A method for calculating blood pressure values for an arterial waveform graph of blood pressure affected by interference

Abstract

The present invention provides a method for calculating blood pressure values for an interfered blood pressure arterial waveform diagram, which can achieve higher-precision blood pressure measurement for interfered blood pressure arterial waveform diagram data without the subjective judgment of technicians. In this application, first, a dynamic electrocardiogram and blood pressure integrated device is used to jointly monitor the electrocardiogram waveform and the blood pressure arterial waveform, which can effectively estimate the pulse wave conduction velocity and further calculate the blood pressure value. Secondly, a convolutional neural network is trained to construct a blood pressure measurement model to express the correlation between the pulse wave conduction velocity and the mean pressure, diastolic pressure, and systolic pressure. The blood pressure measurement model is pre-trained through the PR interval, PP interval, PdH height, and PsH height collected from high-quality arterial waveform diagrams, and then the pre-trained blood pressure measurement model is finely tuned in a targeted manner through the micro-called arterial waveform diagrams in the arterial waveform diagrams to be calculated, and a trained blood pressure measurement model is obtained, which can achieve higher-precision blood pressure measurement for interfered blood pressure arterial waveform diagram data.

Description

Technical Field

[0001] The present invention relates to the technical field of medical artificial intelligence, and specifically to a method for calculating blood pressure values for an interfered blood pressure arterial waveform graph. Background Art

[0002] A dynamic electrocardiogram and blood pressure integrated device is a medical device product that combines dynamic electrocardiogram and dynamic blood pressure. It can not only perform dynamic electrocardiogram but also monitor dynamic blood pressure through the oscillometric method. The oscillometric method is one of the most commonly used non-invasive blood pressure measurement methods. It has good repeatability and small measurement errors, and is the blood pressure measurement method adopted by the vast majority of electronic blood pressure monitors at home and abroad. However, during the blood pressure measurement process using the oscillometric method, the user needs to remain stationary during the air pump inflation and deflation process, which is easily affected by internal and external factors such as jitter, movement status, and arrhythmia, resulting in abnormal longitudinal amplitude of the arterial waveform graph. Once an interfered blood pressure arterial waveform appears, the oscillometric method cannot accurately display the blood pressure at that point, thus forming invalid data. The judgment of blood pressure values for interfered blood pressure arterial waveforms requires the subjective participation of technicians. Summary of the Invention

[0003] In order to solve the problem that the blood pressure values for interfered blood pressure arterial waveform graphs cannot be calculated in the oscillometric blood pressure monitor in the prior art, the present invention provides a method for calculating blood pressure values for interfered blood pressure arterial waveform graphs, which can achieve higher-precision blood pressure measurement for interfered blood pressure arterial waveform graph data without the subjective judgment of technicians.

[0004] The technical solution of the present invention is as follows: A method for calculating blood pressure values for an interfered blood pressure arterial waveform graph, characterized in that it includes the following steps:

[0005] S1: Construct a blood pressure measurement model based on a convolutional neural network;

[0006] S2: Collect N high-quality arterial waveform graphs according to a preset high-quality threshold based on a signal quality assessment method, denoted as: basic arterial waveform graphs; where N≥1000;

[0007] The basic arterial waveform graphs include:

[0008] Static pressure curve: A curve showing the change of static pressure over time;

[0009] Arterial waveform graph: A curve showing the change of dynamic pressure over time;

[0010] Electrocardiogram of channel II: A curve showing the change of the voltage of the electrocardiogram signal in channel II over time;

[0011] S3: Measure the systolic pressure, mean pressure, and diastolic pressure corresponding to each of the basic arterial waveform graphs based on the oscillometric method, denoted as standard prediction results;

[0012] S4: Collect the basic data of each of the basic arterial waveform diagrams to form a basic data sequence, and use the standard prediction result corresponding to the basic data as the label of the basic data sequence;

[0013] The basic data includes: PR interval, PP interval, PdH height, and PsH height;

[0014] The PR interval: the time difference between the peak point of the arterial waveform diagram and the nearest R wave in front of the electrocardiogram;

[0015] The PP interval: the time difference between the peak point of the arterial waveform diagram and the previous peak point;

[0016] The PdH height: the pressure difference between the peak point of the arterial waveform diagram and the next trough point;

[0017] The PsH height: the ordinate pressure value of the static pressure curve corresponding to the abscissa position of the peak point of the arterial waveform diagram;

[0018] S5: Based on the basic data sequence, construct a missing data sequence;

[0019] Preset the number of sequences InNum included in the missing data sequence and the missing peak point value InP corresponding to each missing data sequence;

[0020] Among them, InP is less than the number of peak points included in the basic arterial waveform diagram;

[0021] In the basic data sequence, in each of the basic arterial waveform diagrams, randomly select InP peak points, and set the basic data corresponding to the InP peak points to 0, that is, a missing data sequence is obtained;

[0022] A total of InNum groups of missing data sequences are obtained, and each group includes N missing data sequences;

[0023] S6: Mix and randomly sort the InNum groups of missing data sequences with the basic data sequence to form a training database;

[0024] S7: Pre-train the blood pressure measurement model based on the training database to obtain the pre-trained blood pressure measurement model;

[0025] S8: Obtain the arterial waveform diagram data to be calculated, and based on the signal quality evaluation method, find the micro-called arterial waveform diagrams with better quality according to the preset training quality threshold; Denote the data other than the micro-called arterial waveform diagrams as: interfered arterial waveform diagrams;

[0026] S9: Denote the micro - call arterial waveform diagram as the basic arterial waveform diagram, execute steps S4 - S6, and denote the obtained training database as the micro - call training database;

[0027] S10: Use the micro - call training database to fine - tune the pre - trained blood pressure measurement model to obtain the trained blood pressure measurement model;

[0028] S11: For each of the disturbed arterial waveform diagrams, specify the abscissa of the highest peak and the abscissa of the reliable peak, and send the two abscissas and the disturbed arterial waveform Figure 1 into the trained blood pressure measurement model, and the blood pressure measurement model outputs the prediction result corresponding to each disturbed arterial waveform diagram;

[0029] The prediction result includes the systolic blood pressure, mean blood pressure, and diastolic blood pressure corresponding to the disturbed arterial waveform diagram.

[0030] Its further feature is that:

[0031] The convolutional neural network includes, connected in sequence: an input layer, a convolutional layer, a pooling layer, and 3 fully - connected layers;

[0032] Among them, the number of channels of the input layer is 4, the number of channels of the convolutional layer and the pooling layer is 8, and the number of channels of the fully - connected layer is: 1;

[0033] The acquisition method of the basic data sequence includes:

[0034] a1: Find the highest peak point of the basic arterial waveform diagram and mark it as P0;

[0035] a2: Taking P0 as the center, find 10 peak points respectively before and after;

[0036] If the number of peak points on the left or right side of P0 is less than 10, execute step a4;

[0037] Otherwise, execute step a3;

[0038] a3: Denote the peak points before and after P0 as: P -10 , P -9 , P -8 ...P -1 and P1, P2...P 10 ;

[0039] Calculate the PR interval, PP interval, PdH height, and PsH height corresponding to the peak points, which are the basic data; execute step a5;

[0040] a4: Calculate all the basic data corresponding to all the existing peak points, and at the same time set the basic data corresponding to the missing peak points to 0 to obtain 21 groups of data, which are the basic data corresponding to the basic arterial waveform diagram; execute step a5;

[0041] a5: Calculate the basic data of all 21 peak points, and respectively form a PR interval sequence, a PP interval sequence, a PdH height sequence, and a PsH height sequence; record the PR interval sequence, the PP interval sequence, the PdH height sequence, and the PsH height sequence as the basic data sequence;

[0042] The signal quality assessment method includes the following steps:

[0043] b1: Obtain the arterial waveform diagram to be analyzed;

[0044] b2: Calculate the scatter degree Sp and the peak amplitude perturbation rate PError corresponding to the obtained arterial waveform diagram to be analyzed;

[0045] b3: Judge whether the arterial waveform diagram to be analyzed is spurious interference data according to the scatter degree Sp and the peak amplitude perturbation rate PError;

[0046] If it is spurious interference data, the quality of the arterial waveform diagram to be analyzed is set as data with poor indicators;

[0047] Otherwise, execute step b4;

[0048] b4: Calculate the treatment evaluation parameter Qua of the arterial waveform diagram to be analyzed:

[0049] Qua = SP * α + PError * β;

[0050] Where α is the scatter degree weight coefficient and β is the perturbation rate weight coefficient;

[0051] b5: Compare Qua with the preset quality judgment threshold QH;

[0052] When Qua is less than QH, the quality of the data corresponding to the arterial waveform diagram to be analyzed is judged as high-quality data, otherwise, the data corresponding to the arterial waveform diagram to be analyzed is judged as data with poor indicators;

[0053] In step S5, InNum is set to 4; the corresponding InP values include: InP1 = 10, InP2 = 15, InP3 = 18, InP4 = 20;

[0054] In step b3, the method for judging the spurious interference data includes the following steps:

[0055] bS1: Obtain the arterial waveform diagram to be analyzed;

[0056] The horizontal axis of the arterial waveform diagram to be analyzed represents the position of the pulse beat caused by the heartbeat, and the vertical axis represents the pressure value collected by the pressure sensor of the dynamic blood pressure monitor;

[0057] bS2: Mark the peak points and valley points on the arterial waveform diagram to be analyzed;

[0058] Based on the arterial waveform diagram to be analyzed, find all the peak points and valley points; starting from the first peak point, pair it with the adjacent valley point;

[0059] Assume that the arterial waveform diagram to be analyzed includes N pairs of peak points and valley points, then the peak points and valley points included in the arterial waveform diagram to be analyzed are expressed as:

[0060] (Peak1, Valley1), (Peak2, Valley2),..., (PeakN, ValleyN);

[0061] The abscissas of the peak points Peak1 - PeakN form a sequence x1, x2,... x N ;

[0062] bS3: Obtain the ordinates of each pair of peak points and valley points, and calculate the difference between the ordinates of each pair of peak points and valley points;

[0063] Then a sequence of the absolute values of the differences between the ordinates of each pair of peak points and valley points is obtained:

[0064] Denoted as the difference sequence: h1, h2,...., h N ;

[0065] bS4: Draw a Lorenz scatter plot, specifically including the following steps:

[0066] ba1: Calculate the difference sequence of the sequence x1, x2,... x N , and record it as △1 = x2 - x1, △2 = x3 - x2,..., Δ N-1 = x N - x N-1 ;

[0067] △ i represents the pulse interval or heartbeat interval, and i is a natural number, i ∈ (1, N - 1);

[0068] ba2: Combine adjacent △ i into pairs to form (△ i , △ i+1 );

[0069] ba3: Construct coordinate axes, and the units of both the horizontal coordinate and the vertical coordinate are ms;

[0070] For each group of (△i , △ i+1 ) the △ in i is the abscissa, and △ i+1 is the ordinate. Plot (△ i , △ i+1 ) into the coordinate axes to obtain a Lorenz scatter plot;

[0071] bS5: The scatter degree Sp of the arterial waveform diagram to be analyzed specifically includes the following steps:

[0072] bb1: Construct a mapping grid diagram;

[0073] Set the heart beat interval threshold and the reasonable interval threshold;

[0074] The heart beat interval threshold represents the maximum value of the reasonable heart beat interval data;

[0075] The reasonable interval threshold means that when the difference between two heart beat intervals of the same data source is less than the reasonable interval threshold, these two heart beat intervals are regarded as the same data during calculation;

[0076] The mapping grid includes: a mapping coordinate system and a grid;

[0077] Both the horizontal axis and the vertical axis of the mapping coordinate system are set to the heart beat interval. The starting points of the horizontal axis and the vertical axis are 0, and the maximum value is the heart beat interval threshold;

[0078] Build the grid between the horizontal axis and the vertical axis with the interval allowable value as the unit;

[0079] bb2: Take out each group (△ i , △ i+1 ), and judge the relationship between △ i and the heart beat interval threshold;

[0080] If △ i > the heart beat interval threshold, then (△ i , △ i+1 ) does not participate in the calculation;

[0081] Otherwise, judge the relationship between △ i+1 and the heart beat interval threshold;

[0082] If △ i+1 > the heart beat interval threshold, then (△ i , △ i+1 ) does not participate in the calculation;

[0083] Otherwise, execute step bb3;

[0084] bb3: Take (△ i , △ i+1 ) with △i is used as the abscissa, and △ i+1 is mapped to the grid of the mapping grid graph as the ordinate;

[0085] bb4: After all (△ i , △ i+1 ) participate in the calculation, count the number of grids successfully mapped in the mapping grid graph;

[0086] Suppose there are n grids successfully mapped in the mapping grid graph, and calculate the scatter degree Sp of the arterial waveform graph to be analyzed;

[0087] Sp = n / (N - 1);

[0088] bS6: Find the effective area of the arterial waveform graph to be analyzed, which specifically includes the following steps:

[0089] bc1: In the difference sequence: h1, h2,...., h N , find the maximum value and denote it as: h Max ;

[0090] bc2: Initialize the serial number j = 1,

[0091] bc3: Compare h j , h j+1 , h j+2 with C1 * h Max respectively;

[0092] If there exist h j , h j+1 , h j+2 all greater than C1 * h Max , then set h j as Pstart and execute step bc6;

[0093] Otherwise, execute step bc4;

[0094] bc4: Judge whether j is greater than Max - 2;

[0095] If j > Max - 2, then judge the arterial waveform graph data corresponding to the difference sequence as false difference interference data;

[0096] Otherwise, execute step bc5;

[0097] bc5: j = j + 1, and loop to execute steps bc3 - bc4;

[0098] bc6: Judge the relationship between j and N;

[0099] If j > N - 2, the arterial waveform data to be analyzed corresponding to the difference sequence is determined as spurious interference data;

[0100] Otherwise, execute step bc7;

[0101] bc7: j = j + 1;

[0102] bc8: Compare h j , h j+1 , h j+2 with C1 * h Max respectively;

[0103] where C1 is the effective peak - valley difference threshold;

[0104] If there exists h j , h j+1 , h j+2 all greater than C1 * h Max , then set h j+2 as Pend and execute step bc10;

[0105] Otherwise, execute step bc9;

[0106] bc9: Judge whether j is greater than N - 2;

[0107] If j > N - 2, the arterial waveform data to be analyzed corresponding to the difference sequence is determined as interference signal;

[0108] Otherwise, j = j + 1, and loop to execute steps bc8 - bc9;

[0109] bc10: In the difference sequence: h1, h2,...., h N the data points between Pstart and Pend are considered as the effective data region of the arterial waveform to be analyzed;

[0110] bS7: Perform data fitting on the effective data region, specifically including the following steps:

[0111] bd1: Take out the data including Pstart to Pend in the difference sequence, denoted as the effective data sequence,

[0112] The effective data sequence is: h Pstart , h Pstart+1 ....h max..... h Pend-1 , h Pend ;

[0113] bd2: Take out the data on the left side of h max. : h Pstart , h Pstart+1 ....hmax-1 , centered on h max. , perform mirroring to obtain the left - hand fitting data sequence: h Pstart , h Pstart+1 ....h max-1 , h max , h max-1 ....h Pstart+1 , h Pstart ;

[0114] The abscissas corresponding to the left - hand fitting data sequence are: x Pstart , x Pstart+1 ....x max-1 , x max , x max- 1....x Pstart ;

[0115] bd3: Extract the data on the right - hand side of h max. : h max+1..... h Pend-1 , h Pend , centered on h max. , perform mirroring to obtain the right - hand fitting data sequence: h Pend , h Pend-1 ....h max+1 , h max , h max+1..... h Pend-1 , h Pend ;

[0116] The abscissas corresponding to the right - hand fitting data sequence are: x Pend , x Pend-1. ...x max+1 , x max , x max+ 1....x Pend-1 , x Pend ;

[0117] bS8: Calculate the peak amplitude perturbation rate PError corresponding to the arterial waveform diagram to be analyzed;

[0118] be1: Calculate the number of left - hand interference points ErrorLeft;

[0119] Using the left - hand fitting data sequence and the abscissas corresponding to the left - hand fitting data sequence as the ordinates and abscissas of the data points, use the least - squares method to obtain the quadratic fitting curve h = ax 2 + bx + c;

[0120] Count the number of points whose distance between the ordinate of the left - hand valid data points and the fitting curve is greater than C2 * h Max , and record it as the number of left - hand interference points ErrorLeft;

[0121] where C2 is the interference point judgment threshold;

[0122] be2: Calculate the number of interference points on the right, ErrorRight;

[0123] Using the right - hand fitting data sequence and the corresponding abscissa of the right - hand fitting data sequence as the ordinate and abscissa of the data points, the quadratic fitting curve h = ax 2 + bx + c is obtained by using the least - squares method;

[0124] Count the number of points where the distance between the ordinate of the valid data points on the right and the fitting curve is greater than C2 * h Max and record it as the number of interference points on the right, ErrorRight;

[0125] be3: Calculate the peak amplitude perturbation rate PError;

[0126] PError=(ErrorRight + ErrorLeft) / (Pend - Pstart + 1);

[0127] bS9: Judge the nature of the arterial waveform diagram to be analyzed;

[0128] bf1: Compare the scatter degree Sp of the arterial waveform diagram to be analyzed with the preset scatter degree threshold C3;

[0129] bf2: Compare the peak amplitude perturbation rate PError with the preset perturbation rate threshold C4;

[0130] bf3: When the following two conditions are simultaneously met, the arterial waveform diagram to be analyzed is judged as spurious interference data;

[0131] Sp > C3 and PError > C4;

[0132] The heart - beat interval threshold is set to 2000 ms; the reasonable interval threshold is set to 100 ms;

[0133] The effective peak - valley difference threshold C1 is 1 / 3; the interference point judgment threshold C2 is 1 / 8; the scatter degree threshold C3 is 30%; the perturbation rate threshold C4 is 20%;

[0134] The scatter degree weight coefficient α takes a value of 25%; the perturbation rate weight coefficient β takes a value of 75%; the quality judgment threshold QH takes a value of 20%.

[0135] The present invention provides a method for calculating blood pressure values for disturbed blood pressure arterial waveforms. First, the dynamic ECG and blood pressure two-in-one device is used to monitor both the ECG waveform and the blood pressure arterial waveform in a linked manner, which can effectively estimate the pulse wave conduction velocity and further calculate the blood pressure value. Secondly, a blood pressure measurement model is constructed by training a convolutional neural network to express the correlation between the pulse wave conduction velocity and the mean pressure, diastolic pressure, and systolic pressure. The principle of the correlation between the pulse wave conduction velocity and the mean pressure, diastolic pressure, and systolic pressure is used to pre-train the blood pressure measurement model through the PR interval, PP interval, PdH height, and PsH height collected from the high-quality arterial waveform. The number of layers of the convolutional neural network of the blood pressure measurement model, the number of channels in each layer, and the size of the convolution kernel are determined. Then, the pre-trained blood pressure measurement model is fine-tuned by fine-tuning the arterial waveform data to be calculated, and the weight coefficients in the connections in the model are adaptively adjusted to obtain a trained blood pressure measurement model. The model does not require the subjective judgment of technicians during the entire process, and can achieve higher-precision blood pressure measurement for disturbed blood pressure arterial waveform data. Based on the technical solution of the present application, the oscillometric method is used to measure accurate blood pressure values when the arterial waveform signal quality is good, providing sufficient standard data for the blood pressure measurement model constructed based on the convolutional neural network to train, learn and express the correlation between pulse wave conduction velocity and mean pressure, diastolic pressure and systolic pressure. Not only can the blood pressure value be determined by the pulse wave velocity method when the oscillometric method fails to calculate the blood pressure due to abnormal longitudinal amplitude of the arterial waveform, thereby avoiding the defects of the oscillometric method in measuring blood pressure, but also solving the problem in the prior art that the pulse wave velocity method requires correction by other blood pressure measurement methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0136] Figure 1 This is an embodiment of a blood pressure measurement model;

[0137] Figure 2 is an embodiment of a basal arterial waveform diagram;

[0138] Figure 3 Example 2 of the basic arterial waveform diagram;

[0139] Figure 4 This is a flow chart of the method for identifying artifacts in a blood pressure arterial waveform signal in this application;

[0140] Figure 5 is an embodiment of a blood pressure arterial waveform diagram;

[0141] Figure 6 An embodiment of peaks and valleys of a blood pressure arterial waveform;

[0142] Figure 7 An embodiment of a mapped grid graph;

[0143] Figure 8An embodiment of the effective region of the blood pressure arterial waveform diagram;

[0144] Figure 9 An embodiment of the quadratic fitting curve;

[0145] Figure 10 An embodiment calculated based on the method of the present application. Specific implementation mode

[0146] The present invention includes a blood pressure value calculation method for a disturbed blood pressure arterial waveform diagram, which includes the following steps.

[0147] S1: Construct a blood pressure measurement model based on a convolutional neural network;

[0148] In the technical solution of the present application, a classical convolutional neural network is used to construct a blood pressure measurement model. The convolutional neural network includes, in sequence: an input layer, a convolutional layer, a pooling layer, and three fully connected layers; the number of layers of the convolutional neural network, the number of channels in each layer, and the size of the convolutional kernel are all hyperparameters, which are adjusted according to the training effect.

[0149] In order to train the blood pressure measurement model, it is necessary to first construct training data. In the technical solution of the present application, the training of the blood pressure measurement model includes two stages: pre-training and fine-tuning training; in the pre-training stage, the hyperparameters of the blood pressure measurement model are trained using the full range of historical data to determine the structure of the neural network; in the fine-tuning training stage, the data with better treatment in the data source to be calculated is used to fine-tune the parameters in the model through online learning: the link weight coefficient.

[0150] In order to implement pre-training, it is necessary to pre-construct a pre-learning database.

[0151] S2: According to a preset high-quality threshold, collect N high-quality arterial waveform diagrams based on a signal quality evaluation method, denoted as: basic arterial waveform diagrams; where N≥1000 to ensure that there is enough training data to train the model and ensure the accuracy of the model; in this embodiment, N is taken as 1000.

[0152] The basic arterial waveform diagrams are based on the data collected by a dynamic electrocardiogram and blood pressure integrated device. The original arterial waveform, static pressure curve data output by the pressure sensor in the dynamic electrocardiogram and blood pressure integrated device, and the electrocardiogram waveform output by the electrocardiogram electrode patch. The systolic pressure, mean pressure, and diastolic pressure results measured by the oscillometric method are used as the gold standard and are the standard blood pressure values of this group of data. As Figure 2 shown, the basic arterial waveform diagrams include:

[0153] Static pressure curve: The curve of the static pressure changing with time;

[0154] Arterial waveform diagram: The curve of the dynamic pressure changing with time;

[0155] ECG waveform of channel II: The curve of the voltage of the ECG signal in channel II changing with time.

[0156] S3: Based on the oscillometric method, measure the systolic pressure, mean pressure, and diastolic pressure corresponding to each basic arterial waveform diagram, and record them as the standard prediction results; input them into the model together with the training data in the form of labels to train the model.

[0157] S4: Process the data of the basic arterial waveform diagram.

[0158] Collect the basic data of each basic arterial waveform diagram respectively to form a basic data sequence, and use the standard prediction result corresponding to the basic data as the label of the basic data sequence;

[0159] ... The R-wave position is obtained by the QRS wave localization algorithm, such as Figure 3 the position of the fork mark shown in the figure. Find the abscissa and ordinate of the peak point and valley point in the arterial waveform diagram, and then the data such as the peak-to-valley difference can be calculated, and further the basic data value can be calculated;

[0160] The basic data includes: PR interval, PP interval, PdH height, and PsH height;

[0161] PR interval: The time difference between the peak point of the arterial waveform diagram and the nearest R wave in front of the electrocardiogram;

[0162] PP interval: The time difference between the peak point of the arterial waveform diagram and the previous peak point;

[0163] PdH height: The pressure difference between the peak point of the arterial waveform diagram and the next valley point;

[0164] PsH height: The ordinate pressure value of the static pressure curve corresponding to the abscissa position of the peak point of the arterial waveform diagram.

[0165] The acquisition method of the basic data sequence specifically includes the following steps:

[0166] a1: Find the highest peak point of the basic arterial waveform diagram and mark it as P0;

[0167] a2: Search for 10 peak points respectively before and after P0;

[0168] If the number of peak points on the left or right side of P0 is less than 10, execute step a4;

[0169] Otherwise, execute step a3;

[0170] a3: Denote the peak points before and after P0 as: P -10 , P -9 , P -8 ... P -1 and P1, P2... P 10 ;

[0171] Calculate the PR interval, PP interval, PdH height, and PsH height corresponding to the peak points, which are the basic data; execute step a5;

[0172] PR interval: PR -10 , PR -9 , PR -8 ...PR -1 , PR0, PR1, PR2...PR 10

[0173] PP interval: PP -10 , PP -9 , PP -8 ...PP -1 , PP0, PP1, PP2...PP 10

[0174] PdH interval: PdH -10 , PdH -9 , PdH -8 ...PdH -1 , PdH0, PdH1, PdH2...PdH 10

[0175] PsH interval: PsH -10 , PsH -9 , PsH -8 ...PsH -1 , PsH0, PsH1, PsH2...PsH 10 ;

[0176] a4: Calculate all the basic data corresponding to all the existing peak points, and at the same time set the basic data corresponding to the missing peak points to 0, obtaining twenty - one groups of data, which are the basic data corresponding to the basic arterial waveform diagram; execute step a5;

[0177] a5: Calculate the basic data of all 21 peak points, respectively forming a PR - interval sequence, a PP - interval sequence, a PdH - height sequence, and a PsH - height sequence; denote the PR - interval sequence, the PP - interval sequence, the PdH - height sequence, and the PsH - height sequence as the basic - data sequence.

[0178] In the disturbed blood - pressure arterial waveform diagram, since the vertical height of the arterial waveform diagram is disturbed, the PdH - height and PsH - height data are not reliable. At this time, the correct blood - pressure value needs to be obtained relying on the PR - interval and PP - interval. In order to construct training data for calculating the blood - pressure value only through the PR - interval and PP - interval (related to the pulse - wave conduction velocity) when the PdH - height and PsH - height data are disturbed or incomplete. According to the following method, incomplete data are constructed by a random method and together with the complete data to form a database for training.

[0179] like Figure 3 As shown, in Figure 2 A schematic diagram of the absolute value of the height difference between the peak and the trough is added below the basic arterial waveform diagram. In the figure, the point marked PdH0 is the absolute value of the height difference between the peak point P0 and its adjacent peak and trough; PR0 is the absolute value of the height difference between P0 and P -1 The time difference (horizontal coordinate) is PP0, and the time difference between P0 and the rear valley point (located between P0 and P1). In the basic arterial waveform diagram, PsH0 is the vertical coordinate (pressure value) of the P0 horizontal coordinate (time) on the static pressure curve. Among them, referring to the basic arterial waveform diagram, it can be seen that there is a time difference between the horizontal coordinate position corresponding to each R wave in the electrocardiogram of channel II and the corresponding horizontal coordinate on the arterial waveform diagram. For the same data source, this time difference is basically fixed. In this application, based on this fixed time difference, the data relationship between the PR interval, PP interval, PdH height, and PsH height is used to determine the missing blood pressure height value of the interfered blood pressure arterial wave.

[0180] S5: Based on the basic data sequence, construct the incomplete data sequence;

[0181] The number of sequences InNum included in the preset incomplete data sequence and the incomplete peak value InP corresponding to each incomplete data sequence;

[0182] Wherein, InP is less than the number of peak points included in the basic arterial waveform;

[0183] In each basic arterial waveform graph in the basic data sequence, randomly select InP peak points, and set the basic data corresponding to the InP peak points to 0, thus obtaining an incomplete data sequence;

[0184] A total of InNum groups of incomplete data sequences are obtained, each group including N incomplete data sequences.

[0185] In this embodiment, InNum is set to 4; the corresponding InP values include: InP1=10, InP2=15, InP3=18, InP4=20;

[0186] Take out N=1000 groups of data, and for each group of data, randomly select 10 peak points out of the 21 peak points, and set the PR interval, PP interval, PdH height, and PsH height of the corresponding peak points to 0, thus forming N=1000 groups of incomplete data.

[0187] Take out N=1000 groups of data, and for each group of data, randomly select 15 peak points out of the 21 peak points, and set the PR interval, PP interval, PdH height, and PsH height of the corresponding peak points to 0, forming N=1000 groups of incomplete data.

[0188] Take out N = 1000 groups of data. For each group of data, randomly select 18 out of 21 peak points, and set the PR interval, PP interval, PdH height, and PsH height of the corresponding peak points to 0 to form N = 1000 groups of incomplete data.

[0189] Take out N = 1000 groups of data. For each group of data, randomly select 20 out of 21 peak points, and set the PR interval, PP interval, PdH height, and PsH height of the corresponding peak points to 0 to form N = 1000 groups of incomplete data.

[0190] Mix and randomly sort the above 4000 groups of incomplete data and the original 1000 groups of basic data sequences to form a pre-training database.

[0191] S6: Mix and randomly sort the InNum groups of incomplete data sequences and the basic data sequences to form a training database.

[0192] S7: Pre-train the blood pressure measurement model based on the training database to obtain the pre-trained blood pressure measurement model.

[0193] As Figure 1 The embodiment of the blood pressure measurement model shown is a model obtained after training for a blood pressure monitor of the CB-2302-A model with 12 leads and two independent pacing channels. In this application, a classic convolutional neural network is used to train the database. The number of layers of the convolutional neural network, the number of channels in each layer, and the size of the convolutional kernel are all hyperparameters and are adjusted according to the training effect. In this embodiment, the number of layers of the convolutional neural network, the number of channels in each layer, and the size of the convolutional kernel are: input layer 4x21, convolutional layer 8x21, pooling layer 8x21, the first and second fully connected layers 1x10, and the last fully connected layer (output layer) 1x3. That is, the input of the blood pressure measurement model is: PR interval, PP interval, PdH height, PsH height, and the output is: systolic blood pressure, mean blood pressure, and diastolic blood pressure corresponding to the arterial waveform diagram.

[0194] S8: Obtain the data of the arterial waveform diagram to be calculated. According to the preset quality threshold for training, based on the signal quality evaluation method, find the micro-called arterial waveform diagram with better quality; record the data other than the micro-called arterial waveform diagram as: the disturbed arterial waveform diagram.

[0195] Conduct a 24-hour ambulatory blood pressure monitoring to form a case. A case generally contains 50 - 100 groups of blood pressure measurement records. Select the cases with signal quality greater than a certain threshold and retrain them according to the above training method.

[0196] S9: Record the micro-called arterial waveform diagram as: the basic arterial waveform diagram, and execute steps S4 - S6. Record the obtained training database as: the micro-called training database.

[0197] S10: Fine-tune the pre-trained blood pressure measurement model using the micro-call training database to obtain a trained blood pressure measurement model.

[0198] After pre-training, the obtained pre-trained model is already a basically formed blood pressure measurement model. However, the pulse wave velocity has a good correlation with the mean blood pressure, diastolic blood pressure, and systolic blood pressure. Currently, this correlation cannot be accurately expressed by theory. In the prior art, other blood pressure measurement methods are usually used for calibration, and the calibration parameters for each person are different. Therefore, in this application, high-quality data in the data source of the arterial waveform diagram to be calculated is used to retrain the model and adjust the parameters, mainly training the weight coefficients in the connection, so that the model becomes a targeted model, thereby ensuring the accuracy of the prediction results.

[0199] S11: For each disturbed arterial waveform diagram, specify the abscissa of the highest peak and the abscissa of the reliable peak, and input the two abscissas and the disturbed arterial waveform Figure 1 into the trained blood pressure measurement model, and the blood pressure measurement model outputs the prediction result corresponding to each disturbed arterial waveform diagram;

[0200] The prediction results include: the systolic blood pressure, mean blood pressure, and diastolic blood pressure corresponding to the disturbed arterial waveform diagram.

[0201] Because only the information of some peaks in the disturbed arterial waveform diagram is reliable, and the ordinate pressure values in the disturbed part are inaccurate. In this application, as long as the abscissa of the highest peak and the abscissa of the reliable peak are specified and input into the blood pressure measurement model, the model will calculate and fill in the corresponding PR interval / PP interval, determine the number of each wave peak through the position of the electrocardiogram R peak, and then calculate the corresponding systolic blood pressure, mean blood pressure, and diastolic blood pressure.

[0202] The specific calculation process is as Figure 10 shown. Specify the abscissa of the highest pulse peak as X (the peak point marked as h max in the figure), and calculate the PP, PR, PsH, and PdH of the h max peak, and record them as: PR0, PP0, PdH0, PsH0;

[0203] Specify the abscissa of a reliable pulse peak in front of the highest peak as X front , and the number of electrocardiogram QRS waves between the abscissa X front of this pulse peak and the abscissa X of the highest pulse peak is n = 6. Calculate the PP, PR, PsH, and PdH of this peak, and record them as: PR -n , PP -n , PdH -n , PsH -n ;

[0204] Specify the horizontal coordinate of a reliable pulse peak in front of the highest peak as X back , the pulse peak abscissa X back The number of ECG QRS peaks between the horizontal coordinate X and the highest pulse peak is n = 1. Calculate the PP, PR, PsH, and PdH of this peak, which are recorded as: PR n PP n 、PdH n 、PsH n ;

[0205] The input values of the remaining pulse peaks are all set to 0 and input into the trained blood pressure measurement model to calculate the blood pressure values corresponding to all the horizontal coordinates of the disturbed arterial waveform.

[0206] In step S2, N high-quality arterial waveforms need to be collected based on a preset high-quality threshold and a signal quality assessment method. In step S8, a fine-tuned arterial waveform with good quality needs to be found based on a preset training quality threshold and a signal quality assessment method.

[0207] The signal quality assessment method includes the following steps:

[0208] b1: Get the arterial waveform to be analyzed;

[0209] b2: Calculate the scatter Sp and peak amplitude disturbance rate PError corresponding to the arterial waveform to be analyzed;

[0210] b3: Determine whether the arterial waveform to be analyzed is artifact interference data based on the scatter Sp and peak amplitude disturbance rate PError;

[0211] If it is artifact interference data, the quality of the arterial waveform to be analyzed is set as poor indicator data;

[0212] Otherwise, execute step g4;

[0213] b4: Calculate the treatment evaluation parameter Qua of the arterial waveform to be analyzed:

[0214] Qua=SP*α+PError*β;

[0215] Among them, α is the weight coefficient of the scatter degree, and β is the weight coefficient of the disturbance rate;

[0216] b5: Compare Qua with the preset quality judgment threshold QH;

[0217] When Qua is less than QH, the quality of the data corresponding to the arterial waveform to be analyzed is judged to be good quality data; otherwise, the quality of the data corresponding to the arterial waveform to be analyzed is judged to be poor quality data.

[0218] The scatter degree weight coefficient α takes a value of 25%; the perturbation rate weight coefficient β takes a value of 75%; even if the scatter degree of the arterial waveform diagram to be analyzed is large, when the peak amplitude perturbation rate PError of the arterial waveform diagram is small, that is, the data points are relatively consistent with the fitting curve, it indicates that the waveform of the analyzed arterial waveform diagram is relatively complete, and the data of the arterial waveform diagram to be analyzed can reflect the heartbeat trend of the data source and can also be used as valid data for calculation.

[0219] In this embodiment, in step S2, a large number of arterial waveform diagrams are used as the waveform diagrams to be analyzed, and a preset high-quality threshold is used as the quality judgment threshold QH. Through steps b1 to b5, N high-quality arterial waveform diagrams are found; in step S8, the data of the arterial waveform diagram to be calculated is used as the waveform diagram to be analyzed, and a preset training quality threshold is used as the quality judgment threshold QH. Through steps b1 to b5, a micro-called arterial waveform diagram with better quality is found.

[0220] Among them, the high-quality threshold in step S2 and the training quality threshold in step S8 can be the same or different. The specific value is set according to actual needs. In this embodiment, the two are set to the same value, both taking a value of 20%. That is, the quality judgment threshold QH takes a value of 20%. QH taking a value of 100% represents completely scattered rhythm (the maximum scatter degree Sp) and all points deviating from the fitting curve (the maximum peak amplitude perturbation rate PError). Generally, the normal rhythm is relatively neat, the scatter degree sp < 20%, and most data points (at least 80% of the data points) can fall on the fitting line. Therefore, the peak amplitude perturbation rate PError < 20%. The value of QH represents the combination of Sp and PError. Therefore, QH taking a value less than 20% represents better signal quality and is not a false difference. In practical applications, users can dynamically change the QH value threshold according to needs to screen data with better signal quality. To ensure that among about 50 groups of blood pressure measurement records in a patient's case, at least 30 groups are used as normal data and the others are used as false differences.

[0221] In step b3, the method for judging false difference interference data is as Figure 4 shown and it includes the following steps.

[0222] bS1: Obtain the arterial waveform diagram to be analyzed;

[0223] As Figure 5As shown, it is an embodiment of a blood pressure arterial waveform diagram as the arterial waveform diagram to be analyzed. The horizontal axis of the arterial waveform diagram to be analyzed is the position of the pulse beat caused by the heartbeat, and the vertical axis is the pressure value collected by the pressure sensor of the dynamic sphygmomanometer. In actual work, the horizontal axis of the arterial waveform diagram to be analyzed is the position of the pulse beat caused by the heartbeat, represented by the number of sampling points or sampling time; the sampling rate of the arterial waveform diagram to be analyzed is SAMPLE, and the sampling rate SAMPLE defines the number of samples extracted from the continuous signal per second and composed into a discrete signal, with the unit of hertz (Hz).

[0224] bS2: Mark the peak points and valley points of the arterial waveform diagram to be analyzed;

[0225] Based on the arterial waveform diagram to be analyzed, find all the peak points and valley points; starting from the first peak point, pair it with the adjacent valley point;

[0226] Suppose: The arterial waveform diagram to be analyzed includes N pairs of peak points and valley points, then the peak points and valley points included in the arterial waveform diagram to be analyzed are expressed as:

[0227] (Peak1, Valley1), (Peak2, Valley2),..., (PeakN, ValleyN);

[0228] The abscissas of the peak points Peak1 - PeakN form a sequence x1, x2,... x N .

[0229] Regarding the method for confirming the peak points and valley points of the arterial waveform diagram, based on the existing technology, methods such as the oscillometric blood pressure measurement algorithm can be used to complete it. The obtained (Peak1, Valley1) is as Figure 6 shown.

[0230] bS3: Obtain the ordinates of each pair of peak points and valley points, and calculate the difference between the ordinates of each pair of peak points and valley points;

[0231] Then the sequence of the absolute values of the differences between the ordinates of each pair of peak points and valley points is obtained:

[0232] Recorded as the difference sequence: h1, h2,...., h N .

[0233] bS4: Draw a Lorenz scatter plot;

[0234] ba1: Calculate the difference sequence of the sequence x1, x2,... x N and record it as △1 = x2 - x1, △2 = x3 - x2,... △ N-1 = x N - x N-1 ;

[0235] △ iIt represents the pulse interval or heart beat interval, where i is a natural number and i ∈ (1, N - 1);

[0236] ba2: Combine adjacent △ i in pairs to form (Δ i , Δ i+1 );

[0237] ba3: Construct a coordinate axis, where the units of both the horizontal and vertical coordinates are ms;

[0238] Using each △ i in each group of (Δ i+1 , Δ i ) as the horizontal coordinate and △ i+1 as the vertical coordinate, plot (Δ i , Δ i+1 ) into the coordinate axis to obtain the Lorenz scatter plot;

[0239] The degree of scatter in the distribution of the Lorenz scatter plot is used to determine whether it belongs to severe arrhythmia or interference. In this application, the degree of scatter in the distribution of the Lorenz scatter plot is used as one of the bases for evaluating the signal quality.

[0240] bS5: The scatter degree Sp of the arterial waveform diagram to be analyzed, specifically including the following steps:

[0241] bb1: Construct a mapping grid diagram;

[0242] Set the heart beat interval threshold and the reasonable interval threshold;

[0243] The heart beat interval threshold represents the maximum value of reasonable heart beat interval data;

[0244] The reasonable interval threshold means that when the difference between two heart beat intervals from the same data source is less than the reasonable interval threshold, these two heart beat intervals are regarded as the same data during calculation;

[0245] The mapping grid includes: a mapping coordinate system and a grid;

[0246] Both the horizontal and vertical axes of the mapping coordinate system are set to the heart beat interval, with the starting points of the horizontal and vertical axes being 0 and the maximum value being the heart beat interval threshold;

[0247] Between the horizontal and vertical axes, construct a grid with the interval allowance value as the unit.

[0248] Because, in actual work, the horizontal axis of the arterial waveform diagram to be analyzed is the position of the pulse beat caused by the heartbeat, which can be represented by the number of sampling points or the sampling time. Therefore, the units of the heartbeat interval threshold and the reasonable interval threshold can be the sampling time or the number of sampling points; the specific values of the heartbeat interval threshold and the reasonable interval threshold are set according to research needs. In this embodiment, when analyzing most arterial waveform diagrams, the maximum value of the corresponding human heartbeat interval is 2000s. Therefore, the heartbeat interval threshold is set to 2000ms; in general research, when the difference between two heartbeat intervals is not greater than 100ms, they are processed as the same trend data and can be mapped in the same grid. In this embodiment, the reasonable interval threshold is set to 100ms, and the specific mapping grid diagram is as shown in Figure 7 shown; Figure 7 The part between 1000ms and 2000ms in it is omitted for representation.

[0249] bb2: Take out each group (Δ i , Δ i+1 ), and judge the relationship between △ i and the heartbeat interval threshold;

[0250] If △ i > the heartbeat interval threshold, then (Δ i , Δ i+1 ) does not participate in the calculation;

[0251] Otherwise, judge the relationship between △ i+1 and the heartbeat interval threshold;

[0252] If △ i+1 > the heartbeat interval threshold, then (Δ i , Δ i+1 ) does not participate in the calculation;

[0253] Otherwise, execute step bb3;

[0254] bb3: Map (Δ i , Δ i+1 ) to the grid of the mapping grid diagram with △ i as the abscissa and △ i+1 as the ordinate;

[0255] bb4: When all (Δ i , Δ i+1 ) have participated in the calculation, count the number of grids successfully mapped in the mapping grid diagram;

[0256] Suppose there are n grids successfully mapped in the mapping grid diagram, and calculate the scatter degree Sp of the arterial waveform diagram to be analyzed;

[0257] Sp = n / (N - 1).

[0258] In specific implementation, when the horizontal and vertical axis coordinate units of the constructed mapping grid graph are the same as the horizontal axis unit of the arterial waveform graph to be analyzed, the Lorenz scatter plot corresponding to the arterial waveform graph to be analyzed can be directly mapped into the mapping grid graph. If the two are not unified, unit conversion is required before mapping. The value range of Sp is When the value of SP is , it means that all heart rate intervals △ i are mapped in the same grid, indicating that there are no abnormalities in both the data source and the data acquisition process, and no spurious signal is generated; when the value of SP is , it means that all heart rate intervals are different, and the scatter degree Sp of the arterial waveform graph to be analyzed is the largest. This section of data includes a large number of spurious signals.

[0259] For example: in step b3, when the abscissa on the horizontal axis of the arterial waveform graph to be analyzed is represented by the number of sampling points, and the unit of the grid of the mapping grid graph is ms, it is necessary to first calculate RR i and RR i+1 for (Δ i and Δ i+1 ) respectively:

[0260]

[0261]

[0262] Taking RR i and RR i+1 as the abscissa and ordinate, map (Δ i , Δ i+1 ) into the grid of the mapping grid graph;

[0263] For example: round down △1 / (0.1*SAMPLE) and denote it as RR1, round down △2 / (0.1*SAMPLE) and denote it as RR2, and map (RR1, RR2) into the mapping grid graph with the unit of ms.

[0264] bS6: Find the effective region of the arterial waveform graph to be analyzed. When in the difference sequence: h1, h2,...., h N , the maximum value is h Max ; in the difference sequence, find a continuous section of data Pstart~Pend, where Pstart and the two consecutive data on the right are both greater than C1*h Max , and Pend and the two consecutive data on the left are both less than C1*h Max , then: the data points between Pstart and Pend are considered as the effective data region of the arterial waveform graph to be analyzed. Specifically, as shown in Figure 8 .

[0265] Specifically, it includes the following steps:

[0266] bc1: In the difference sequence: h1, h2,...., h N find the maximum value and denote it as: h Max ;

[0267] When specifically implemented, if the number of h Max is greater than 1, the first maximum point is used for subsequent calculations.

[0268] bc2: Initialize the serial number j = 1,

[0269] bc3: Compare h j , h j+1 , h j+2 with C1*h Max respectively;

[0270] If there exist h j , h j+1 , h j+2 all greater than C1*h Max , then set h j as Pstart and execute step bc6;

[0271] Otherwise, execute step bc4;

[0272] bc4: Judge whether j is greater than Max - 2;

[0273] If j > Max - 2, then judge the arterial waveform data to be analyzed corresponding to the difference sequence as interference signal, that is, it is found that when the acquisition speed of the arterial waveform is too slow or too fast, there are not enough data points on the left side of the highest point h Max , resulting in measurement interference data, and as long as Pstart cannot be found, the data can be judged as interference data.

[0274] Otherwise, execute step bc5;

[0275] bc5: j = j + 1, and loop to execute steps bc3 - bc4;

[0276] bc6: Judge the relationship between j and N;

[0277] If j > N - 2, then judge the arterial waveform data to be analyzed corresponding to the difference sequence as interference signal;

[0278] Otherwise, execute step bc7;

[0279] bc7: j = j + 1;

[0280] bc8: h j , h j+1 , hj+2 Compare with C1*h respectively Max for comparison;

[0281] Among them, C1 is the effective peak-valley difference threshold, and the specific value is set according to research needs; in this embodiment, the effective peak-valley difference threshold C1 is 1 / 3, that is, when the difference in the vertical coordinates of the peak point and the valley point on the arterial waveform diagram to be analyzed is less than 1 / 3 of the highest point h Max , it is determined that the data point is an invalid data point;

[0282] If there is h j , h j+1 , h j+2 all greater than C1*h Max , then set h j+2 to Pend and execute step bc10;

[0283] Otherwise, execute step bc9;

[0284] bc9: Determine whether j is greater than N - 2;

[0285] If j > N - 2, then determine the data of the arterial waveform diagram to be analyzed corresponding to the difference sequence as interference signal, that is, when Pstart is found but Pend cannot be found, it means that there are not enough data points on the right side of the highest point h Max , and it is also determined as interference data;

[0286] Otherwise, j = j + 1, and loop to execute steps bc8 - bc9;

[0287] bc10: In the difference sequence: h1, h2,...., h N , the data points between Pstart and Pend are considered as the effective data area of the arterial waveform diagram to be analyzed.

[0288] Through h Max find the disturbance anomaly caused by the deflation process of the detection tool or the jitter and movement of the measured person, resulting in the arterial waveform acquisition speed being too slow or too fast, so that the number of data points on the left or both sides of the highest point is insufficient, that is, Pstart does not exist during the calculation process, or the measurement interference data caused by the non-existence of Pend. The interference data can be directly determined as spurious data and no subsequent calculation is required. If the number of data points on both sides of h Max is sufficient, then the data points between Pstart and Pend are considered as the effective data area of the arterial waveform diagram to be analyzed, and subsequent calculations are performed based on the data points between Pstart and Pend.

[0289] bS7: As shown in Figure 9 , perform data fitting on the effective data area, which specifically includes the following steps:

[0290] bd1: Retrieve the data from Pstart to Pend in the difference sequence, denoted as the valid data sequence.

[0291] The valid data sequence is: h Pstart , h Pstart+1 ....h max..... h Pend-1 , h Pend ;

[0292] bd2: Retrieve the data to the left of h max .: h Pstart , h Pstart+1 ....h max-1 , with h max. as the center, perform mirroring to obtain the left-side fitting data sequence: h Pstart , h Pstart+1 ....h max-1 , h max , h max - 1. ...h Pstart+1 , h Pstart ;

[0293] The abscissa corresponding to the left-side fitting data sequence is: x Pstart , x Pstart+1 ....x max-1 , x max , x max- 1....x Pstart ;

[0294] bd3: Retrieve the data to the right of h max. : h max+1..... h Pend-1 , h Pend , with h max. as the center, perform mirroring to obtain the right-side fitting data sequence: h Pend , h Pend-1 ....h max+1 , h max , h max+1..... h Pend-1 , h Pend ;

[0295] The abscissa corresponding to the right-side fitting data sequence is: x Pend , x Pend-1 ....x max+1 , x max , x max+1 ....x Pend-1 , x Pend .

[0296] bbS8: Calculate the peak amplitude perturbation rate PError corresponding to the arterial waveform to be analyzed;

[0297] be1: Calculate the number of left interference points ErrorLeft;

[0298] Using the left fitting data sequence and the abscissa corresponding to the left fitting data sequence as the ordinate and abscissa of the data points, obtain the quadratic fitting curve h = ax 2 + bx + c;

[0299] Count the number of points where the distance between the ordinate of the left valid data points and the fitting curve is greater than C2 * h Max and denote it as the number of left interference points ErrorLeft;

[0300] where C2 is the interference point judgment threshold. In actual work, not all data points can be strictly fitted to the fitting curve. As Figure 9 shown, there are three points below the quadratic fitting curve that are not fitted to the curve; however, the difference is very small and can also reflect the data trend and be used as valid data. In this application, the interference point judgment threshold C2 is used to control the judgment of interference data points to ensure that the technical solution of this application meets the actual production needs and adapts to various different application scenarios. The specific value of C2 is set according to research needs; in this embodiment, the interference point judgment threshold C2 is 1 / 8, that is, when the distance between the ordinate of the valid data points on the arterial waveform to be analyzed and the fitting curve is greater than C2 * h Max the data point is judged as an interference point.

[0301] be2: Calculate the number of right interference points ErrorRight;

[0302] Using the right fitting data sequence and the abscissa corresponding to the right fitting data sequence as the ordinate and abscissa of the data points, obtain the quadratic fitting curve h = ax 2 + bx + c;

[0303] Count the number of points where the distance between the ordinate of the right valid data points and the fitting curve is greater than C2 * h Max and denote it as the number of right interference points ErrorRight.

[0304] be3: Calculate the peak amplitude perturbation rate PError;

[0305] PError = (ErrorRight + ErrorLeft) / (Pend - Pstart + 1).

[0306] The peak amplitude perturbation rate PError represents the proportion of interference points within the data valid region. The larger the proportion of interference points, the greater the likelihood that the arterial waveform to be analyzed is a spurious interference signal.

[0307] bS9: Determine the nature of the arterial waveform to be analyzed;

[0308] Compare the scatter degree Sp of the arterial waveform to be analyzed with a preset scatter degree threshold C3, and compare the peak amplitude perturbation rate PError and a preset perturbation rate threshold C4;

[0309] When the following two conditions are simultaneously met, the arterial waveform to be analyzed is determined to be spurious interference data;

[0310] Sp > C3 and PError > C4.

[0311] The specific values of the scatter degree threshold C3 and the perturbation rate threshold C4 are set according to actual research requirements. In this embodiment, the scatter degree threshold C3 is 30%; the perturbation rate threshold C4 is 20%.

[0312] The method for identifying artifacts in blood pressure arterial waveform signals provided by this application determines the scattering degree Sp of the arterial peak intervals of the arterial waveform diagram to be analyzed through a Lorenz scatter plot, and judges the abnormal situation of the irregular distribution of arterial waveform peaks on the time axis; by finding the effective region of the arterial waveform diagram to be analyzed, the data of the arterial waveform diagram to be analyzed with abnormal peak points and valley points are judged as interference signals; through the mirror image data of the data in the effective region of the arterial waveform diagram to be analyzed, a quadratic fitting curve is obtained based on the least squares method, and then the peak amplitude perturbation rate PError of the arterial waveform diagram to be analyzed is calculated and judged, and the pressure difference between the peak points and valley points in the arterial waveform diagram to be analyzed is judged to determine the proportion of the number of abnormal peak points in the arterial waveform diagram to be analyzed; finally, the artifact interference data is jointly determined by two conditions of Sp and PError; when calculating Sp, by constructing a mapping grid diagram, the Lorenz scatter plot is mapped into the mapping grid diagram, and the scattering degree Sp is obtained by calculating the area hit by the data points in the mapping grid diagram, without the need to judge each data individually, greatly improving the calculation efficiency; at the same time, when drawing the mapping grid diagram, the data accuracy in the calculation process of the scattering degree Sp is adjusted by a reasonable interval threshold, the method is simple and easy to implement, further improving the calculation efficiency. The whole process of this application is based on the actual data situation of the arterial waveform diagram to be analyzed, and is not affected by the personal factors of the analyst, ensuring the rapid and accurate identification of artifact signals in blood artery waveform signals, especially suitable for the identification of blood pressure arterial waveform signals in massive data, greatly improving the artifact identification efficiency of blood pressure arterial waveform signals. At the same time, in the method for evaluating the quality of blood pressure arterial waveform signals provided by this application, based on the Sp and PError data, from two angles of the scattering degree of arterial peak intervals and the proportion of the number of abnormal peak points, quantitative data is used at the same time to judge the data quality of the arterial waveform diagram to be analyzed, without the need for personal subjective judgment by technicians, and the result is accurate and objective; at the same time, the quality judgment standard is adjusted according to the preset quality judgment threshold QH, so that the quality evaluation of this application is applicable to various different application scenarios.

[0313] After adopting the technical solution of the present invention, the irregular distribution of the arterial waveform peaks on the time axis caused by jitter, motion conditions, severe arrhythmia or external vibration interference is evaluated through the hitting area of the arterial peak interval on the Lorenz scatter plot; then the quadratic curve fitting is used to find the proportion of the number of abnormal peak points to evaluate the abnormal arterial waveform peak-valley pressure caused by jitter, motion conditions, uneven inflation or deflation speed; through the quantitative analysis of the data point characteristics included in the arterial waveform diagram to be analyzed, the whole analysis process does not require the subjective judgment of analysts, the result is accurate, and the calculation speed is fast. It is especially suitable for the analysis process of a large amount of blood pressure data, can quickly find all the artifact interference point data, and judge the quality of the blood pressure arterial waveform diagram signal according to the artifact signal, greatly improving the efficiency of data analysis. Then, based on the quality recognition result of the blood pressure arterial waveform diagram signal, the arterial waveform diagram signal with good quality is used to extract the PR interval, PP interval, PdH height and PsH height, and combined with the corresponding systolic blood pressure, mean blood pressure and diastolic blood pressure as labels to train the blood pressure measurement model. According to the good correlation between the pulse wave velocity and the mean blood pressure, diastolic blood pressure and systolic blood pressure, the blood pressure of the interfered arterial waveform diagram is quickly and accurately measured. At the same time, through the homologous data of the arterial waveform diagram data to be calculated, the weight coefficient in the connection of the blood pressure measurement model is trained and fine-tuned to make the model more targeted, so as to ensure that the finally obtained calculation result is more accurate. In the technical solution of this application, from the judgment of the quality of the blood pressure arterial waveform diagram signal to the final calculation of the blood pressure value of the interfered arterial waveform diagram, the whole process does not rely on the subjective judgment of technicians, the result is accurate and the calculation efficiency is high, and it is especially suitable for the processing of a large amount of blood pressure arterial waveform diagram signals.

Claims

1. A blood pressure value calculation method for an arterial waveform graph of disturbed blood pressure, characterized in that, It includes the following steps: S1: Construct a blood pressure measurement model based on a convolutional neural network; S2: According to a preset high-quality threshold, collect N high-quality arterial waveform diagrams based on a signal quality evaluation method, denoted as: basic arterial waveform diagrams; where N≥1000; The basic arterial waveform diagrams include: Static pressure curve: A curve showing the change of static pressure over time; Arterial waveform diagram: A curve showing the change of dynamic pressure over time; ECG waveform of channel II: A curve showing the change of the voltage of the ECG signal in channel II over time; S3: Measure the systolic blood pressure, mean blood pressure, and diastolic blood pressure corresponding to each of the basic arterial waveform diagrams based on the oscillometric method, denoted as the standard prediction results; S4: Respectively collect the basic data of each of the basic arterial waveform diagrams to form a basic data sequence, and use the standard prediction results corresponding to the basic data as the labels of the basic data sequence; The basic data includes: PR interval, PP interval, PdH height, and PsH height; The PR interval: The time difference between the peak point of the arterial waveform diagram and the nearest R wave in front of the electrocardiogram; The PP interval: The time difference between the peak point of the arterial waveform diagram and the previous peak point; The PdH height: The pressure difference between the peak point of the arterial waveform diagram and the next valley point; The PsH height: The ordinate pressure value of the static pressure curve corresponding to the abscissa position of the peak point of the arterial waveform diagram; S5: Based on the basic data sequence, construct a missing data sequence; Preset the number of sequences InNum included in the missing data sequence and the missing peak point value InP corresponding to each missing data sequence; Among them, InP is less than the number of peak points included in the basic arterial waveform diagram; For the basic data sequence, randomly select InP peak points in each of the basic arterial waveform diagrams, and set the basic data corresponding to the InP peak points to 0, that is, a missing data sequence is obtained; A total of InNum groups of missing data sequences are obtained, and each group includes N missing data sequences; S6: Mix and randomly sort the InNum groups of missing data sequences and the basic data sequence to form a training database; S7: Pre-train the blood pressure measurement model based on the training database to obtain the pre-trained blood pressure measurement model; S8: Obtain the data of the arterial waveform diagram to be calculated, and based on the signal quality evaluation method, find the micro-called arterial waveform diagram with better quality according to the preset training quality threshold; Denote the data other than the micro-called arterial waveform diagram as: interfered arterial waveform diagram; S9: Denote the micro-called arterial waveform diagram as: basic arterial waveform diagram, execute steps S4~S6, and denote the obtained training database as: micro-called training database; S10: Use the micro-called training database to fine-tune the pre-trained blood pressure measurement model to obtain the trained blood pressure measurement model; S11: For each of the interfered arterial waveform diagrams, specify the abscissa of the highest peak and the abscissa of the credible peak, and send the two abscissas and the interfered arterial waveform diagram into the trained blood pressure measurement model, and the blood pressure measurement model outputs the prediction results corresponding to each of the interfered arterial waveform diagrams; The predicted results include: systolic blood pressure, mean blood pressure, and diastolic blood pressure corresponding to the disturbed arterial waveform diagram.

2. The blood pressure value calculation method for an arterial waveform diagram of blood pressure affected by interference according to claim 1, characterized in that: The convolutional neural network includes, connected in sequence: an input layer, a convolutional layer, a pooling layer, and three fully connected layers; Among them, the number of channels of the input layer is 4, the number of channels of the convolutional layer and the pooling layer is 8, and the number of channels of the fully connected layer is:

1.

3. The blood pressure value calculation method for an arterial waveform diagram of blood pressure affected by interference according to claim 1, wherein: The acquisition method of the basic data sequence includes the following steps: a1: Find the highest peak point of the basic arterial waveform diagram and mark it as P0; a2: Centered on P0, find 10 peak points before and after respectively; If the number of peak points on the left or right side of P0 is less than 10, execute step a4; Otherwise, execute step a3; a3: Denote the peak points before and after P0 as: P -10 , P -9 , P -8 ... P -1 and P1, P2... P 10 ; Calculate the PR interval, PP interval, PdH height, and PsH height corresponding to the peak points, which are the basic data; execute step a5; a4: Calculate all the basic data corresponding to all existing peak points, and at the same time set the basic data corresponding to the missing peak points to 0 to obtain 21 groups of data, which are the basic data corresponding to the basic arterial waveform diagram; execute step a5; a5: Calculate the basic data of all 21 peak points, and respectively form a PR interval sequence, a PP interval sequence, a PdH height sequence, and a PsH height sequence; record the PR interval sequence, the PP interval sequence, the PdH height sequence, and the PsH height sequence as the basic data sequence.

4. The blood pressure value calculation method for an arterial waveform graph of blood pressure affected by interference according to claim 1, characterized in that: The signal quality assessment method includes the following steps: b1: Obtain the arterial waveform diagram to be analyzed; b2: Calculate the scatter Sp and the peak amplitude perturbation rate PError corresponding to the obtained arterial waveform diagram to be analyzed; b3: Determine whether the arterial waveform diagram to be analyzed is pseudo-interference data according to the scatter Sp and the peak amplitude perturbation rate PError; If it is pseudo-interference data, the quality of the arterial waveform diagram to be analyzed is set as data with poor indicators; Otherwise, execute step b4; b4: Calculate the treatment evaluation parameter Qua of the arterial waveform diagram to be analyzed: Qua = SP*α + PError*β; Among them, α is the scatter weight coefficient, and β is the perturbation rate weight coefficient; b5: Compare Qua with the preset quality judgment threshold QH; When Qua is less than QH, the quality of the data corresponding to the arterial waveform diagram to be analyzed is judged as high-quality data, otherwise, the data corresponding to the arterial waveform diagram to be analyzed is judged as data with poor indicators.

5. The blood pressure value calculation method for an arterial waveform diagram of blood pressure affected by interference according to claim 1, characterized in that: In step S5, InNum is set to 4; the corresponding InP values include: InP1 = 10, InP2 = 15, InP3 = 18, InP4 = 20.

6. The blood pressure value calculation method for an arterial waveform graph of blood pressure affected by interference according to claim 4, characterized in that: In step b3, the judgment method of the pseudo-interference data includes the following steps: bS1: Obtain the arterial waveform diagram to be analyzed; The horizontal axis of the arterial waveform diagram to be analyzed is the pulse beat position caused by the heartbeat, and the vertical axis is the pressure value collected by the pressure sensor of the dynamic blood pressure monitor; bS2: Mark the peak points and valley points of the arterial waveform diagram to be analyzed; Based on the arterial waveform diagram to be analyzed, find all the peak points and valley points; starting from the first peak point, pair it with the adjacent valley point; Assume that the arterial waveform diagram to be analyzed includes N pairs of peak points and valley points, then the peak points and valley points included in the arterial waveform diagram to be analyzed are expressed as: (Peak1, Valley1), (Peak2, Valley2),..., (PeakN, ValleyN); The abscissas of peak points Peak1 - PeakN form a sequence x1, x2,... x N ; bS3: Obtain the ordinates of each pair of peak points and valley points, and calculate the difference of the ordinates of each pair of peak points and valley points; Then a sequence of the absolute values of the differences of the ordinates of each pair of peak points and valley points is obtained: Denoted as the difference sequence: h1, h2,...., h N ; bS4: Draw a Lorenz scatter plot, which specifically includes the following steps: ba1: Calculate the difference sequence of the sequence x1, x2,... x N , and record it as △1 = x2 - x1, △2 = x3 - x2,... △ N-1 = x N - x N-1 ; △ i represents the pulse interval or heart beat interval, i is a natural number, i ∈ (1, N - 1); ba2: Combine adjacent △ i in pairs to form (△ i , △ i+1 ); ba3: Construct coordinate axes, and the units of both the abscissa and the ordinate are ms; Using △ in each group (△ i , △ i+1 ) as the abscissa and △ i as the ordinate, plot (△ i+1 , △ i , △ i+1 ) on the coordinate axes to obtain a Lorenz scatter plot; bS5: The scatter degree Sp of the arterial waveform diagram to be analyzed specifically includes the following steps: bb1: Construct a mapping grid diagram; Set the heartbeat interval threshold and the reasonable interval threshold; The heartbeat interval threshold represents the maximum value of the reasonable heartbeat interval data; The reasonable interval threshold means that when the difference between two heartbeat intervals of the same data source is less than the reasonable interval threshold, these two heartbeat intervals are regarded as the same data during calculation; The mapping grid includes: a mapping coordinate system and a grid; Both the horizontal axis and the vertical axis of the mapping coordinate system are set as the heartbeat interval, the starting points of the horizontal axis and the vertical axis are 0, and the maximum value is the heartbeat interval threshold; Between the horizontal axis and the vertical axis, construct the grid with the interval allowable value as the unit; bb2: Take out each group (△ i , △ i+1 ), and judge the relationship between △ i and the heart rate interval threshold; If △ i > the heart rate interval threshold, then (△ i , △ i+1 ) will not be involved in the calculation; Otherwise, judge the relationship between △ i+1 and the heart rate interval threshold; If △ i+1 > the heart rate interval threshold value, then (△ i , △ i+1 ) does not participate in the calculation; Otherwise, execute step bb3; bb3: Map (△ i , △ i+1 ) to the grid of the said mapping grid graph with △ i as the abscissa and △ i+1 as the ordinate; bb4: When all of (△ i , △ i+1 ) have participated in the calculation, count the number of grids successfully mapped in the mapped grid diagram; Assume that n grids in the mapping grid diagram are successfully mapped, and calculate the scatter degree Sp of the arterial waveform diagram to be analyzed; Sp = n / (N - 1); bS6: Find the effective region of the arterial waveform diagram to be analyzed, which specifically includes the following steps: bc1: In the difference sequence: h1, h2,...., h N find the maximum value and denote it as: h Max ; bc2: Initialize the serial number j = 1, bc3: Compare h j , h j+1 , h j+2 with C1*h Max respectively; If h exists j , h j+1 , h j+2 are all greater than C1 * h Max , then set h j to Pstart and execute step bc6; Otherwise, execute step bc4; bc4: Judge whether j is greater than Max - 2; If j > Max - 2, then judge the arterial waveform diagram data corresponding to the difference sequence as spurious interference data; Otherwise, execute step bc5; bc5: j = j + 1, and loop to execute steps bc3 - bc4; bc6: Judge the relationship between j and N; If j > N - 2, then judge the arterial waveform diagram data corresponding to the difference sequence as spurious interference data; Otherwise, execute step bc7; bc7: j = j + 1; bc8: Compare h j , h j+1 , h j+2 with C1*h Max respectively; Among them, C1 is the effective peak-valley difference threshold; If h exists j 、h j+1 、h j+2 are all greater than C1*h Max , then set h j+2 to Pend and execute step bc10; Otherwise, execute step bc9; bc9: Judge whether j is greater than N - 2; If j > N - 2, then judge the arterial waveform diagram data corresponding to the difference sequence as interference signals; Otherwise, j = j + 1, and loop to execute steps bc8 - bc9; bc10: The difference sequence: h1, h2,...., h N Among them, the data points between Pstart and Pend are considered as the effective data region of the arterial waveform diagram to be analyzed; bS7: Perform data fitting on the effective data region, which specifically includes the following steps: bd1: Take out the data included in Pstart to Pend in the difference sequence and record it as the effective data sequence, The valid data sequence is: h Pstart , h Pstart+1 .... h max ..... h Pend-1 , h Pend ; bd2: Take out h max Data on the left: h Pstart , h Pstart+1 .... h max-1 , with h max as the center for mirroring to obtain the left-side fitting data sequence: h Pstart , h Pstart+1 .... h max-1 , h max , h max-1 .... h Pstart+1 , h Pstart ; The abscissa corresponding to the left - hand fitting data sequence is: x Pstart , x Pstart+1 .... x max-1 , x max , x max-1 ....x Pstart ; bd3: Take out h max Data on the right: h max+1..... h Pend-1 , h Pend , with h max as the center, perform mirroring to obtain the right-side fitting data sequence: h Pend , h Pend-1 .... h max+1 , h max , h max+1..... h Pend-1 , h Pend ; The abscissas corresponding to the right-side fitting data sequence are: x Pend , x Pend-1 .... x max+1 , x max , x max+1 .... x Pend-1 ,x Pend ; bS8: Calculate the peak amplitude perturbation rate PError corresponding to the arterial waveform diagram to be analyzed; be1: Calculate the number of left interference points ErrorLeft; Taking the left - hand side fitting data sequence and the abscissa corresponding to the left - hand side fitting data sequence as the ordinate and abscissa of the data points, the quadratic fitting curve h = ax 2 + bx + c is obtained using the least - squares method; Count the number of points whose vertical coordinate of the valid data points on the left side and the distance from the fitting curve are greater than C2*h Max and denote it as the number of left interference points ErrorLeft; Among them, C2 is the interference point judgment threshold; be2: Calculate the number of right interference points ErrorRight; Taking the right-side fitting data sequence and the abscissa corresponding to the right-side fitting data sequence as the ordinate and abscissa of the data points, the quadratic fitting curve h = ax 2 + bx + c is obtained using the least squares method; Count the number of points where the vertical distance between the valid data points on the right side and the fitting curve is greater than C2*h Max and denote it as the number of interference points on the right side, ErrorRight; be3: Calculate the peak amplitude perturbation rate PError; PError = (ErrorRight + ErrorLeft) / (Pend - Pstart + 1); bS9: Determine the nature of the arterial waveform diagram to be analyzed; bf1: Compare the scatter degree Sp of the arterial waveform diagram to be analyzed with a preset scatter degree threshold C3; bf2: Compare the peak amplitude perturbation rate PError with a preset perturbation rate threshold C4; bf3: When the following two conditions are simultaneously satisfied, the arterial waveform diagram to be analyzed is determined as spurious interference data; Sp > C3 and PError > C4.

7. The method for calculating blood pressure value for an arterial blood pressure waveform graph affected by interference according to claim 6, wherein: The heart rate interval threshold is set to 2000 ms; the reasonable interval threshold is set to 100 ms.

8. The blood pressure value calculation method for an arterial waveform graph of blood pressure affected by interference according to claim 6, characterized in that: The effective peak-to-valley difference threshold C1 is 1 / 3; the interference point judgment threshold C2 is 1 / 8; the scatter degree threshold C3 is 30%; the perturbation rate threshold C4 is 20%.

9. The blood pressure value calculation method for an artery waveform graph of disturbed blood pressure according to claim 6, characterized in that: The scatter degree weight coefficient α takes a value of 25%; the perturbation rate weight coefficient β takes a value of 75%; the quality judgment threshold QH takes a value of 20%.

Images