CN114722874A - Heart beat clustering method based on focus region registration - Google Patents

Abstract

The heart beat clustering method based on focus region registration can quickly obtain the main form of an electrocardiosignal by using a small calculated amount, and meanwhile, can overcome the interference caused by baseline drift and improve the accuracy of the result. According to the technical scheme, through constructing an R-wave position histogram and a histogram screening condition, careless beats are quickly deleted, heart beats with similar characteristics are found, the main morphological peak characteristics are positioned, then heart beat data classified into non-main morphological peaks are directly positioned to heart beat data in the main morphological peak characteristics closest to the heart beat data through time, the similarity of the heart beat data and the heart beat data is calculated, heart beat samples meeting the similarity precision requirement are found, and the heart beat samples and the main morphological peak characteristics form a clustering template.

Description

Heart beat clustering method based on focus region registration

Technical Field

The invention relates to the technical field of medical artificial intelligence, in particular to a heart beat clustering method based on focus region registration.

Background

The method for analyzing the electrocardiosignal heartbeat based on the clustering algorithm plays an important role in the intelligent analysis of the electrocardiosignal and the clinical auxiliary diagnosis. Generally, the electrocardiograph clustering algorithm is to cluster heartbeats in a cluster pair with similar shapes based on the characteristic parameters of heartbeats or template matching to obtain the characteristic waveform of the electrocardiograph signal of the target object for technical personnel to analyze and assist diagnosis. However, the calculation method based on the characteristic parameters is easily interfered by noise and has poor stability, and the accuracy of the obtained characteristic waveform is unstable; the calculation method based on the template matching method has low efficiency and large calculation amount, and is difficult to distinguish heartbeats with relatively close forms. The method of combining template matching and characteristic parameters for analyzing the electrocardiosignal also has the advantage that the defect of large calculation amount caused by the template matching method cannot be solved. Finally, the automatic clustering device for the electrocardiographic waveforms in the prior art has poor adaptability, cannot meet the application under complex polymorphic electrocardiographic signals, has insufficient accuracy of obtained results, and needs to manually participate in the judgment of the finally obtained results.

Disclosure of Invention

In order to solve the problems of large calculation amount or insufficient accuracy of the conventional automatic clustering device for the electrocardiographic waveforms, the invention provides a heart beat clustering method based on focus area registration, which can quickly obtain the main form of the electrocardiographic signals by using smaller calculation amount, and can overcome the interference caused by baseline drift and improve the accuracy of the result.

The technical scheme of the invention is as follows: a heart beat clustering method based on focus area registration comprises the following steps:

s1, performing baseline filtering and high-pass filtering on the original electrocardiographic waveform to remove noise influence;

s2, performing QRS identification to obtain heartbeat data, wherein the voltage sampling value of the baseline position of the electrocardiowave is 0 mv;

s3, obtaining an original heart beat data sequence;

setting: the original electrocardio wave mode comprises N heartbeats in total;

sampling each heart beat, and setting: the sampling time length is T milliseconds;

intercepting a cardiac electric wave of about T/2 milliseconds from each heart beat by taking an R wave as a center to serve as a sample of the heart beat, obtaining N samples of the heart beats in total, and recording the N samples as the samples of the heart beats to be processed;

setting the sampling frequency of each heartbeat sample as F, wherein each heartbeat sample to be processed comprises F T/1000 sampling points;

it is characterized by also comprising the following steps:

s4, constructing an R wave position histogram for all the heart beat samples to be processed;

the abscissa of the R-wave position histogram is a sampling value distribution interval, and the ordinate represents the number of the heart beat samples to be processed, of which the sampling values of the R-wave positions are in the corresponding abscissa interval;

s5, finding the peak characteristics of the main form;

finding an abscissa corresponding to the maximum ordinate value based on the R wave position histogram, and recording the abscissa as Xmax;

respectively searching left and right by taking Xmax as a starting point, and finding out an abscissa which simultaneously meets the following 2 histogram screening conditions;

histogram screening condition one: 1, | X-Xmax | >;

and (5) histogram screening conditions II: the ordinate value of the X point is 0, or when searching to the left, the ordinate value of the X point is less than X-1, or when searching to the right, the ordinate value of the X point is less than X + 1;

when searching leftwards, the found sitting marks meeting the histogram screening condition are made into leftX, and when searching rightwards, the found sitting marks meeting the histogram screening condition are made into rightX; recording the interval [ leftX, rightX ] as the same characteristic interval;

dividing all the heart beat samples to be processed in the histogram into two sets, wherein the R wave sampling value positions fall into samples of [ leftX, rightX ] to form a set A, and the rest form a set B; at this time, the heart beat sample to be processed in the set A is a sample according with the peak characteristic of the main form;

s6, setting a basic characteristic sampling time tb;

the basic feature sampling time indicates that the features of the homomorphic heart beat data are in the same feature interval within the basic feature sampling time tb;

namely, when two cardioverter samples to be processed respectively have Nd feature points in the same feature interval leftwards and rightwards by taking the R wave position as a starting point, the two cardioverter samples are considered as homomorphic cardioverter samples;

Nd ＝ ［（T／2）／tb ］；

s7, removing all the feature points strictly conforming to the main form features in the set A to obtain a feature set of the main form, wherein the method specifically comprises the following steps;

a1, if d sampling points are included in one basic characteristic sampling time tb, taking the R wave position as a starting point, and Nd sampling points are respectively arranged on the left and right of each heart beat sample to be processed;

a 2: taking the R wave as a starting point, finding the j x d sampling points leftwards or rightwards, and marking as the j comparison characteristic points;

wherein j is a natural number not more than Nd, j is 1, 2, …, Nd;

a3, recording the heartbeat sample to be processed in the set A as a characteristic snack beat sample;

a4, constructing a comparison characteristic point position histogram for all the characteristic snack beat samples based on the jth comparison characteristic point;

the abscissa of the comparison feature point position histogram is a sampling value distribution interval, and the ordinate represents the number of the to-be-processed heart beat samples of which the sampling value of the jth comparison feature point is in the corresponding abscissa interval;

a 5: based on the comparison feature point position histogram, finding the abscissa corresponding to the value with the maximum ordinate, and recording the abscissa as Xmax;

searching leftwards and rightwards respectively by taking Xmax as a starting point, and finding out an abscissa which simultaneously meets the first histogram screening condition and the second histogram screening condition;

making a leftwards search for the found sit-up mark meeting the histogram screening condition as a leftwards mark, and making a rightwards search for the found sit-up mark meeting the histogram screening condition as a rightwards mark; recording the interval [ leftX, rightX ] as the same characteristic interval;

a6, dividing all the characteristic snack beat samples into two sets, and comparing the sampling value positions of the characteristic points to fall into [ LeftX, RightX]Is named set A_jThe rest of the components form a set B_j；

At this time set A_jThe heart beat sample to be processed is a sample with the same characteristics as the j x d th comparison characteristics;

a7 grouping A_jIn place of said set A, at j<When the Nd is performed, j equals j +1, and then steps a 3-a 7 are executed in a circulating mode;

until j is Nd, performing step a 8;

a 8-set A that will be the final result_NdAs a final set A, the set A at this time is a sample set strictly conforming to the main morphological characteristics;

all the sets B_jThe heart beat sample to be processed in (1) is put into the set B;

s8, setting a similarity threshold value T hd;

the similarity threshold T hd is the precision of the clustering template, namely the similarity of the two heartbeats is greater than the coincidence threshold T hd, and the two heartbeats are determined to accord with the same characteristic and belong to the same clustering set;

s9, based on the similarity threshold T hd, taking out samples meeting the similarity requirement from the set B and putting the samples into the set A to obtain a clustering template, wherein the calculation process comprises the following steps:

setting that Nb heart beat samples to be processed are shared in the set B;

taking out each heart beat sample to be processed one by one, recording the heart beat sample as a heart beat i, searching the heart beat sample to be processed which is closest to the heart beat i in time in the set A, and recording the heart beat sample as a heart beat iNear;

calculating to obtain the similarity SIM between the heart beat i and the heart beat iNear, and if the similarity SIM is greater than the similarity threshold value Thd, moving the sample from the set B to the set A;

and calculating all Nb heartbeat samples to be processed in the set B to obtain a sample contained in the set A, namely a clustering template.

It is further characterized in that:

it also includes the following steps:

s10, taking the sample data in the set B as the heart beat sample to be processed, circularly executing the steps S4-S15, and continuously calculating to obtain other clustering templates;

in step S9, the method for calculating the similarity between the cardiac beat i and the cardiac beat ienar includes the following steps:

b1, finding out all characteristic peak areas of the heart beat i and the heart beat iNear respectively;

setting: the set of all the characteristic peak regions included in the heartbeat i is mFt;

the set of all characteristic peak regions comprised by the heart beat iNear is nFt;

b2, calculating the correlation degree of the characteristic peak set mFt and the heart beat iNear, and recording as follows: correlation degree mSim;

b3, calculating the correlation degree of the characteristic peak set nFt and the heart beat i, and recording as follows: correlation nSim;

b4, copying the smaller value of the mSim and the nSim to the similarity SIM;

in step b1, the method for calculating the characteristic peak region includes:

c1, setting the search range to be about 300 milliseconds at R;

setting: searching the heart beat participating in the characteristic peak area as a heart beat to be searched;

c2, finding all sampling points within 300 milliseconds before and after the R wave of the heart beat to be searched;

setting and taking x as a center corresponding to each sampling point x respectively, wherein the range is as follows: a window of [ x-len, x + len ] is recorded as a retrieval window;

c3, finding out all sampling points meeting the following peak point conditions in the retrieval window, and recording as peak points;

peak point condition one: point x is maximum or minimum;

peak point condition two: the difference between the highest point and the lowest point in the window is more than 0.3 mv;

c4, respectively searching sampling points leftwards and rightwards by taking the peak point as a center, finding out the sampling points meeting the following conditions, and respectively recording as xLeft and xRight:

the time of the distance x is more than 200ms, or the sampling value of the point is more than the sampling value of the point < 0;

the region [ xLeft, xRight ] forms a characteristic peak region with x as a peak point;

c5, sorting all the found characteristic peak areas according to the found sequence;

in steps b2 and b3, the method for calculating the correlation degree comprises the following steps:

d1, recording the feature peak set participating in calculation as: a set xFt, wherein the heartbeats participating in the calculation are marked as heartbeats xx;

d2, taking out the characteristic peak area with the sequence number xi according to the sequence of the characteristic peak areas in the set xFt, and recording the characteristic peak area as follows: a characteristic peak area to be calculated;

setting the width of the characteristic peak region as LenI and the corresponding waveform as iWvFt;

d3, finding all starting points S corresponding to the characteristic peak area to be calculated in the heartbeat xx;

d4, taking a waveform region nWvS with the same width in the heart beat xx by taking each starting point S as a starting point, and calculating the Pearson correlation coefficient Cor of the waveforms iWvFt and nWvS:

the match value iValue = pearson correlation coefficient Cor;

d 5: finding the maximum matching value from all the matching values iValue corresponding to the characteristic peak area to be calculated, and setting the maximum matching value as the matching value corresponding to the characteristic peak area to be calculated;

d6, setting the value of the correlation degree of the set xFt and the heartbeat xx as the minimum value in the matching values iValue corresponding to all the characteristic peak areas to be calculated;

the searching method of the starting point S comprises the following steps:

e1, confirming the relation between the sequence number xi of the characteristic peak area to be calculated and 1;

if xi is 1: step e2 is executed;

otherwise xi >1, go to step e 3;

e2, the starting point S corresponding to the heart beat xx of the characteristic peak area to be calculated with the sequence number xi is:

all sampling points 300ms in front of the R wave to 300ms behind the R wave of the cardiorespiratory beat xx;

e3, wherein the starting point S corresponding to the characteristic peak area to be calculated with the sequence number xi in the heartbeat xx is:

all sampling points with sequence numbers xi-1 between 100ms in front of and 100ms behind the matching position iPoint corresponding to the characteristic peak area to be calculated;

the calculation method of the matching position iPoint comprises the following steps:

f1, calculating the Peak to be calculated with the sequence number xi-1, the Peak to be calculated, corresponding to each starting point S in the heartbeat xx, the Peak to be calculated characteristic peak to be calculated, the Peak to be calculated, the Peak to be calculated, the Peak to calculate, the Peak to be calculated, the Peak to be calculated, the Peak, the;

amplitude variance ValChange = | MaxMin1-MaxMin0|/MaxMin 0;

wherein: MaxMin0 is the maximum minimum difference of all waveforms iWvFt included in set xFt, MaxMin1 is the maximum minimum difference of nWvS of all waveforms included in beat xx;

f2, finding the starting point S which meets the following condition, namely maxS:

ValChange <0.2 and Pearson correlation coefficient Cor is maximum;

matching position iPoint = maxS + LenI-1.

The heart beat clustering method based on focus region registration provided by the invention has the advantages that through constructing an R wave position histogram and a histogram screening condition, careless beats are quickly deleted, heart beats with similar characteristics are found, a main morphological peak value characteristic is positioned, then heart beat data classified into non-main morphological peak values are directly positioned to heart beat data in the main morphological peak value characteristic closest to the heart beat data through time, the similarity of the heart beat data and the heart beat data is calculated, heart beat samples meeting the similarity precision requirement are found, and the heart beat samples and the main morphological peak value characteristic form a clustering template; in the calculation process, when the peak characteristics of the main form are positioned, a large number of QRS samples contained in the main form can be directly judged to be similar, and further correlation calculation is not needed, so that the matching speed is greatly increased; in the process of generating the clustering template based on the main morphological peak value characteristics, the heartbeat data participating in calculation is quickly positioned through time, and irrelevant heartbeat data does not participate in calculation, so that the calculation amount is further greatly reduced, and meanwhile, the matching precision is improved. Meanwhile, in the process of calculating the similarity, the two heart beat data participating in calculation are respectively divided according to the characteristic peak regions, and when the two heart beat data are compared, the interference of baseline drift to the similarity is reduced, the main peak feature and the secondary peak feature are ensured to be included in the result, and the accuracy of the matching result of the waveform similarity of the two heart beat data is improved.

Drawings

FIG. 1 is a sample of a cardiac beat to be processed;

in FIG. 2, A is a schematic diagram of the R-wave position of each heartbeat, and B is an embodiment of a histogram of the R-wave positions;

FIG. 3 is an embodiment of peak feature search for a dominant mode;

FIG. 4 is a sample of set A and set A_jA schematic diagram of (a);

FIG. 5 is a schematic diagram showing the position relationship between the heartbeat i and the heartbeat iNear on the heartbeat to be processed;

FIG. 6 is an embodiment of a characteristic peak region calculation;

FIG. 7 is a first embodiment of similarity calculation;

FIG. 8 is a second embodiment of similarity calculation;

fig. 9 is a flowchart of a heart beat clustering method based on focus region registration.

Detailed Description

The invention relates to a heart beat clustering method based on focus area registration, which comprises the following steps.

And S1, performing baseline filtering and high-pass filtering on the original electrocardiographic waveform to remove noise influence.

After filtering, the voltage sampling value of the baseline position of the cardiac wave is considered to be 0 mv. Then QRS identification is carried out to obtain N heart beats.

And S2, performing QRS identification to obtain heartbeat data, wherein the voltage sampling value of the baseline position of the electrocardiowave is 0 mv.

S3, obtaining an original heart beat data sequence;

setting: the original electrocardio wave mode comprises N heartbeats in total;

sampling each heart beat, and setting: the sampling time length is T milliseconds;

intercepting a cardiac electric wave of about T/2 milliseconds from each heart beat by taking an R wave as a center to serve as a sample of the heart beat, obtaining N samples of the heart beats in total, and recording the N samples as the samples of the heart beats to be processed;

and setting the sampling frequency of each heartbeat sample as F, wherein each heartbeat sample to be processed comprises F T/1000 sampling points.

In the embodiment shown in fig. 1, T =1000ms and F =128, there are: for each heartbeat, a cardiac wave of 500ms or so is taken out as a sample of the heartbeat around the R wave, and N samples of heartbeats are obtained in total. Each sample is 1000ms in duration, assuming a sampling rate of 128, i.e., 128 samples are contained.

S4, constructing an R wave position histogram for all heart beat samples to be processed;

the abscissa of the R-wave position histogram is a sampling value distribution interval, and the ordinate represents the number of the heart beat samples to be processed, of which the sampling values of the R-wave positions are in the corresponding abscissa interval.

As shown in a diagram in fig. 2, the sampling value of each heart beat R-wave position is counted. The distribution of the sampling values is counted and a histogram is drawn to obtain a B diagram in fig. 2, which is an R-wave position histogram in this embodiment.

The histogram abscissa x =122 represents 900 samples in which the sampling value of the R-wave position is [1.05mv,1.1mv ]). Each abscissa point represents a sampling value distribution interval: 0 represents < -5mv,1 represents [ -5mv, -5+0.05mv),. i represents [ -5+ (i-1) × 0.05mv, -5+ i × 0.05mv ],. 200 represents [5.95mv,5mv),201 represents > =5 mv. The ordinate represents the number of heart beats in which "the sampling value of the R-wave position" is in the corresponding abscissa interval.

S5, finding the peak characteristics of the main form;

based on the R-wave position histogram, finding the abscissa corresponding to the maximum ordinate value, and recording the abscissa as Xmax;

respectively searching left and right by taking Xmax as a starting point, and finding out an abscissa which simultaneously meets the following 2 histogram screening conditions;

histogram screening condition one: 1, | X-Xmax | >;

and (5) histogram screening conditions II: the ordinate value of the X point is 0, or when searching to the left, the ordinate value of the X point is less than X-1, or when searching to the right, the ordinate value of the X point is less than X + 1.

Making a leftwards search for the found sit-up mark meeting the histogram screening condition as a leftwards mark, and making a rightwards search for the found sit-up mark meeting the histogram screening condition as a rightwards mark; recording the interval [ leftX, rightX ] as the same characteristic interval;

dividing all heart beat samples to be processed in the histogram into two sets, wherein the positions of R wave sampling values fall into samples of [ leftX, rightX ] to form a set A, and the rest of the R wave sampling values form a set B; at this time, the heartbeat sample to be processed in the set A is a sample which accords with the peak characteristic of the main form.

As shown in fig. 3, an embodiment is found for the peak feature of the dominant mode.

Finding the abscissa point Xmax =122 corresponding to the histogram ordinate value (900) at maximum.

Respectively searching left and right by taking an abscissa Xmax =122 as a starting point, and if the abscissa X simultaneously satisfies the following two conditions:

firstly, the method comprises the following steps: 1 | X-Xmax | >1

Secondly, the method comprises the following steps: "the ordinate value of the X point is 0" or "the ordinate value of the X point is smaller than X + -1" (take-1 to the left, take +1 to the right) the ordinate of the point;

when the ordinate value of the X point is 0, the subsequent point enters the statistical range of another 'peak' no matter whether the subsequent point is upward or downward; when the ordinate value of the X point is less than X +/-1, the Y point also indicates that the shape range from the X +/-1 coordinate to another 'peak' is entered; in the embodiment of fig. 3, the next point "X ± 1" of the abscissa X that meets the condition is also counted in the current main form peak value, and a slight change occurs in the adjacent sampling points due to disturbance in the actual data acquisition process, but two different peaks do not occur. Then, in this embodiment, left boundary point LeftX =120 and right boundary point RightX =125 are obtained, all samples in the histogram are divided into two sets, the sample whose R-wave sampling value position falls in the interval [ LeftX, RightX ] constitutes set a, and the rest constitutes set B.

S6, setting a basic characteristic sampling time tb;

the basic characteristic sampling time indicates that the characteristics of the heart beat data in the same form are in the same characteristic interval within the basic characteristic sampling time tb;

namely, when two cardioverter samples to be processed respectively have Nd feature points in the same feature interval towards the left and the right by taking the R wave position as a starting point, the two cardioverter samples are considered as homomorphic cardioverter samples;

Nd ＝ ［（T／2）／tb ］。

in this embodiment, tb is set to 30ms, i.e., it is considered that the waveform within 30ms can represent the characteristics of the heartbeat sample to be processed. When the value of T is 1000, the value of T,

（1000/2）/30 = 16.667

after rounding the decimal place, Nd =16, that is, when two heartbeats start from the R-wave position, 16 points are located in the same characteristic interval to the left or to the right respectively, it is considered that the two heartbeats have the same form and belong to the same cluster set.

S7, removing all the feature points strictly conforming to the main form features in the set A to obtain the feature set of the main form, which specifically comprises the following steps.

a1, setting that d sampling points are included in one basic characteristic sampling time tb, and each heartbeat sample to be processed takes the R wave position as a starting point and has Nd sampling points to the left and the right respectively.

a 2: taking the R wave as a starting point, finding the j x d sampling points leftwards or rightwards, and marking as the j comparison characteristic points;

wherein j is a natural number not more than Nd, j is 1, 2, …, Nd.

In this embodiment, taking the R wave as the center, the values of j left or right are: j ═ 1, 2, …, 16;

taking the R wave as the center on the abscissa, and taking the sampling points participating in calculation in the left direction and the right direction as follows: 4,8,12,......,64.

a3, recording the heart beat sample to be processed included in the set A as a characteristic snack beat sample.

a4, constructing a comparison characteristic point position histogram for all characteristic snack swatches on the basis of the jth comparison characteristic point;

the abscissa of the comparison feature point position histogram is a sampling value distribution interval, and the ordinate represents the number of the heartbeat samples to be processed, of which the sampling value of the jth comparison feature point is in the corresponding abscissa interval.

a 5: based on the comparison of the feature point position histogram, finding the abscissa corresponding to the maximum ordinate value, and recording as Xmax;

searching leftwards and rightwards respectively by taking Xmax as a starting point, and finding out an abscissa which simultaneously meets a first histogram screening condition and a second histogram screening condition;

making a leftwards search for the found sit-up mark meeting the histogram screening condition as a leftwards mark, and making a rightwards search for the found sit-up mark meeting the histogram screening condition as a rightwards mark; the interval [ LeftX, RightX ] is marked as the same characteristic interval.

a6 dividing all the characteristic snapshot samples into two sets, and comparing the positions of the characteristic point sampling values to fall into [ leftX, rightX ]]Is named set A_jThe rest of the components form a set B_j；

At this time set A_jThe heart beat sample to be processed in (1) is a sample with the same characteristics as the j x d th comparison characteristics.

As shown in FIG. 4, samples of set A and set A_jSchematic representation of (a).

a7 grouping A_jIn j instead of the set A<When the Nd is performed, j equals j +1, and then steps a 3-a 7 are executed in a circulating mode;

until j becomes Nd, step a8 is performed.

a 8-set A that will be the final result_NdAs a final set A, the set A at this time is a sample set strictly conforming to the main morphological characteristics;

all the sets B_jThe heart beat sample to be processed in (1) is put into the set B.

S8, setting a similarity threshold value T hd;

the similarity threshold T hd is the precision of the clustering template, i.e. the similarity of two heartbeats is greater than the coincidence threshold T hd, then the same features are considered to be coincided, and the same clustering set is belonged to.

The specific value of the T hd is set according to the precision requirement of the calculation, and the precision of the result is controlled according to the actual requirement through the T hd, so that the method is suitable for different application scenes. In this embodiment, Thd =95%, when the similarity degree threshold of the two waveforms is 95% or more, it is determined that the waveforms conform to the same feature and belong to the same cluster set.

S9, based on the similarity threshold T hd, taking out samples meeting the similarity requirement from the set B and putting the samples into the set A to obtain a clustering template, wherein the calculation process comprises the following steps:

setting that the set B is provided with Nb heart beat samples to be processed;

taking out each heart beat sample to be processed one by one from the set B, recording the heart beat sample as a heart beat i, searching the heart beat sample to be processed which is closest to the heart beat i in time in the set A, and recording the heart beat sample as a heart beat iNear; as shown in fig. 5, it is a schematic diagram of the position relationship between the heartbeat i and the heartbeat iinear on the heartbeat to be processed;

calculating to obtain the similarity SIM between the heart beat i and the heart beat iNear, and if the similarity SIM is greater than a similarity threshold value Thd, moving the sample from the set B to the set A;

and calculating the Nb heartbeat samples to be processed in the set B to obtain the samples contained in the set A, namely the cluster template.

And S10, taking the sample data in the set B as the heart beat sample to be processed, circularly executing the steps S4-S15, and continuously calculating to obtain other clustering templates.

In step S9, the method for calculating the similarity between the cardiac beat i and the cardiac beat ienar includes the following steps:

b1, finding out all characteristic peak areas of the heart beat i and the heart beat iNear respectively;

setting: the set of all the characteristic peak regions included in the heartbeat i is mFt;

the set of all characteristic peak regions comprised by the heart beat iNear is nFt;

b2, calculating the correlation degree of the characteristic peak set mFt and the heart beat iNear, and recording as follows: correlation degree mSim;

b3, calculating the correlation degree of the characteristic peak set nFt and the heart beat i, and recording as follows: correlation nSim;

b4 copying the smaller of the mSim and nSim values to the similarity SIM.

In the method, a smaller value of mSim and nSim is used as a similarity SIM between the heart beat i and the heart beat iNear, and is compared with a degree threshold T hd, so that the similarity obtained through calculation in the method is ensured to be strict, and the precision of the cluster set obtained in the method is further ensured.

In step b1, the method for calculating the characteristic peak area includes:

c1, setting the search range to be about 300 milliseconds at R;

setting: searching the heart beat participating in the characteristic peak area as a heart beat to be searched;

c2, finding all sampling points within 300 milliseconds before and after the R wave of the heart beat to be searched;

setting and taking x as a center corresponding to each sampling point x respectively, wherein the range is as follows: a window of [ x-len, x + len ] is recorded as a retrieval window;

c3, finding out all sampling points meeting the following peak point conditions in the retrieval window, and recording as peak points;

peak point condition one: point x is maximum or minimum;

peak point condition two: the difference between the highest point and the lowest point in the window is more than 0.3 mv;

c4, searching sampling points leftwards and rightwards respectively by taking the peak point as a center, finding out the sampling points meeting the following conditions, and respectively recording as xLeft and xRight:

the time of the distance x is more than 200ms, or the sampling value of the point is more than the sampling value of the previous point and is less than 0;

the region [ xLeft, xRight ] forms a characteristic peak region with x as a peak point;

c5, sorting all the found characteristic peak areas according to the found sequence.

The comparison is performed using samples within 300 seconds after the R wavefront because 300ms represents the size of the region to be matched, and the method looks at the waveform over the time of a complete cardiac cycle of a heartbeat. In practical application, the PQRST wave process of one heartbeat cycle can be included 300ms after the R wavefront, so that the region waveform 300ms after the R wavefront is taken for calculation.

The waveform comprises a P wave, a QRS wave and a T wave in a complete cardiac cycle time range of one heart beat. In the method, the similarity of the P wave is not required to be considered in the matching process, and only the similarity of the T wave of the QRS wave is considered. The limit of the height of the normal P wave in the electrocardiogram is 0.3mv, and the difference between the highest point and the lowest point in the window is more than 0.3mv, which indicates that the window is not the window with the peak caused by the P wave as the main part. The height and duration of the QRST wave are specifically ranged in order to find the individual peaks and their corresponding left and right boundaries caused by the QRST wave in the electrocardiogram. In practical application, because the peak point of the QRST wave is not more than 200ms away from its boundary, the method limits the passage of xLeft and xRight for 200ms, and the time when a certain point satisfies "the distance x is more than 200 ms" is not found.

The method examines the similarity of QRS wave and T wave in a complete cardiac cycle time range of a heart beat. Len is set here to search for peaks corresponding to QRS and T waves using a windowing method. Len is taken as data smaller than the length of the heartbeat, and in the embodiment, Len takes 100ms and can cover a peak area caused by QRST waves.

As shown in fig. 6, the position relationship of the characteristic peak regions in the set mFt and the set nFt in the heartbeat data is shown schematically.

In steps b2 and b3, the method for calculating the degree of correlation comprises the following steps:

d1, recording the feature peak set participating in calculation as: a set xFt, wherein the heartbeats participating in the calculation are marked as heartbeats xx;

d2, according to the sequence of the characteristic peak areas in the set xFt, taking out the characteristic peak area with the sequence number xi, and recording as follows: a characteristic peak area to be calculated;

setting the width of the characteristic peak region as LenI and the corresponding waveform as iWvFt;

d3, finding all starting points S corresponding to the characteristic peak area to be calculated in the heartbeat xx;

d4 taking a waveform region nWvS with the same width in the heart beat xx by taking each starting point S as a starting point, and calculating the Pearson correlation coefficient Cor of the waveforms iWvFt and nWvS:

the match value iValue = pearson correlation coefficient Cor;

d 5: finding the maximum matching value from all the matching values iValue corresponding to the characteristic peak area to be calculated, and setting the maximum matching value as the matching value corresponding to the characteristic peak area to be calculated;

d6, setting the value of the correlation degree between the set xFt and the heartbeat xx as the minimum value in all matching values iValue corresponding to the characteristic peak area to be calculated;

when the similarity is calculated, the characteristic peak areas in the set xFt are taken out one by one, for each characteristic peak area, the S starting point is used as the starting point on the waveform of the heartbeat xx, LenI is used as the width, all areas with the same width are searched in a sliding mode, and then the Pearson Correlation Coefficient (Pearson Correlation Coefficient) of each area on the heartbeat xx and the characteristic peak area to be calculated is calculated;

for the pearson correlation coefficient, a larger absolute value indicates a stronger correlation. Finding out the most similar area of each characteristic peak area to be calculated on the heartbeat xx through the maximum matching value; then, the correlation degree between the set xFt and the heartbeat xx is represented by the smallest pearson correlation coefficient corresponding to all the characteristic peak areas, and the calculated correlation degree is ensured to be the strictest correlation degree.

The searching method of the starting point S comprises the following steps:

e1, confirming the relation between the serial number xi of the characteristic peak area to be calculated and 1;

if xi is 1: step e2 is executed;

otherwise xi >1, go to step e 3;

e2, the starting point S corresponding to the heart beat xx of the characteristic peak area to be calculated with the sequence number xi is:

all sampling points 300ms in front of the R wave to 300ms behind the R wave of the cardiorespiratory beat xx;

e3, the corresponding starting point S of the characteristic peak area to be calculated with the sequence number xi in the heartbeat xx is:

and all sampling points with the sequence number xi-1 between 100ms in front of and 100ms behind the matching position iPoint corresponding to the characteristic peak area to be calculated.

When a starting point S is determined, when a characteristic peak area to be calculated with the sequence number of 1 is searched, the characteristic peak area to be calculated is searched within the range of 300ms before and after the R wave of the heart beat to be searched, once the characteristic peak area to be calculated with the sequence number of 1 is found, the characteristic peak area to be calculated 2 is searched within 100ms before and after the characteristic peak area to be calculated with the sequence number of 1, similarly, the characteristic peak area to be calculated … … is searched within 100ms before and after the characteristic peak area to be calculated with the sequence number of 2, the position relation between the characteristic peak area found on the heart beat xx and the previous characteristic peak area is ensured to be consistent with the position relation between the characteristic peak areas to be calculated in the set xFt, and the result of the similarity between the heart beat i and the heart beat iNear finally obtained is ensured to be accurate.

The calculation method of the matching position iPoint comprises the following steps:

f1, calculating the Peak region to be calculated with the serial number xi-1, and the Peak region to be calculated, corresponding to each starting point S in the heartbeat xx, and calculating the Peak correlation coefficient Cor and the amplitude difference ValChange;

amplitude variance ValChange = | MaxMin1-MaxMin0|/MaxMin 0;

wherein: MaxMin0 is the maximum minimum difference of all waveforms iWvFt included in set xFt, MaxMin1 is the maximum minimum difference of nWvS of all waveforms included in beat xx;

if only the pearson correlation coefficient Cor of the waveforms iWvFt and nWvS is considered when calculating the similarity, even if the Cor is large (close to 100%), only the rising and falling trends of the waveforms iWvFt and nWvS can be consistent, and the amplitudes of the two waveforms cannot be ensured to be consistent; according to the invention, amplitude difference ValChange is obtained by calculation through MaxMin0 and MaxMin1, and two similar waveforms found based on the method are ensured by combining with the Pearson correlation coefficient Cor, so that not only the waveform trend is constant, but also the waveform amplitudes are consistent. And further ensuring the accuracy of the clustering template calculated based on the method.

f2, finding a starting point S which meets the following condition, namely maxS:

ValChange <0.2 and pearson correlation coefficient Cor is maximal;

matching position iPoint = maxS + LenI-1.

As shown in fig. 7, the first embodiment of similarity calculation is shown.

And (3) taking out the characteristic peak area with the sequence number of 1 on the heart beat m, and calculating to obtain 3 characteristic peak areas: characteristic peak areas 1-3 are arranged from left to right on the abscissa according to the found sequence; and taking out the characteristic peak calculation region with the serial number of 1 as a characteristic peak region to be calculated. And on the waveform of the heart beat n, respectively taking all sampling points 300ms before the R wave and 300ms behind the R wave as starting points S, and calculating the Pearson correlation coefficients one by one.

Assuming that, by calculation, on a heartbeat n, the pearson correlation coefficient of the region with the starting point S of-300 ms and the characteristic peak region to be calculated is 0%, and the pearson correlation coefficient of the region with the starting point S of-100 ms and the characteristic peak region to be calculated is 30%, the matching value iValue of the characteristic peak region 1 is = 30%.

As shown in fig. 8, the second embodiment of similarity calculation is shown.

In addition to the embodiment shown in fig. 7, the calculation feature peak region with the extraction number 2 is continuously extracted as the feature peak region to be calculated. And searching the 2 nd characteristic peak area to be calculated within 100ms before and after the matching position iPoint corresponding to the characteristic peak area to be calculated with the serial number of 1 on the waveform of the heartbeat n, and calculating the lowest correlation degree of the characteristic peak set mFt of the heartbeat m and the heartbeat n one by one.

The starting point S in fig. 7 is an area of-100 ms, and if the condition "ValChange <0.2 and the pearson correlation coefficient Cor is maximum" is met, the starting point S is-100 ms and is used as maxS, and the matching position iPoint of the characteristic peak area to be calculated with the serial number 1 is calculated. Subsequent calculations are then made with all sample points within 100ms before and after the iPoint as starting points S.

After the technical scheme of the invention is used, on the aspect of calculation efficiency, the disordered waveforms are quickly eliminated by adopting threading stripping of the focus area, and the peak characteristics of the main form are checked. Since a large number of QRS samples contained in the main morphology can be directly judged to be similar, further correlation calculation is not needed, and the matching speed is greatly enhanced. In the matching accuracy, all characteristic peak areas on the heartbeat data are found, all the characteristic peak areas are taken as focuses, the similarity between two QRS waves on different heartbeat data is measured by the similarity in a 'focus area correlation' mode, and the defects that the baseline drift anti-interference capability is weak and the similarity of secondary peaks is neglected due to the fact that the correlation coefficient matching is carried out on the whole QRS interval in the traditional method are avoided.

Claims (7)

1. A heart beat clustering method based on focus region registration comprises the following steps:

s1, performing baseline filtering and high-pass filtering on the original electrocardiographic waveform to remove noise influence;

s2, performing QRS identification to obtain heartbeat data, wherein the voltage sampling value of the baseline position of the electrocardiowave is 0 mv;

s3, obtaining an original heart beat data sequence;

setting: the original electrocardio wave mode comprises N heartbeats in total;

sampling each heart beat, and setting: the sampling time length is T milliseconds;

intercepting a cardiac electric wave of about T/2 milliseconds from each heart beat by taking an R wave as a center to serve as a sample of the heart beat, obtaining N samples of the heart beats in total, and recording the N samples as the samples of the heart beats to be processed;

setting the sampling frequency of each heartbeat sample as F, wherein each heartbeat sample to be processed comprises F T/1000 sampling points;

it is characterized by also comprising the following steps:

s4, constructing an R wave position histogram for all the heart beat samples to be processed;

the abscissa of the R-wave position histogram is a sampling value distribution interval, and the ordinate represents the number of the heart beat samples to be processed, of which the sampling values of the R-wave positions are in the corresponding abscissa interval;

s5, finding the peak characteristics of the main form;

finding an abscissa corresponding to the maximum ordinate value based on the R wave position histogram, and recording the abscissa as Xmax;

respectively searching left and right by taking Xmax as a starting point, and finding out an abscissa which simultaneously meets the following 2 histogram screening conditions;

histogram screening condition one: 1, | X-Xmax | >;

and (5) histogram screening conditions II: the ordinate value of the X point is 0, or when searching to the left, the ordinate value of the X point is less than X-1, or when searching to the right, the ordinate value of the X point is less than X + 1;

making a leftwards search for the found sit-up mark meeting the histogram screening condition as a leftwards mark, and making a rightwards search for the found sit-up mark meeting the histogram screening condition as a rightwards mark; recording the interval [ leftX, rightX ] as the same characteristic interval;

dividing all the heart beat samples to be processed in the histogram into two sets, wherein the R wave sampling value positions fall into samples of [ leftX, rightX ] to form a set A, and the rest form a set B; at this time, the heart beat sample to be processed in the set A is a sample according with the peak characteristic of the main form;

s6, setting a basic characteristic sampling time tb;

the basic feature sampling time indicates that the features of the homomorphic heart beat data are in the same feature interval within the basic feature sampling time tb;

namely, when two cardioverter samples to be processed respectively have Nd feature points in the same feature interval leftwards and rightwards by taking the R wave position as a starting point, the two cardioverter samples are considered as homomorphic cardioverter samples;

Nd ＝ ［（T／2）／tb ］；

s7, removing all the feature points strictly conforming to the main form features in the set A to obtain a feature set of the main form, wherein the method specifically comprises the following steps;

a1, if d sampling points are included in one basic characteristic sampling time tb, taking the R wave position as a starting point, and Nd sampling points are respectively arranged on the left and right of each heart beat sample to be processed;

a 2: taking the R wave as a starting point, finding the j x d sampling points leftwards or rightwards, and marking as the j comparison characteristic points;

wherein j is a natural number not more than Nd, j is 1, 2, …, Nd;

a3, recording the heartbeat sample to be processed in the set A as a characteristic snack beat sample;

a4, constructing a comparison characteristic point position histogram for all the characteristic snack beat samples based on the jth comparison characteristic point;

the abscissa of the comparison feature point position histogram is a sampling value distribution interval, and the ordinate represents the number of the to-be-processed heart beat samples of which the sampling value of the jth comparison feature point is in the corresponding abscissa interval;

a 5: based on the comparison feature point position histogram, finding the abscissa corresponding to the value with the maximum ordinate, and recording the abscissa as Xmax;

searching leftwards and rightwards respectively by taking Xmax as a starting point, and finding out an abscissa which simultaneously meets the first histogram screening condition and the second histogram screening condition;

making a leftwards search for the found sit-up mark meeting the histogram screening condition as a leftwards mark, and making a rightwards search for the found sit-up mark meeting the histogram screening condition as a rightwards mark; recording the interval [ leftX, rightX ] as the same characteristic interval;

a6 dividing all the characteristic snack swatches into two sets, and comparing the positions of the characteristic point sampling values to fall into [ leftX, rightX]Is named set A_jThe rest of the components form a set B_j；

Set A at this time_jThe heart beat sample to be processed is a sample with the same characteristics as the j x d th comparison characteristics;

a7 grouping A_jIn place of said set A, at j<When the Nd is performed, j equals j +1, and then steps a 3-a 7 are executed in a circulating mode;

until j is Nd, performing step a 8;

a 8-set A that will be the final result_NdAs a final set A, the set A at this time is a sample set strictly conforming to the main morphological characteristics;

all the sets B_jThe heart beat sample to be processed in (1) is put into the set B;

s8, setting a similarity threshold value T hd;

the similarity threshold T hd is the precision of the clustering template, namely the similarity of the two heartbeats is greater than the coincidence threshold T hd, and the two heartbeats are determined to accord with the same characteristic and belong to the same clustering set;

s9, based on the similarity threshold T hd, taking out samples meeting the similarity requirement from the set B and putting the samples into the set A to obtain a clustering template, wherein the calculation process comprises the following steps:

setting that Nb heart beat samples to be processed are shared in the set B;

taking out each heart beat sample to be processed one by one, recording the heart beat sample as a heart beat i, searching the heart beat sample to be processed which is closest to the heart beat i in time in the set A, and recording the heart beat sample as a heart beat iNear;

calculating to obtain the similarity SIM between the heart beat i and the heart beat iNear, and if the similarity SIM is greater than the similarity threshold Thd, moving the sample from the set B to the set A;

and calculating all Nb heartbeat samples to be processed in the set B to obtain a sample contained in the set A, namely a clustering template.

2. The heart beat clustering method based on focus area registration according to claim 1, characterized in that: it also includes the following steps:

and S10, taking the sample data in the set B as the heart beat sample to be processed, circularly executing the steps S4-S15, and continuously calculating to obtain other clustering templates.

3. The heart beat clustering method based on focus area registration according to claim 1, characterized in that: in step S9, the method for calculating the similarity between the cardiac beat i and the cardiac beat ienar includes the following steps:

b1, finding out all characteristic peak areas of the heart beat i and the heart beat iNear respectively;

setting: the set of all the characteristic peak regions included in the heartbeat i is mFt;

the set of all characteristic peak regions comprised by the heart beat iNear is nFt;

b2, calculating the correlation degree of the characteristic peak set mFt and the heart beat iNear, and recording as follows: correlation degree mSim;

b3, calculating the correlation degree of the characteristic peak set nFt and the heart beat i, and recording as follows: correlation nSim;

b4, copying the smaller value of the mSim and the nSim to the similarity SIM.

4. The heart beat clustering method based on focus area registration according to claim 3, characterized in that: in step b1, the method for calculating the characteristic peak region includes:

c1, setting the search range to be about 300 milliseconds at R;

setting: searching the heart beat participating in the characteristic peak area as a heart beat to be searched;

c2, finding all sampling points within 300 milliseconds before and after the R wave of the heart beat to be searched;

setting and taking x as a center corresponding to each sampling point x respectively, wherein the range is as follows: a window of [ x-len, x + len ] is recorded as a retrieval window;

c3, finding out all sampling points meeting the following peak point conditions in the retrieval window, and recording as peak points;

peak point condition one: point x is maximum or minimum;

peak point condition two: the difference between the highest point and the lowest point in the window is more than 0.3 mv;

c4, respectively searching sampling points leftwards and rightwards by taking the peak point as a center, finding out the sampling points meeting the following conditions, and respectively recording as xLeft and xRight:

the time of the distance x is more than 200ms, or the sampling value of the point is more than the sampling value of the point < 0;

the region [ xLeft, xRight ] constitutes a characteristic peak region with x as a peak point;

and c5, sorting all the found characteristic peak areas according to the found sequence.

5. The heart beat clustering method based on focus area registration according to claim 3, characterized in that: in steps b2 and b3, the method for calculating the correlation degree comprises the following steps:

d1, recording the feature peak set participating in calculation as: xFt, recording the heartbeats participating in the calculation as heartbeats xx;

d2, according to the sequence of the characteristic peak areas in the set xFt, taking out the characteristic peak area with the sequence number xi, and recording as follows: a characteristic peak area to be calculated;

setting the width of the characteristic peak region as LenI and the corresponding waveform as iWvFt;

d3, finding all starting points S corresponding to the characteristic peak area to be calculated in the heartbeat xx;

d4, taking a waveform region nWvS with the same width in the heart beat xx by taking each starting point S as a starting point, and calculating the Pearson correlation coefficient Cor of the waveforms iWvFt and nWvS:

matching value iValue = pearson correlation coefficient Cor;

d 5: finding the maximum matching value from all the matching values iValue corresponding to the characteristic peak area to be calculated, and setting the maximum matching value as the matching value corresponding to the characteristic peak area to be calculated;

d6, setting the value of the correlation degree between the set xFt and the heartbeat xx as the minimum value in the matching values iValue corresponding to all the characteristic peak areas to be calculated.

6. The heart beat clustering method based on focus area registration according to claim 5, characterized in that: the searching method of the starting point S comprises the following steps:

e1, confirming the relation between the serial number xi of the characteristic peak area to be calculated and 1;

if xi is 1: step e2 is executed;

otherwise xi >1, go to step e 3;

e2, wherein the starting point S corresponding to the heart beat xx in the characteristic peak area to be calculated with the sequence number xi is as follows:

all sampling points 300ms in front of the R wave to 300ms behind the R wave of the cardiorespiratory beat xx;

e3, wherein the starting point S corresponding to the characteristic peak area to be calculated with the sequence number xi in the heartbeat xx is as follows:

and all sampling points with sequence numbers xi-1 between 100ms in front of and 100ms behind the matching position iPoint corresponding to the characteristic peak area to be calculated.

7. The heart beat clustering method based on focus area registration according to claim 6, characterized in that: the calculation method of the matching position iPoint comprises the following steps:

f1, calculating the Peak to be calculated with the sequence number xi-1, the Peak to be calculated, corresponding to each starting point S in the heartbeat xx, the Peak to be calculated characteristic peak to be calculated, the Peak to be calculated, the Peak to be calculated, the Peak to calculate, the Peak to be calculated, the Peak to be calculated, the Peak, the;

amplitude variance ValChange = | MaxMin1-MaxMin0|/MaxMin 0;

wherein: MaxMin0 is the maximum minimum difference of all waveforms iWvFt included in set xFt, MaxMin1 is the maximum minimum difference of nWvS of all waveforms included in beat xx;

f2, finding the starting point S which meets the following condition, namely maxS:

ValChange <0.2 and Pearson correlation coefficient Cor is maximum;

matching position iPoint = maxS + LenI-1.

Images