CN110584654A - Multi-mode convolutional neural network-based electrocardiosignal classification method - Google Patents
Multi-mode convolutional neural network-based electrocardiosignal classification method Download PDFInfo
- Publication number
- CN110584654A CN110584654A CN201910955726.3A CN201910955726A CN110584654A CN 110584654 A CN110584654 A CN 110584654A CN 201910955726 A CN201910955726 A CN 201910955726A CN 110584654 A CN110584654 A CN 110584654A
- Authority
- CN
- China
- Prior art keywords
- branch
- electrocardiosignals
- neural network
- convolutional neural
- electrocardio
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 47
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 45
- 230000008569 process Effects 0.000 claims description 24
- 238000005070 sampling Methods 0.000 claims description 21
- 238000000605 extraction Methods 0.000 claims description 18
- 238000013145 classification model Methods 0.000 claims description 13
- 230000009466 transformation Effects 0.000 claims description 9
- 238000001914 filtration Methods 0.000 claims description 7
- 238000007781 pre-processing Methods 0.000 claims description 7
- 238000001514 detection method Methods 0.000 claims description 6
- 230000008014 freezing Effects 0.000 claims description 6
- 238000007710 freezing Methods 0.000 claims description 6
- 230000004927 fusion Effects 0.000 claims description 6
- 238000002372 labelling Methods 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 238000004806 packaging method and process Methods 0.000 claims description 3
- 230000011218 segmentation Effects 0.000 claims description 3
- 238000000844 transformation Methods 0.000 claims description 3
- 230000006870 function Effects 0.000 abstract description 20
- 230000000694 effects Effects 0.000 abstract description 2
- 238000013528 artificial neural network Methods 0.000 description 3
- 208000019622 heart disease Diseases 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000747 cardiac effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000001228 spectrum 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]
-
- 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
- A61B5/366—Detecting abnormal QRS complex, e.g. widening
-
- 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
- A61B5/7203—Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal
-
- 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
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
- A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Biophysics (AREA)
- General Health & Medical Sciences (AREA)
- Cardiology (AREA)
- Veterinary Medicine (AREA)
- Public Health (AREA)
- Animal Behavior & Ethology (AREA)
- Surgery (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- Pathology (AREA)
- Biomedical Technology (AREA)
- Heart & Thoracic Surgery (AREA)
- Medical Informatics (AREA)
- Signal Processing (AREA)
- Psychiatry (AREA)
- Physiology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Fuzzy Systems (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The invention discloses an electrocardiosignal classification method based on a multi-mode convolutional neural network, which is characterized in that a multi-mode convolutional neural network which is composed of three branches and has medical index attention is obtained by training a large number of electrocardio data sets with labeled classification results, and the multi-mode convolutional neural network is respectively a first branch, a second branch and a third branch, wherein the first branch is used for processing pre-extracted electrocardio characteristic information which can effectively assist the multi-mode convolutional neural network in final prediction; the first branch is used for processing original electrocardiosignals; the third branch circuit fuses the outputs of the first branch circuit and the second branch circuit and classifies the types of the input electrocardiosignals. The method of the invention combines the loss functions of the three branches, adopts a non-end-to-end training mode, and introduces the attention of the traditional medical indexes with important values, thereby accurately predicting the category of the electrocardiosignals and obtaining good electrocardio classification effect.
Description
Technical Field
The invention relates to the technical field of electrocardiosignal classification, in particular to an electrocardiosignal classification method based on a multi-mode convolutional neural network.
Background
The classification result of the electrocardiosignals is an important auxiliary means and reference information for doctors to diagnose heart diseases, and with the rise of deep learning, methods for classifying the electrocardiosignals by using a neural network are successively appeared at present, and breakthrough results are obtained. However, most of the existing methods for classifying electrocardiosignals by using a neural network directly analyze the electrocardiosignals, but the analysis indexes developed by the current medicine are abandoned, and the whole model is a black box for people and cannot meet the actual requirement of providing auxiliary diagnosis information for doctors.
Disclosure of Invention
The invention provides an electrocardiosignal classification method based on a multi-mode convolution neural network, which aims to solve the problem that the existing electrocardiosignal classification method is difficult to realize high accuracy.
In order to achieve the above purpose, the technical means adopted is as follows:
an electrocardiosignal classification method based on a multi-mode convolutional neural network comprises the following steps:
s1, acquiring an original electrocardiosignal and preprocessing the electrocardiosignal; the electrocardiosignals are divided into a training set and a target set, the electrocardiosignals of the training set are electrocardiosignals of labeled types, and the electrocardiosignals of the target set are electrocardiosignals of unlabeled types;
s2, performing electrocardio medical feature extraction and downsampling processing on the preprocessed electrocardiosignals respectively;
s3, constructing a multi-mode convolutional neural network concerned by medical indexes, inputting data which comprises the training set and is subjected to electrocardio medical feature extraction and labeling corresponding to the type of the training set, and data subjected to down-sampling processing, and outputting the data as a classification result of electrocardiosignals;
s4, defining a loss function according to a training target of the multi-mode convolutional neural network, and solving an optimal value by using a back propagation algorithm to obtain a final electrocardiosignal classification model;
s5, classifying the input target signals through the electrocardiosignal classification model to obtain a classification result of the target signals; the input target signal comprises data of the target set after electrocardio medical characteristic extraction and data after down-sampling processing.
In the scheme, a multi-mode convolutional neural network with medical index attention is designed, a large amount of electrocardio data with labeled classification results are input for end-to-end training, and the pre-extracted electrocardio characteristic information effectively assists the network to carry out final prediction, so that the category of the electrocardio signal is accurately predicted, and the electrocardio signal can be used as reference information for judging subsequent heart diseases.
The first branch processes pre-extracted electrocardio characteristic information for being used as a medical index to pay attention, so that the network is effectively assisted to carry out electrocardio detection; the second branch circuit processes original electrocardiosignals; the third branch circuit fuses the outputs of the first branch circuit and the second branch circuit and classifies the target electrocardiosignals.
Preferably, the preprocessing the electrocardiographic signal in step S1 specifically includes:
s11, performing analog-to-digital conversion on the obtained electrocardiosignals to obtain electrocardio digital signals;
s12, carrying out segmentation processing on the electrocardio digital signals to obtain a plurality of electrocardio digital signals with the duration of L seconds; wherein L is less than or equal to 60 seconds;
s13, filtering each section of electrocardio digital signal; in the preferred scheme, the frequency spectrum range of the filtering is between 3Hz and 45Hz, and the filtering is used for filtering noise which is a central electric signal of the filtering, so as to further guarantee the accuracy of subsequent detection;
preferably, the medical electrocardiographic features extracted in step S2 include: 4 groups of characteristics of R wave amplitude, RR interval, P wave amplitude and PR interval.
Preferably, after the down-sampling process in step S2, the frequency of the electrocardiographic signal is reduced to 100 Hz. In the preferred scheme, the electrocardiosignal is subjected to down-sampling processing, so that the data volume can be simplified, and the calculation efficiency can be improved.
Preferably, the step S2 of respectively performing electrocardiographic medical feature extraction on the preprocessed electrocardiographic signals includes:
s21, carrying out QRS wave group detection on the preprocessed electrocardiosignals by using a Pan-Tompkins algorithm, and positioning the position of an R wave;
s22, retrieving the maximum value of the electrocardio amplitude value in a period of time before the R wave occurs according to the R wave position positioned in the step S21, namely the P wave position; because the PR interval of the electrocardiosignals is usually between 0.12s and 0.20s, the P wave position of the electrocardiosignals can be extracted by searching the maximum value of the electrocardio amplitude in a period of time before the R wave occurs;
s23, according to the R wave position and the P wave position located in the steps S21 and S22, the amplitude of the R wave and the amplitude of the P wave in the electrocardiosignals are respectively determined, and 4 groups of characteristics of the R wave amplitude, the RR interval, the P wave amplitude and the PR interval are obtained through calculation;
s24, aligning 4 groups of calculated features, wherein the alignment is performed by taking the length of the longest segment as a standard and performing self-filling data alignment until the lengths of the 4 groups of features are consistent;
and S25, splicing and packaging the 4 groups of feature data after being completed.
Preferably, the multi-modal convolutional neural network with medical index attention described in step S3 includes three branches, where a first branch is used to process the training set for performing the electrocardiographic feature extraction and label the training set with a corresponding type, a second branch is used to process data of the training set for performing down-sampling processing, and a third branch is used to perform feature fusion on information of the first branch and the second branch and perform electrocardiographic signal classification.
Preferably, the step S4 of defining the loss function according to the training target of the multi-modal convolutional neural network includes:
s41, equating the training process of the multi-mode convolutional neural network to be the following minimum solving process:
wherein P represents a probability; x ═ X1,...,XNRepresenting data after the down-sampling processing is carried out on the training set; t ═ T1,...,TN-labels representing the training set; y represents data of the training set after the electrocardio medical characteristics are extracted; z represents the fusion characteristic of the first branch and the second branchPerforming sign; w is aa,wb,wcNetwork parameters respectively corresponding to three branches in the multi-mode convolutional neural network;
s42, according to the Bayes formula and the principle of conditional probability and joint distribution, P (X, T) and network parameters waIndependently of one another, the following formula is obtained:
wherein X ═ { X ═ X1,...,XNRepresenting data after the down-sampling processing is carried out on the training set; t ═ T1,...,TN-labels representing the training set; w is aaNetwork parameters representing a first branch in the multi-modal convolutional neural network;
wherein P (X | w)a) Taken as a constant, the following equation is obtained:
P(wa|X,T)＝P(T|X,wa)P(wa)
same pair P (w)bI Y, T) and P (w)c| Z, T) are transformed as above, the solving process of the maximum value of step S41 is converted into:
s43. if i is not equal to j, Xi，XjIndependently of one another, and P (w) is a Gaussian distribution satisfying a mean value of 0The process of obtaining the w-optimum solution is a process of minimizing the following loss function:
whereinRepresents the total loss function;
s44, carrying out logarithmic transformation on the loss function of the step S43 to obtain:
wherein gamma isaIs a constant;
s45, dividing the loss function subjected to logarithmic transformation in the step S44 intoThree portions corresponding to the three branches, respectively, whereinRepresented by the formula:
whereinAfter cross entropy expression, the following formula is obtained:
wherein etaT,ηaIs a parameter; t isi,jThe presence of a real label is indicated,representing a class predicted by the first branch;
same reason pairRespectively making the above transformations, thenThe overall loss function for the three sections is:
wherein α, β, χ represent three parameters, respectively.
Preferably, the step S4 of solving the optimal value by using a back propagation algorithm to obtain the final classification model of the electrocardiographic signal specifically includes the following steps:
s46, freezing network parameters of the second branch and the third branch, setting the loss weights of the second branch and the third branch to be 0, and updating the weight of the first branch by using a gradient descent and back propagation algorithm on the total loss function obtained in the step S45 until the network parameters of the first branch are converged;
s47, freezing network parameters of the first branch, setting the loss weight of the first branch to 0, resetting the loss weights of the second branch and the third branch, and performing weight updating on the second branch and the third branch by using a gradient descent and back propagation algorithm on the total loss function obtained in the step S45 until the multi-mode convolutional neural network is converged to obtain a final electrocardiosignal classification model.
Compared with the prior art, the technical scheme of the invention has the beneficial effects that:
the invention provides an electrocardiosignal classification method based on a multi-mode convolutional neural network, which is characterized in that the multi-mode convolutional neural network concerned by medical indexes is designed, wherein a first branch circuit processes pre-extracted electrocardio characteristic information which is used as the medical indexes to be concerned, so that the network is effectively assisted to carry out electrocardio detection; the second branch circuit processes original electrocardiosignals; the third branch circuit fuses the outputs of the first branch circuit and the second branch circuit and classifies the target electrocardiosignals. The method of the invention combines the loss functions of the three branches, adopts a non-end-to-end training mode, and introduces the attention of the traditional medical indexes with important values, thereby accurately predicting the category of the electrocardiosignals and obtaining good electrocardio classification effect.
Drawings
Fig. 1 is a general flow chart of the present invention.
Fig. 2 is a flowchart of a process of classifying the cardiac signal of the target signal.
Detailed Description
The drawings are for illustrative purposes only and are not to be construed as limiting the patent;
for the purpose of better illustrating the embodiments, certain features of the drawings may be omitted, enlarged or reduced, and do not represent the size of an actual product;
it will be understood by those skilled in the art that certain well-known structures in the drawings and descriptions thereof may be omitted.
The technical solution of the present invention is further described below with reference to the accompanying drawings and examples.
Example 1
An electrocardiosignal classification method based on a multi-mode convolutional neural network is shown in fig. 1, and comprises the following steps:
s1, acquiring an original electrocardiosignal and preprocessing the electrocardiosignal; the electrocardiosignals are divided into a training set and a target set, the electrocardiosignals of the training set are electrocardiosignals of labeled types, and the electrocardiosignals of the target set are electrocardiosignals of unlabeled types;
s2, performing electrocardio medical feature extraction and downsampling processing on the preprocessed electrocardiosignals respectively;
s3, constructing a multi-mode convolutional neural network concerned by medical indexes, inputting data which comprises the training set and is subjected to electrocardio medical feature extraction and labeling corresponding to the type of the training set, and data subjected to down-sampling processing, and outputting the data as a classification result of electrocardiosignals;
s4, defining a loss function according to a training target of the multi-mode convolutional neural network, and solving an optimal value by using a back propagation algorithm to obtain a final electrocardiosignal classification model;
s5, classifying the input target signals through the electrocardiosignal classification model to obtain a classification result of the target signals; the input target signal comprises data of the target set after electrocardio medical characteristic extraction and data after down-sampling processing.
Example 2
An electrocardiosignal classification method based on a multi-mode convolutional neural network is shown in fig. 1, and comprises the following steps:
s1, acquiring an original electrocardiosignal and preprocessing the electrocardiosignal; the electrocardiosignals are divided into a training set and a target set, the electrocardiosignals of the training set are electrocardiosignals of labeled types, and the electrocardiosignals of the target set are electrocardiosignals of unlabeled types;
the preprocessing of the electrocardiosignals specifically comprises the following steps:
s11, performing analog-to-digital conversion on the obtained electrocardiosignals to obtain electrocardio digital signals;
s12, carrying out segmentation processing on the electrocardio digital signals to obtain a plurality of electrocardio digital signals with the duration of L seconds; wherein L is less than or equal to 60 seconds;
and S13, filtering each section of electrocardio digital signal.
S2, performing electrocardio medical feature extraction and downsampling processing on the preprocessed electrocardiosignals respectively; the extracted electrocardio-medical characteristics comprise: 4 groups of characteristics of R wave amplitude, RR interval, P wave amplitude and PR interval; the frequency of the electrocardiosignals after the down-sampling treatment is reduced to 100 Hz;
the specific steps of respectively carrying out electrocardio medical characteristic extraction on the preprocessed electrocardiosignals comprise:
s21, carrying out QRS wave group detection on the preprocessed electrocardiosignals by using a Pan-Tompkins algorithm, and positioning the position of an R wave;
s22, retrieving the maximum value of the electrocardio amplitude value in a period of time before the R wave occurs according to the R wave position positioned in the step S21, namely the P wave position;
s23, according to the R wave position and the P wave position located in the steps S21 and S22, the amplitude of the R wave and the amplitude of the P wave in the electrocardiosignals are respectively determined, and 4 groups of characteristics of the R wave amplitude, the RR interval, the P wave amplitude and the PR interval are obtained through calculation;
s24, aligning 4 groups of calculated features, wherein the alignment is performed by taking the length of the longest segment as a standard and performing self-filling data alignment until the lengths of the 4 groups of features are consistent; the data alignment is carried out in a self-filling mode, namely, firstly, a segment with the longest length in 4 groups of characteristic data is obtained as a standard, then, the data of the part with short sequences and insufficient is copied, the data of the part with short sequences is filled in the missing part, and the data is copied from the first point until the length of the segment is equal to that of the longest segment;
and S25, splicing and packaging the 4 groups of feature data after being completed.
S3, constructing a multi-mode convolutional neural network concerned by medical indexes, wherein the network comprises three branches, a first branch is used for processing the training set to perform electrocardio medical characteristic extraction and labeling corresponding to the type, a second branch is used for processing data of the training set for down-sampling processing, a third branch is used for performing characteristic fusion on the information of the first branch and the second branch and classifying electrocardiosignals, the input of the network comprises data of the training set subjected to electrocardio medical characteristic extraction and labeling corresponding to the type and data subjected to down-sampling processing, and the output of the network is a classification result of the electrocardiosignals;
s4, defining a loss function according to a training target of the multi-mode convolutional neural network, and solving an optimal value by using a back propagation algorithm to obtain a final electrocardiosignal classification model; the method specifically comprises the following steps:
s41, equating the training process of the multi-mode convolutional neural network to be the following minimum solving process:
wherein P represents a probability; x ═ X1,...,XNRepresenting data after the down-sampling processing is carried out on the training set; t ═ T1,...,TN-labels representing the training set; y represents data of the training set after the electrocardio medical characteristics are extracted; z represents the fusion characteristic of the first branch and the second branch; w is aa,wb,wcNetwork parameters respectively corresponding to three branches in the multi-mode convolutional neural network;
s42, according to a Bayes formula and the principles of conditional probability and joint distribution,and P (X, T) and the network parameter waIndependently of one another, the following formula is obtained:
wherein X ═ { X ═ X1,...,XNRepresenting data after the down-sampling processing is carried out on the training set; t ═ T1,...,TN-labels representing the training set; w is aaNetwork parameters representing a first branch in the multi-modal convolutional neural network;
wherein P (X | w)a) Taken as a constant, the following equation is obtained:
P(wa|X,T)＝P(T|X,wa)P(wa)
same pair P (w)bI Y, T) and P (w)c| Z, T) are transformed as above, the solving process of the maximum value of step S41 is converted into:
s43. if i is not equal to j, Xi，XjIndependently of one another, and P (w) is a Gaussian distribution satisfying a mean value of 0The process of obtaining the w-optimum solution is a process of minimizing the following loss function:
whereinRepresents the total loss function;
s44, carrying out logarithmic transformation on the loss function of the step S43 to obtain:
wherein gamma isaIs a constant;
s45, dividing the loss function subjected to logarithmic transformation in the step S44 intoThree portions corresponding to the three branches, respectively, whereinRepresented by the formula:
whereinAfter cross entropy expression, the following formula is obtained:
wherein etaT,ηaIs a parameter; t isi,jThe presence of a real label is indicated,representing a class predicted by the first branch;
same reason pairRespectively making the above transformations, thenThe overall loss function for the three sections is:
wherein alpha, beta and chi respectively represent three parameters;
s46, freezing network parameters of the second branch and the third branch, setting the loss weights of the second branch and the third branch to be 0, and updating the weight of the first branch by using a gradient descent and back propagation algorithm on the total loss function obtained in the step S45 until the network parameters of the first branch are converged;
s47, freezing network parameters of the first branch, setting the loss weight of the first branch to be 0, resetting the loss weights of the second branch and the third branch, and performing weight updating on the second branch and the third branch by using a gradient descent and back propagation algorithm on the total loss function obtained in the step S45 until the multi-mode convolutional neural network is converged to obtain a final electrocardiosignal classification model;
s5, as shown in the figure 2, classifying the input target signals through the electrocardiosignal classification model to obtain a classification result of the target signals; the input target signal comprises data of the target set after electrocardio medical characteristic extraction and data after down-sampling processing.
The terms describing positional relationships in the drawings are for illustrative purposes only and are not to be construed as limiting the patent;
it should be understood that the above-described embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. Other variations and modifications will be apparent to persons skilled in the art in light of the above description. And are neither required nor exhaustive of all embodiments. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention should be included in the protection scope of the claims of the present invention.
Claims (8)
1. The method for classifying the electrocardiosignals based on the multi-mode convolutional neural network is characterized by comprising the following steps of:
s1, acquiring an original electrocardiosignal and preprocessing the electrocardiosignal; the electrocardiosignals are divided into a training set and a target set, the electrocardiosignals of the training set are electrocardiosignals of labeled types, and the electrocardiosignals of the target set are electrocardiosignals of unlabeled types;
s2, performing electrocardio medical feature extraction and downsampling processing on the preprocessed electrocardiosignals respectively;
s3, constructing a multi-mode convolutional neural network concerned by medical indexes, inputting data which comprises the training set and is subjected to electrocardio medical feature extraction and labeling corresponding to the type of the training set, and data subjected to down-sampling processing, and outputting the data as a classification result of electrocardiosignals;
s4, defining a loss function according to a training target of the multi-mode convolutional neural network, and solving an optimal value by using a back propagation algorithm to obtain a final electrocardiosignal classification model;
s5, classifying the input target signals through the electrocardiosignal classification model to obtain a classification result of the target signals; the input target signal comprises data of the target set after electrocardio medical characteristic extraction and data after down-sampling processing.
2. The multi-modal convolutional neural network based electrocardiograph signal classification method according to claim 1, wherein the preprocessing the electrocardiograph signal in step S1 specifically comprises:
s11, performing analog-to-digital conversion on the obtained electrocardiosignals to obtain electrocardio digital signals;
s12, carrying out segmentation processing on the electrocardio digital signals to obtain a plurality of electrocardio digital signals with the duration of L seconds; wherein L is less than or equal to 60 seconds;
and S13, filtering each section of electrocardio digital signal.
3. The multi-modal convolutional neural network based electrocardiograph signal classification method according to claim 1, wherein the electrocardiograph medical features extracted in the step S2 include: 4 groups of characteristics of R wave amplitude, RR interval, P wave amplitude and PR interval.
4. The method for classifying electrocardiosignals based on the multi-mode convolutional neural network as claimed in claim 1, wherein the frequency of the electrocardiosignals is reduced to 100Hz after the down-sampling process in step S2.
5. The multi-modal convolutional neural network-based classification method for electrocardiograph signals according to claim 3, wherein the specific steps of performing electrocardiograph medical feature extraction on the preprocessed electrocardiograph signals in step S2 respectively comprise:
s21, carrying out QRS wave group detection on the preprocessed electrocardiosignals by using a Pan-Tompkins algorithm, and positioning the position of an R wave;
s22, retrieving the maximum value of the electrocardio amplitude value in a period of time before the R wave occurs according to the R wave position positioned in the step S21, namely the P wave position;
s23, according to the R wave position and the P wave position located in the steps S21 and S22, the amplitude of the R wave and the amplitude of the P wave in the electrocardiosignals are respectively determined, and 4 groups of characteristics of the R wave amplitude, the RR interval, the P wave amplitude and the PR interval are obtained through calculation;
s24, aligning 4 groups of calculated features, wherein the alignment is performed by taking the length of the longest segment as a standard and performing self-filling data alignment until the lengths of the 4 groups of features are consistent;
and S25, splicing and packaging the 4 groups of feature data after being completed.
6. The method for classifying electrocardiosignals based on the multi-modal convolutional neural network as claimed in claim 1, wherein the multi-modal convolutional neural network with medical index attention in step S3 comprises three branches, wherein a first branch is used for processing the training set to perform the extraction of the electrocardio-medical features and labeling the electrocardio-medical features corresponding to the types of the electrocardio-medical features, a second branch is used for processing the data of the training set for down-sampling processing, and a third branch is used for performing feature fusion on the information of the first branch and the second branch and classifying the electrocardiosignals.
7. The method for classifying electrocardiosignals based on the multi-modal convolutional neural network as claimed in claim 6, wherein the step S4 of defining the loss function according to the training target of the multi-modal convolutional neural network comprises:
s41, equating the training process of the multi-mode convolutional neural network to be the following minimum solving process:
wherein P represents a probability; x ═ X1,...,XNRepresenting data after the down-sampling processing is carried out on the training set; t ═ T1,...,TN-labels representing the training set; y represents data of the training set after the electrocardio medical characteristics are extracted; z represents the fusion characteristic of the first branch and the second branch; w is aa,wb,wcNetwork parameters respectively corresponding to three branches in the multi-mode convolutional neural network;
s42, according to the Bayes formula and the principle of conditional probability and joint distribution, P (X, T) and network parameters waIndependently of one another, the following formula is obtained:
wherein X ═ { X ═ X1,...,XNRepresenting data after the down-sampling processing is carried out on the training set; t ═ T1,...,TN-labels representing the training set; w is aaNetwork parameters representing a first branch in the multi-modal convolutional neural network;
wherein P (X | w)a) Taken as a constant, the following equation is obtained:
P(wa|X,T)＝P(T|X,wa)P(wa)
same pair P (w)bI Y, T) and P (w)c| Z, T) are transformed as above, the solving process of the maximum value of step S41 is converted into:
s43. if i is not equal to j, Xi，XjIndependently of each other, and P (w) isGaussian distribution with mean 0The process of obtaining the w-optimum solution is a process of minimizing the following loss function:
whereinRepresents the total loss function;
s44, carrying out logarithmic transformation on the loss function of the step S43 to obtain:
wherein gamma isaIs a constant;
s45, dividing the loss function subjected to logarithmic transformation in the step S44 intoThree portions corresponding to the three branches, respectively, whereinRepresented by the formula:
whereinAfter cross entropy expression, the following formula is obtained:
wherein etaT,ηaIs a parameter; t isi,jThe presence of a real label is indicated,representing a class predicted by the first branch;
same reason pairRespectively making the above transformations, thenThe overall loss function for the three sections is:
wherein α, β, χ represent three parameters, respectively.
8. The method for classifying electrocardiosignals based on the multi-modal convolutional neural network as claimed in claim 7, wherein the step S4 of performing the optimal value solution by using the back propagation algorithm to obtain the final electrocardiosignal classification model specifically comprises the following steps:
s46, freezing network parameters of the second branch and the third branch, setting the loss weights of the second branch and the third branch to be 0, and updating the weight of the first branch by using a gradient descent and back propagation algorithm on the total loss function obtained in the step S45 until the network parameters of the first branch are converged;
s47, freezing network parameters of the first branch, setting the loss weight of the first branch to 0, resetting the loss weights of the second branch and the third branch, and performing weight updating on the second branch and the third branch by using a gradient descent and back propagation algorithm on the total loss function obtained in the step S45 until the multi-mode convolutional neural network is converged to obtain a final electrocardiosignal classification model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910955726.3A CN110584654A (en) | 2019-10-09 | 2019-10-09 | Multi-mode convolutional neural network-based electrocardiosignal classification method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910955726.3A CN110584654A (en) | 2019-10-09 | 2019-10-09 | Multi-mode convolutional neural network-based electrocardiosignal classification method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110584654A true CN110584654A (en) | 2019-12-20 |
Family
ID=68866248
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910955726.3A Pending CN110584654A (en) | 2019-10-09 | 2019-10-09 | Multi-mode convolutional neural network-based electrocardiosignal classification method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110584654A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112568908A (en) * | 2020-12-14 | 2021-03-30 | 上海数创医疗科技有限公司 | Electrocardiogram waveform positioning and classifying model device adopting multi-scale visual field depth learning |
| CN113180690A (en) * | 2021-04-21 | 2021-07-30 | 上海数创医疗科技有限公司 | Electrocardio diagnosis model and electrocardio detection device |
| CN113384277A (en) * | 2020-02-26 | 2021-09-14 | 京东方科技集团股份有限公司 | Electrocardiogram data classification method and classification system |
| CN114818837A (en) * | 2022-06-29 | 2022-07-29 | 电子科技大学 | Electroencephalogram signal intelligent processing circuit based on multistage neural network and block calculation |
| CN114970608A (en) * | 2022-05-06 | 2022-08-30 | 中国科学院自动化研究所 | Human-computer interaction method and system based on electro-oculogram signals |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105286853A (en) * | 2015-11-17 | 2016-02-03 | 安徽华米信息科技有限公司 | Disease detection method and device based on wearable device and wearable device |
| CN106529402A (en) * | 2016-09-27 | 2017-03-22 | 中国科学院自动化研究所 | Multi-task learning convolutional neural network-based face attribute analysis method |
| CN106980854A (en) * | 2017-03-29 | 2017-07-25 | 珠海习悦信息技术有限公司 | Number-plate number recognition methods, device, storage medium and processor |
| CN107811626A (en) * | 2017-09-10 | 2018-03-20 | 天津大学 | A kind of arrhythmia classification method based on one-dimensional convolutional neural networks and S-transformation |
| CN109271522A (en) * | 2018-11-20 | 2019-01-25 | 深圳大学 | Comment sensibility classification method and system based on depth mixed model transfer learning |
| CN109684906A (en) * | 2018-05-31 | 2019-04-26 | 北京林业大学 | The method of detection red turpentine beetle based on deep learning |
-
2019
- 2019-10-09 CN CN201910955726.3A patent/CN110584654A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105286853A (en) * | 2015-11-17 | 2016-02-03 | 安徽华米信息科技有限公司 | Disease detection method and device based on wearable device and wearable device |
| CN106529402A (en) * | 2016-09-27 | 2017-03-22 | 中国科学院自动化研究所 | Multi-task learning convolutional neural network-based face attribute analysis method |
| CN106980854A (en) * | 2017-03-29 | 2017-07-25 | 珠海习悦信息技术有限公司 | Number-plate number recognition methods, device, storage medium and processor |
| CN107811626A (en) * | 2017-09-10 | 2018-03-20 | 天津大学 | A kind of arrhythmia classification method based on one-dimensional convolutional neural networks and S-transformation |
| CN109684906A (en) * | 2018-05-31 | 2019-04-26 | 北京林业大学 | The method of detection red turpentine beetle based on deep learning |
| CN109271522A (en) * | 2018-11-20 | 2019-01-25 | 深圳大学 | Comment sensibility classification method and system based on depth mixed model transfer learning |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113384277A (en) * | 2020-02-26 | 2021-09-14 | 京东方科技集团股份有限公司 | Electrocardiogram data classification method and classification system |
| CN112568908A (en) * | 2020-12-14 | 2021-03-30 | 上海数创医疗科技有限公司 | Electrocardiogram waveform positioning and classifying model device adopting multi-scale visual field depth learning |
| CN113180690A (en) * | 2021-04-21 | 2021-07-30 | 上海数创医疗科技有限公司 | Electrocardio diagnosis model and electrocardio detection device |
| CN113180690B (en) * | 2021-04-21 | 2022-12-09 | 上海数创医疗科技有限公司 | Electrocardio diagnosis model judgment method and electrocardio detection device |
| CN114970608A (en) * | 2022-05-06 | 2022-08-30 | 中国科学院自动化研究所 | Human-computer interaction method and system based on electro-oculogram signals |
| CN114818837A (en) * | 2022-06-29 | 2022-07-29 | 电子科技大学 | Electroencephalogram signal intelligent processing circuit based on multistage neural network and block calculation |
| CN114818837B (en) * | 2022-06-29 | 2022-10-14 | 电子科技大学 | Electroencephalogram signal intelligent processing circuit based on multistage neural network and block calculation |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110584654A (en) | Multi-mode convolutional neural network-based electrocardiosignal classification method | |
| CN111772619B (en) | Heart beat identification method based on deep learning, terminal equipment and storage medium | |
| CN110522444B (en) | Electrocardiosignal identification and classification method based on Kernel-CNN | |
| CN110992351B (en) | sMRI image classification method and device based on multi-input convolution neural network | |
| CN112667080A (en) | Electroencephalogram signal unmanned platform intelligent control method based on deep convolution countermeasure network | |
| CN111626114B (en) | Electrocardiosignal arrhythmia classification system based on convolutional neural network | |
| CN110638430B (en) | Method for building cascade neural network ECG signal arrhythmia classification model | |
| CN112043260B (en) | Electrocardiogram classification method based on local mode transformation | |
| CN111956208B (en) | ECG signal classification method based on ultra-lightweight convolutional neural network | |
| CN113679393B (en) | ECG data feature generation model based on contrast predictive coding | |
| CN115294075A (en) | OCTA image retinal vessel segmentation method based on attention mechanism | |
| US20230225663A1 (en) | Method for predicting multi-type electrocardiogram heart rhythms based on graph convolution | |
| CN114417836A (en) | Deep learning-based Chinese electronic medical record text semantic segmentation method | |
| CN115908947A (en) | Multi-modal periodontal image identification method and system based on inverse deductive learning | |
| CN114881105A (en) | Sleep staging method and system based on transformer model and contrast learning | |
| CN116269212A (en) | Multi-mode sleep stage prediction method based on deep learning | |
| CN115062655A (en) | Multi-lead electrocardiosignal analysis method and multi-lead myocardial infarction analysis system | |
| CN113855042B (en) | Sequence labeling-based multi-lead electrocardiosignal classification method integrating depth and medical features | |
| CN111543985A (en) | Brain control hybrid intelligent rehabilitation method based on novel deep learning model | |
| CN112990270B (en) | Automatic fusion method of traditional feature and depth feature | |
| CN112957052B (en) | Multi-lead electrocardiosignal classification method based on NLF-CNN lead fusion depth network | |
| CN110613445A (en) | DWNN framework-based electrocardiosignal identification method | |
| CN114862865B (en) | Vessel segmentation method and system based on multi-view coronary angiography sequence image | |
| CN116211315A (en) | Single-lead electrocardiosignal auxiliary diagnosis method and diagnosis terminal | |
| CN115836849A (en) | Arrhythmia classification method based on comparison test time adaptation |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191220 |