the single-cycle pulse wave between points AN and A(N+1) is denoted as PN, and the set of these single-cycle pulse waves is denoted as set P, that is, P = {Pi|i = 1,2,…,N}.

4. The method for pulse wave quality classification and stable waveform extraction according to claim 1, characterized in that: The step (3) specifically includes the following steps: (3a) For each single-cycle pulse wave, calculate the time interval set T of its pulse cycle, which are denoted as T1, T2, ..., T(N-1), TN in sequence: ； Wherein, i is an integer from 1 to N, and Ai is the starting point to be determined; (3b) In the time interval set T, for the i-th time interval Ti, if the starting point Ai is a virtual position point, that is, Ai coincides with the position of the maximum difference point Posi, Ai == Posi, then the time interval Ti is removed from the time interval set T, otherwise the time interval Ti is retained, and the set of all retained time intervals Ti is recorded as the clustering set Set1; (3c) Perform cluster analysis on cluster set Set1: Use K-means algorithm to cluster Set1 into 3 clusters, select the cluster with the largest number of middle clusters as the clustering result, obtain the clustering result of time interval T, record the corresponding single-cycle pulse wave waveform set as P_T = {Pi| i=1,2,…,N_T and N_T <= N}, and count the number N_T; the constraint is: whether N_T satisfies the threshold c×N, if , c is the clustering coefficient, , then go to step (4), otherwise mark the single-cycle pulse wave set P as the fourth-level waveform, and subsequent analysis cannot be performed, and go to step (6).

5. The method for pulse wave quality classification and stable waveform extraction according to claim 1, characterized in that: The step (4) specifically includes the following steps: (4a) For each single-cycle pulse wave P_Ti, the differential threshold method is used to identify the main wave point position in the waveform, and the time interval TPi from the starting point A_T to the main wave point is calculated; the differential threshold method for main wave point position identification means that within a time window after the starting point A_T, the first point where the first-order differential waveform S2 is less than 0 is the main wave point; (4b) All the time intervals TPi of the main wave points are merged into the cluster set Set2, and cluster analysis is performed on the cluster set Set2. The corresponding single-cycle pulse wave waveform set is recorded as P_TP = {Pi| i=1,2,…,N_TP and N_TP <= N_T}, and the number N_TP is counted; the constraint refers to whether N_TP satisfies the threshold c×N_T, where c is the clustering coefficient, c∈(0.5,1): If , then go to step (5), otherwise, mark the single-cycle pulse wave set P_T as the third-level waveform, and cannot perform subsequent analysis, and go to step (6).

6. The method for pulse wave quality classification and stable waveform extraction according to claim 1, characterized in that: The step (5) specifically includes the following steps: (5a) For each single-cycle pulse wave P_TPi, the differential threshold method is used to identify the dicrotic wave point position in the waveform, and the time interval TPNi from the starting point A_TP to the dicrotic wave point is calculated; the differential threshold method for dicrotic wave point position identification means that within a time window after the main wave point, the first point where the first-order differential waveform S2 is greater than 0 is the dicrotic wave point. If there is no such point, the local maximum value point of the first-order differential waveform S2 is taken as the dicrotic wave point; (5b) All the time intervals TPNi of the dicrotic points are merged into the cluster set Set3, and cluster analysis is performed on the cluster set Set3. The corresponding single-cycle pulse wave waveform set is recorded as P_TPD = {Pi| i=1,2,…,N_TPD and N_TPD <=N_TP}, and the number N_TPD is counted. The constraint refers to whether N_TPD satisfies the threshold c×N_TP, where c is the clustering coefficient, c∈(0.5,1): If , then mark the single-cycle pulse wave set P_TPD as the first-level waveform, the dicrotic wave characteristics are stable, and go to step (6); otherwise, mark the single-cycle pulse wave set P_TPD as the second-level waveform, the main wave characteristics are stable, and go to step (6).

Images