CN113598788B - Heartbeat recognition pre-judging model - Google Patents
Heartbeat recognition pre-judging model Download PDFInfo
- Publication number
- CN113598788B CN113598788B CN202110701837.9A CN202110701837A CN113598788B CN 113598788 B CN113598788 B CN 113598788B CN 202110701837 A CN202110701837 A CN 202110701837A CN 113598788 B CN113598788 B CN 113598788B
- Authority
- CN
- China
- Prior art keywords
- heartbeat
- data
- model
- wave crest
- judging
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000001914 filtration Methods 0.000 claims abstract description 16
- 230000002861 ventricular Effects 0.000 claims abstract description 15
- 238000000034 method Methods 0.000 claims abstract description 13
- 238000012549 training Methods 0.000 claims abstract description 11
- 208000000418 Premature Cardiac Complexes Diseases 0.000 claims description 10
- 238000004364 calculation method Methods 0.000 claims description 7
- 238000012935 Averaging Methods 0.000 claims description 3
- 206010015856 Extrasystoles Diseases 0.000 claims description 3
- 238000000605 extraction Methods 0.000 claims description 3
- 210000002569 neuron Anatomy 0.000 claims description 3
- 230000002028 premature Effects 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 101100134058 Caenorhabditis elegans nth-1 gene Proteins 0.000 claims 1
- 238000013136 deep learning model Methods 0.000 abstract description 5
- 238000012216 screening Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 230000007774 longterm Effects 0.000 description 2
- 206010042602 Supraventricular extrasystoles Diseases 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 238000002565 electrocardiography Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 208000019622 heart disease Diseases 0.000 description 1
- 238000002513 implantation Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/316—Modalities, i.e. specific diagnostic methods
- A61B5/318—Heart-related electrical modalities, e.g. electrocardiography [ECG]
- A61B5/346—Analysis of electrocardiograms
- A61B5/349—Detecting specific parameters of the electrocardiograph cycle
- A61B5/352—Detecting R peaks, e.g. for synchronising diagnostic apparatus; Estimating R-R interval
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Abstract
The application relates to a heartbeat identification pre-judging model, which is characterized by comprising the following steps of: step one, collecting data, wherein the data set contains at least 3000 ventricular early heartbeats in total; step two, filtering the data and cutting the data into heartbeats; step three, training a plurality of models by taking the previous heartbeat data as input and taking the next heartbeat data as output; and step four, judging new data. The heartbeat recognition algorithm based on the deep learning model and the correlation is provided, and the method not only considers the relation before and after the heartbeat, but also considers the electrocardio characteristics of individuals, so that the judgment accuracy is improved; in addition, for heart beat of uncertain type, the method provided by the patent also prompts the heart beat; the misleading of software to doctors is reduced.
Description
Technical Field
The application belongs to the technical field of electrocardiosignal processing, and particularly relates to a heartbeat recognition pre-judging model.
Background
The electrocardiogram has the characteristics of non-implantation and low cost, so that the electrocardiogram is widely applied to screening of heart diseases. However, since some diseases are intermittent, long-term dynamic electrocardiographic monitoring is often required to be discovered. One of the important tasks in the analysis of long-term electrocardiography is to pick up two types of ectopic beats, i.e. supraventricular premature beats. However, because the signals are often long, it is time consuming and labor consuming to manually select these abnormal heartbeats, and computer-aided screening is necessary.
The existing auxiliary screening method is to judge the ectopic heartbeat by utilizing the waveform characteristics, but other waveform positioning algorithms except the R wave crest position in the electrocardiogram have poor accuracy and certain limitations. In recent years, due to the development of deep learning technology, classifying heartbeats by using a deep learning model has become a research hotspot. However, most deep learning models do not consider the personal characteristics of the ectopic heart beat, and the influence of the previous heart beat on the current heart beat judgment is ignored to some extent. Thus resulting in current deep learning models that vary in lead form or are less accurate at the patient's discretion.
Disclosure of Invention
The invention aims to solve the technical problems that: the heartbeat identification pre-judging model aims at solving the problem that the accuracy rate of classifying heartbeats by adopting a deep learning model in the prior art is not high, and accordingly is provided.
The technical scheme adopted for solving the technical problems is as follows:
a heartbeat identification pre-judgment model, comprising the steps of:
step one, collecting data, wherein the data set contains at least 3000 ventricular early heartbeats in total;
step two, filtering the data and cutting off the data into heartbeats: firstly, filtering; secondly, marking the position of each R wave crest in the electrocardiosignal, cutting off the first time period before the occurrence of the R wave crest to the second time period after the occurrence of the R wave crest as a heartbeat, sequentially arranging and numbering the heartbeats in each electrocardiosignal, marking each heartbeat, marking the data of the normal heartbeat as 0, marking the data of the early heartbeat on the room as 1, marking the data of the early heartbeat in the room as 2, and marking the data of other heartbeat types as 3;
step three, training a plurality of models by taking the previous heartbeat data as input and taking the next heartbeat data as output;
judging new data:
after new data are generated, firstly, an R wave crest in an electrocardiogram is extracted by adopting any R wave extraction algorithm, a first time period before the occurrence of the R wave crest to a second time period after the occurrence of the R wave crest are cut off to be used as a heartbeat, n sections of heartbeats are generated, the first heartbeat type is not judged, and the second heartbeat is judged in such a way that if the interval RR2q between the R wave crest position of the second heartbeat and the first R wave crest position is smaller than the interval RR2h between the R wave crest position of the second heartbeat and the third R wave crest position by more than 0.12 seconds and the QRS width is larger than 0.08 seconds, the heart beat is regarded as a ventricular premature heart beat; an supraventricular early heartbeat is considered if RR2q < RR2h-0.12 seconds and QRS width is equal to or less than 0.08 seconds; other conditions are considered normal heartbeats;
the calculation method of the QRS width comprises the following steps: performing median filtering on the heartbeat data, wherein a window of 0.16 seconds is selected as a median filtering parameter; (2) averaging with a window of 0.04 seconds; the processed heartbeat data is marked as an ecg2, the original electrocardiograph data and a first intersection point q of the ecg2 before the position of the R wave crest, and the interval between the first intersection points s appearing after the R wave crest is marked as the QRS width;
for the judgment of the m+1th heartbeat, firstly, selecting a model according to the m-th heartbeat type, predicting the data of the m+1th heartbeat according to the 4 models trained in the step three, respectively marking as yy0, yy1, yy2 and yy3,
performing correlation calculation on yy0, yy1, yy2 and yy3 and the actual m+1st heartbeat data y3 respectively; if yy0 correlation is highest, judging that the (m+1) th heartbeat is a normal heartbeat; if the yy1 correlation is highest, judging that the m+1st heartbeat is an on-room early heartbeat; if the yy2 correlation is highest, judging that the m+1st heartbeat is an ventricular premature beat; if the yy3 correlation is highest, judging that the (m+1) th heartbeat is other heartbeats; n is a natural number, m is a natural number greater than or equal to 2 and less than n;
and sequentially judging the type of the next heartbeat until the model is selected to judge the nth heartbeat according to the type of the nth heartbeat to be 1 st heartbeat.
Preferably, in the heartbeat recognition pre-judging model of the present invention, in the step three, the heartbeat data with the front data marked as a and the heartbeat data with the rear data marked as B are used for training the a-B model.
Preferably, in the heartbeat recognition pre-judging model of the present invention, in the step three, a model framework adopts a Seq2Seq architecture:
the framework consists of two RNN models; the first RNN encodes data through feature learning, encodes data with certain dimension into an information vector c, decodes the data through the information vector c, and finally outputs data with the same dimension as the input dimension;
the model will be the formerHeartbeat data x= (x) 1 ,x 2 ...x i ...x 350 ) As input, where x i Representing the i-th frame heartbeat data; the latter heartbeat data y= (y) 1 ,y 2 ....y i ....y 350 ) As an output, where y i Representing predicted i-th frame heartbeat data;
h＝(h 0 ,h 1 ...h i ....h 350 ) For hidden state parameters learned during model coding, where h i Representing an ith hidden state parameter; h '= (h' 0 ,h' 1 ...h' i ....h' 350 ) Is a hidden state parameter in the model decoding process, wherein h' i Representing an ith hidden state parameter;
encoding process of model: h is a 0 Is set as zero vector and heartbeat data x for the preset initial hiding state parameter 1 At the same time, input into the first RNN to learn the hidden state parameter h i In combination with heartbeat data x i+1 Sequentially downwards transmitting, and finally outputting an information vector c containing all information;
decoding process of model: the information vector c is input into the second RNN as an initial hidden state, and only the hidden layer state of the last neuron is received without receiving other inputs, and finally the predicted heartbeat data y is output.
Preferably, the heartbeat recognition pre-judging model of the present invention, the first time period is 0.3 seconds, and the second time period is 0.4 seconds.
Preferably, in the heartbeat identification pre-judging model, in the fourth step, when the model is selected according to the mth heartbeat type, if the mth heartbeat is a normal heartbeat, a 0-0 model, a 0-1 model, a 0-2 model and a 0-3 model are selected; if the mth heartbeat is the early-ventricular heartbeat, selecting a model 1-0, a model 1-1, a model 1-2 and a model 1-3; if the mth heartbeat is the ventricular early heartbeat, selecting a 2-0 model, a 2-1 model, a 2-2 model and a 2-3 model; and if the mth heartbeat is of other heartbeat types, selecting a 3-0 model, a 3-1 model, a 3-2 model and a 3-3 model.
Preferably, in the heartbeat identification pre-judging model of the present invention, in the fourth step, the computer judging type from the third heartbeat to the n-1 th heartbeat is used as the final result.
Preferably, in the heartbeat identification pre-judging model, in the first step, at least 2 thousands of electrocardiographic data containing ectopic heartbeats and normal heartbeats are collected when data are collected; each piece of electrocardiographic data has a length of at least 20 hours.
Preferably, in the heartbeat recognition pre-judging model of the invention, in the first step, when data is collected, the sampling frequency is 500Hz, and if the data is not 500Hz, the data is firstly resampled to 500Hz.
Preferably, in the heartbeat recognition pre-judging model, in the second step, a band-pass filter of 0.1-100Hz is adopted when filtering is carried out.
Preferably, in the heartbeat identification pre-judging model of the present invention, in the step two, a manual or program is adopted in the step of marking each R wave position in the electrocardiograph signal and marking each heartbeat.
The beneficial effects of the invention are as follows: the method not only considers the relation before and after the heartbeat, but also considers the electrocardio characteristics of the individual, thereby improving the judgment accuracy.
Drawings
The technical scheme of the application is further described below with reference to the accompanying drawings and examples.
FIG. 1 is a flowchart of overall steps of a heartbeat identification pre-judgment model in an embodiment of the present application;
FIG. 2 is a flowchart of a heartbeat identification pre-determination model step four in an embodiment of the present application;
fig. 3 is a logic schematic diagram of a heartbeat identification pre-judgment model step three Seq2Seq architecture according to an embodiment of the present application.
Detailed Description
It should be noted that, in the case of no conflict, the embodiments and features in the embodiments may be combined with each other.
In the description of the present application, it should be understood that the terms "center," "longitudinal," "transverse," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the drawings, merely to facilitate description of the present application and simplify the description, and do not indicate or imply that the devices or elements referred to must have a specific orientation, be configured and operated in a specific orientation, and therefore should not be construed as limiting the scope of protection of the present application. Furthermore, the terms "first," "second," and the like, are used for descriptive purposes only and are not to be construed as indicating or implying a relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first", "a second", etc. may explicitly or implicitly include one or more such feature. In the description of the invention, unless otherwise indicated, the meaning of "a plurality" is two or more.
In the description of the present application, it should be noted that, unless explicitly specified and limited otherwise, the terms "mounted," "connected," and "connected" are to be construed broadly, and may be either fixedly connected, detachably connected, or integrally connected, for example; can be mechanically or electrically connected; can be directly connected or indirectly connected through an intermediate medium, and can be communication between two elements. The specific meaning of the terms in this application can be understood by those of ordinary skill in the art in a specific context.
The technical solutions of the present application will be described in detail below with reference to the accompanying drawings in combination with embodiments.
Examples
The embodiment provides a heartbeat recognition pre-judging model, as shown in fig. 1, comprising the following steps:
step one, collecting data, wherein the data set contains at least 3000 ventricular early heartbeats in total;
step two, filtering the data and cutting off the data into heartbeats: firstly, filtering; secondly, marking the position of each R wave crest in the electrocardiosignal, cutting off the first time period before the occurrence of the R wave crest to the second time period after the occurrence of the R wave crest as a heartbeat, sequentially arranging and numbering the heartbeats in each electrocardiosignal, marking each heartbeat, marking the data of the normal heartbeat as 0, marking the data of the early heartbeat on the room as 1, marking the data of the early heartbeat in the room as 2, and marking the data of other heartbeat types as 3;
step three, training a plurality of models by taking the previous heartbeat data as input and taking the next heartbeat data as output;
judging new data:
after new data are generated, firstly, an R wave crest in an electrocardiogram is extracted by adopting any R wave extraction algorithm, a first time period before the occurrence of the R wave crest to a second time period after the occurrence of the R wave crest are cut off to be used as a heartbeat, n sections of heartbeats are generated, the first heartbeat type is not judged, and the second heartbeat is judged in such a way that if the interval RR2q between the R wave crest position of the second heartbeat and the first R wave crest position is smaller than the interval RR2h between the R wave crest position of the second heartbeat and the third R wave crest position by more than 0.12 seconds and the QRS width is larger than 0.08 seconds, the heart beat is regarded as a ventricular premature heart beat; an supraventricular early heartbeat is considered if RR2q < RR2h-0.12 seconds and QRS width is equal to or less than 0.08 seconds; other conditions are considered normal heartbeats;
the calculation method of the QRS width comprises the following steps: (1) Performing median filtering on the heartbeat data, wherein a window of 0.16 seconds is selected as a median filtering parameter; (2) averaging with a window of 0.04 seconds; the processed heartbeat data is marked as an ecg2, the original electrocardiograph data and a first intersection point q of the ecg2 before the position of the R wave crest, and the interval between the first intersection points s appearing after the R wave crest is marked as the QRS width;
for the judgment of the m+1 (m > =2) th heartbeat, firstly, selecting a model according to the m-th heartbeat type, predicting the data of the m+1 th heartbeat according to the 4 models trained in the third step, respectively marking as yy0, yy1, yy2 and yy3,
performing correlation calculation on yy0, yy1, yy2 and yy3 and the actual m+1st heartbeat data y3 respectively; if yy0 correlation is highest, judging that the m+1st heartbeat is a normal heartbeat; if the yy1 correlation is highest, judging that the m+1st heartbeat is an on-room early heartbeat; if the yy2 correlation is highest, judging that the m+1st heartbeat is an ventricular premature beat; if the yy3 correlation is highest, judging that the (m+1) th heartbeat is other heartbeats; n is a natural number, m is a natural number greater than or equal to 2 and less than n;
and sequentially judging the type of the next heartbeat until the model is selected to judge the nth heartbeat according to the type of the nth heartbeat to be 1 st heartbeat.
Specifically, in step three, as shown in fig. 2, according to the second heartbeat type, a model is selected, if the second heartbeat is a normal heartbeat, a 0-0 model, a 0-1 model, a 0-2 model and a 0-3 model are selected, and if the second heartbeat is an supraventricular heartbeat, a model 1-0 model, a 1-1 model, a 1-2 model and a 1-3 model are selected; and the other is the same. Predicting data of a third heartbeat according to the 4 trained models, respectively marking the data as yy0, yy1, yy2 and yy3, and respectively carrying out correlation calculation on yy0, yy1, yy2 and yy3 and the actual third heartbeat data y 3; if yy0 correlation is highest, diagnosing the third heartbeat as a normal heartbeat; if yy1 correlation is highest, diagnosing the third heartbeat as an supraventricular premature beat; if yy2 correlation is highest, diagnosing the third heartbeat as an ventricular premature beat; if yy3 correlation is highest, the third heartbeat is diagnosed as the other heartbeat. And according to the third heartbeat type, selecting a model to diagnose the fourth heartbeat until all the later heartbeats are diagnosed.
Preferably, in the heartbeat recognition pre-judging model of the present embodiment, in the step three, the heartbeat data with the front data marked as a and the heartbeat data with the rear data marked as B are used to train out the a-B model. Such as: training a 0-0 model by using data of normal heartbeat before and after use; training a 0-1 model by using signals of which the front is normal heartbeat data and the back is indoor early heartbeat data; training a 0-2 model by using data of normal heartbeat in front and early ventricular heartbeat in back; training a 0-3 model by using data with normal heartbeats in front and other types of heartbeats in the latter;
similarly, training a 1-0 model, a 1-1 model, a 1-2 model and a 1-3 model by using corresponding data; 2-0 model, 2-1 model, 2-2 model, 2-3 model, 3-0 model, 3-1 model, 3-2 model, 3-3 model.
Preferably, in the heartbeat recognition pre-judging model of the present embodiment, in step three, a model framework adopts a Seq2Seq architecture: as shown in the figure 3 of the drawings,
the framework consists of two RNN (recurrent neural network) models; the first RNN encodes data through feature learning, encodes data with certain dimension into an information vector c, decodes the data through the information vector c, and finally outputs data with the same dimension as the input dimension;
the model will have previous heartbeat data x= (x) 1 ,x 2 ...x i ...x 350 ) As input, where x i Representing the i-th frame heartbeat data; the latter heartbeat data y= (y) 1 ,y 2 ....y i ....y 350 ) As an output, where y i Representing predicted i-th frame heartbeat data;
h＝(h 0 ,h 1 ...h i ....h 350 ) For hidden state parameters learned during model coding, where h i Representing an ith hidden state parameter; h '= (h' 0 ,h' 1 ...h' i ....h' 350 ) Is a hidden state parameter in the model decoding process, wherein h' i Representing an ith hidden state parameter;
encoding process of model: h is a 0 Is set as zero vector and heartbeat data x for the preset initial hiding state parameter 1 At the same time, input into the first RNN to learn the hidden state parameter h i In combination with heartbeat data x i+1 Sequentially downwards transmitting, and finally outputting an information vector c containing all information;
decoding process of model: the information vector c is input into the second RNN as an initial hidden state, and only the hidden layer state of the last neuron is received without receiving other inputs, and finally the predicted heartbeat data y is output.
Preferably, the heartbeat recognition pre-judging model of the present embodiment, the first time period is 0.3 seconds, and the second time period is 0.4 seconds.
Preferably, in the fourth step, when selecting the model according to the mth heartbeat type, if the mth heartbeat is a normal heartbeat, selecting a 0-0 model, a 0-1 model, a 0-2 model and a 0-3 model; if the mth heartbeat is the early-ventricular heartbeat, selecting a model 1-0, a model 1-1, a model 1-2 and a model 1-3; if the mth heartbeat is the ventricular early heartbeat, selecting a 2-0 model, a 2-1 model, a 2-2 model and a 2-3 model; and if the mth heartbeat is of other heartbeat types, selecting a 3-0 model, a 3-1 model, a 3-2 model and a 3-3 model.
Preferably, in the heartbeat recognition pre-judging model of the present embodiment, in the fourth step, since the analysis result is not affected by removing the first two heartbeats when analyzing the long-time electrocardiographic signal, the computer judging type from the third heartbeat to the n-1 th heartbeat is used as the final result.
Preferably, in the heartbeat recognition pre-judging model of the present embodiment, in the first step, when collecting data, at least 2 tens of thousands of electrocardiograph data including ectopic heartbeat+normal heartbeat are collected; each piece of electrocardiographic data is at least 20 hours long (4800 heartbeats per hour on average).
Preferably, in the heartbeat recognition pre-judging model of the present embodiment, in the first step, when data is collected, the sampling frequency is 500Hz, and if there is data which is not 500Hz, it is firstly resampled to 500Hz.
Preferably, in the heartbeat recognition pre-judging model of the embodiment, in the second step, a band-pass filter of 0.1-100Hz is adopted when filtering is performed.
Preferably, in the heartbeat recognition pre-judging model of the present embodiment, in the second step, a manual or program is adopted in the step of marking each R wave position in the electrocardiograph signal and marking each heartbeat.
With the above-described preferred embodiments according to the present application as a teaching, the related workers can make various changes and modifications without departing from the scope of the technical idea of the present application. The technical scope of the present application is not limited to the contents of the specification, and must be determined according to the scope of claims.
Claims (8)
1. A heartbeat identification pre-judgment model, characterized by comprising the steps of:
step one, collecting data, wherein the data set contains at least 3000 ventricular early heartbeats in total;
step two, filtering the data and cutting off the data into heartbeats: firstly, filtering; secondly, marking the position of each R wave crest in the electrocardiosignal, cutting off the first time period before the occurrence of the R wave crest to the second time period after the occurrence of the R wave crest as a heartbeat, sequentially arranging and numbering the heartbeats in each electrocardiosignal, marking each heartbeat, marking the data of the normal heartbeat as 0, marking the data of the early heartbeat on the room as 1, marking the data of the early heartbeat in the room as 2, and marking the data of other heartbeat types as 3;
step three, training a plurality of models by taking the previous heartbeat data as input and taking the next heartbeat data as output;
judging new data:
after new data are generated, firstly, an R wave crest in an electrocardiogram is extracted by adopting any R wave extraction algorithm, a first time period before the occurrence of the R wave crest to a second time period after the occurrence of the R wave crest are cut off to be used as a heartbeat, n sections of heartbeats are generated, the type of the first heartbeat is not judged, and the judgment mode of the second heartbeat is that if the interval RR2q between the R wave crest position of the second heartbeat and the first R wave crest position is smaller than the interval RR2h between the R wave crest position of the second heartbeat and the third R wave crest position by more than 0.12 seconds and the QRS width is larger than 0.08 seconds, the heart beat is regarded as a ventricular premature heart beat; an supraventricular early heartbeat is considered if RR2q < RR2h-0.12 seconds and QRS width is equal to or less than 0.08 seconds; other conditions are considered normal heartbeats;
the calculation method of the QRS width comprises the following steps: (1) Performing median filtering on the heartbeat data, wherein a window of 0.16 seconds is selected as a median filtering parameter; (2) averaging with a window of 0.04 seconds; the processed heartbeat data is marked as an ecg2, the original electrocardiograph data and a first intersection point q of the ecg2 before the position of the R wave crest, and the interval between the first intersection points s appearing after the R wave crest is marked as the QRS width;
for the judgment of the m+1th heartbeat, firstly, selecting a model according to the type of the m-th heartbeat, predicting the data of the m+1th heartbeat according to the 4 models trained in the third step, respectively marking as yy0, yy1, yy2 and yy3,
performing correlation calculation on yy0, yy1, yy2 and yy3 and the actual m+1st heartbeat data y3 respectively; if yy0 correlation is highest, judging that the m+1st heartbeat is a normal heartbeat; if the yy1 correlation is highest, judging that the m+1st heartbeat is an on-room early heartbeat; if the yy2 correlation is highest, judging that the m+1st heartbeat is an ventricular premature beat; if the yy3 correlation is highest, judging that the (m+1) th heartbeat is other heartbeats; n is a natural number, m is a natural number greater than or equal to 2 and less than n;
sequentially judging the type of the next heartbeat until the model is selected to judge the nth heartbeat according to the type of the nth-1 heartbeat;
in the third step, training an A-B model by using the heartbeat data with the front data marked as A and the heartbeat data with the rear data marked as B;
in step three, the model framework adopts a Seq2Seq architecture:
the framework consists of two RNN models; the first RNN encodes data through feature learning, encodes data with certain dimension into an information vector c, decodes the data through the information vector c, and finally outputs data with the same dimension as the input dimension;
the model will have previous heartbeat data x= (x) 1 ,x 2 ...x i ...x 350 ) As input, where x i Representing the i-th frame heartbeat data; the latter heartbeat data y= (y) 1 ,y 2 ....y i ....y 350 ) As an output, where y i Representing predicted i-th frame heartbeat data;
h＝(h 0 ,h 1 ...h i ....h 350 ) For hidden state parameters learned during model coding, where h i Representing an ith hidden state parameter; h '= (h' 0 ,h' 1 ...h' i ....h' 350 ) Is a hidden state parameter in the model decoding process, wherein h' i Representing an ith hidden state parameter;
encoding process of model: h is a 0 Is set as zero vector and heartbeat data x for the preset initial hiding state parameter 1 At the same time, input into the first RNN to learn the hidden state parameter h i In combination with heartbeat data x i+1 Sequentially downwards transmitting, and finally outputting an information vector c containing all information;
decoding process of model: the information vector c is input into the second RNN as an initial hidden state, and only the hidden layer state of the last neuron is received without receiving other inputs, and finally the predicted heartbeat data y is output.
2. The beat recognition pre-judgment model of claim 1, wherein the first time period is 0.3 seconds and the second time period is 0.4 seconds.
3. The heartbeat identification pre-judgment model of claim 2 wherein in step four, when selecting the model according to the type of the mth heartbeat, if the mth heartbeat is a normal heartbeat, selecting the model 0-0, the model 0-1, the model 0-2, the model 0-3; if the mth heartbeat is the early-ventricular heartbeat, selecting a model 1-0, a model 1-1, a model 1-2 and a model 1-3; if the mth heartbeat is the ventricular early heartbeat, selecting a 2-0 model, a 2-1 model, a 2-2 model and a 2-3 model; and if the mth heartbeat is of other heartbeat types, selecting a 3-0 model, a 3-1 model, a 3-2 model and a 3-3 model.
4. A heartbeat identification pre-judgment model as claimed in claim 3, wherein in step four, the computer judgment type from the third heartbeat to the n-1 th heartbeat is taken as the final result.
5. The model of any one of claims 1-4, wherein in step one, at least 2 tens of thousands of electrocardiographic data including ectopic heart beats + normal heart beats are collected when data is collected; each piece of electrocardiographic data has a length of at least 20 hours.
6. The model of claim 5, wherein in step one, the sampling frequency is 500Hz when collecting data, and if there is data other than 500Hz, it is first resampled to 500Hz.
7. The model of any one of claims 1-4, wherein in step two, a band pass filter of 0.1-100Hz is used for filtering.
8. The heart beat recognition pre-judgment model of claim 7, wherein in the step two, the step of marking each R-wave position in the electrocardiographic signal and marking each heart beat is performed manually or by a program.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110701837.9A CN113598788B (en) | 2021-06-24 | 2021-06-24 | Heartbeat recognition pre-judging model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110701837.9A CN113598788B (en) | 2021-06-24 | 2021-06-24 | Heartbeat recognition pre-judging model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113598788A CN113598788A (en) | 2021-11-05 |
| CN113598788B true CN113598788B (en) | 2024-02-13 |
Family
ID=78303663
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110701837.9A Active CN113598788B (en) | 2021-06-24 | 2021-06-24 | Heartbeat recognition pre-judging model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113598788B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109620206A (en) * | 2018-12-26 | 2019-04-16 | 上海数创医疗科技有限公司 | The recognition methods of atrial fibrillation artificial intelligence and device comprising dystopy heartbeat judgement |
| CN109620207A (en) * | 2018-12-26 | 2019-04-16 | 上海数创医疗科技有限公司 | Based on the ventricular premature beat heartbeat recognition methods and device for improving convolutional neural networks |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107951485B (en) * | 2017-11-27 | 2019-06-11 | 深圳市凯沃尔电子有限公司 | Ambulatory ECG analysis method and apparatus based on artificial intelligence self study |
-
2021
- 2021-06-24 CN CN202110701837.9A patent/CN113598788B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109620206A (en) * | 2018-12-26 | 2019-04-16 | 上海数创医疗科技有限公司 | The recognition methods of atrial fibrillation artificial intelligence and device comprising dystopy heartbeat judgement |
| CN109620207A (en) * | 2018-12-26 | 2019-04-16 | 上海数创医疗科技有限公司 | Based on the ventricular premature beat heartbeat recognition methods and device for improving convolutional neural networks |
Non-Patent Citations (2)
| Title |
|---|
| 一种心律失常分类算法;徐朝浪;;电子世界(第04期);全文 * |
| 基于深度学习和模糊C均值的心电信号分类方法;吴志勇;丁香乾;许晓伟;鞠传香;;自动化学报(第10期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113598788A (en) | 2021-11-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10758139B2 (en) | Automatic method to delineate or categorize an electrocardiogram | |
| US11147500B2 (en) | Electrocardiogram processing system for delineation and classification | |
| EP3878364A1 (en) | An automatic method to delineate or categorize an electrocardiogram | |
| US20100217144A1 (en) | Diagnostic and predictive system and methodology using multiple parameter electrocardiography superscores | |
| US11617528B2 (en) | Systems and methods for reduced lead electrocardiogram diagnosis using deep neural networks and rule-based systems | |
| CN107837082A (en) | Electrocardiogram automatic analysis method and device based on artificial intelligence self study | |
| CN110403600B (en) | Intelligent analysis method and system for paroxysmal atrial fibrillation based on difference time scatter diagram | |
| CN111297349A (en) | Machine learning-based heart rhythm classification system | |
| US11331034B2 (en) | Automatic method to delineate or categorize an electrocardiogram | |
| CN110638430B (en) | Method for building cascade neural network ECG signal arrhythmia classification model | |
| KR20140063100A (en) | Apparatus and methods for remote cardiac disease management | |
| CN113095302B (en) | Depth model for arrhythmia classification, method and device using same | |
| JP2023544242A (en) | Electrocardiogram processing system for detecting and/or predicting cardiac events | |
| JP2003079591A (en) | Method and device for displaying biological information | |
| US11571161B2 (en) | Systems and methods for electrocardiogram diagnosis using deep neural networks and rule-based systems | |
| CN115640507A (en) | Abnormal data screening method based on electrocardio-heart sound joint analysis | |
| CN115062655A (en) | Multi-lead electrocardiosignal analysis method and multi-lead myocardial infarction analysis system | |
| US20220104750A1 (en) | Electrocardiogram processing system for delineation and classification | |
| CN113598788B (en) | Heartbeat recognition pre-judging model | |
| EP4216232A1 (en) | Methods and system for cardiac arrhythmia prediction using transformer-based neural networks | |
| CN110934579A (en) | Deep learning-based electrocardiogram data classification method | |
| CN112842355B (en) | Electrocardiosignal heart beat detection and identification method based on deep learning target detection | |
| US20230190203A1 (en) | Human health risk assessment method | |
| Rayavarapu et al. | Synthesis of ECG signals using Generative Adversarial Networks | |
| Andersson | Acceleration and deceleration and baseline estimation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |