CN112883803A - Deep learning-based electrocardiosignal classification method and device and storage medium - Google Patents
Deep learning-based electrocardiosignal classification method and device and storage medium Download PDFInfo
- Publication number
- CN112883803A CN112883803A CN202110076938.1A CN202110076938A CN112883803A CN 112883803 A CN112883803 A CN 112883803A CN 202110076938 A CN202110076938 A CN 202110076938A CN 112883803 A CN112883803 A CN 112883803A
- Authority
- CN
- China
- Prior art keywords
- electrocardiosignal
- data
- original
- electrocardiosignals
- deep learning
- 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.)
- Granted
Links
Images
Classifications
-
- 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
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Abstract
The invention relates to an electrocardiosignal classification method and device based on deep learning and a computer readable storage medium, wherein the method comprises the following steps: collecting original electrocardiosignals, carrying out R wave detection on the original electrocardiosignals to obtain R point coordinates, segmenting the original electrocardiosignals according to the R point coordinate graph and the total heart beat number in the original electrocardiosignals to obtain electrocardiosignal data segments, and forming a data set according to the electrocardiosignal data segments; constructing a neural network model, and training the neural network model by using the data set to obtain the neural network model for electrocardiosignal classification; and re-collecting the original electrocardiosignals, and classifying the original electrocardiosignals by the neural network model for classifying the electrocardiosignals to obtain the types of the original electrocardiosignals. The electrocardiosignal classification method based on deep learning improves the classification precision of the electrocardiosignals.
Description
Technical Field
The invention relates to the technical field of electrocardiosignal classification, in particular to an electrocardiosignal classification method and device based on deep learning and a computer readable storage medium.
Background
The deep learning method has the advantages of higher speed and higher precision compared with the traditional signal processing method for the classification of the electrocardiosignals, the classification of the electrocardiosignals can provide reference for the diagnosis of doctors, and the auxiliary diagnosis can reduce the workload of the doctors in the presence of a large amount of data to be diagnosed.
The current classification method for electrocardiosignals mainly comprises a traditional signal processing method and a deep learning method. The traditional digital signal processing method classifies the electrocardiosignal data, the parameters, the threshold values and the like are required to be set to manually extract the characteristics of the signals, the parameters, the threshold values and the like have practical physical meanings, so the classification method has strong interpretability, but the original signals often have the interferences of baseline drift, noise and the like due to the fixed numerical values of the parameters, the threshold values and the like, and the manual characteristic extraction method has poor robustness, so the identification effect is far lower than that of a deep learning method.
The existing deep learning method not only can greatly add useless information into data and disperse the attention of a neural network to key parts in the data, but also can increase the size of a data set so as to slow down the training speed, and also can lose important characteristics in original data, reduce the recognition precision of the neural network and enable the classification precision of electrocardiosignals to be lower.
Disclosure of Invention
In view of the above, there is a need to provide an electrocardiographic signal classification method and apparatus based on deep learning, and a computer-readable storage medium, so as to solve the problem of low classification accuracy of electrocardiographic signals in the prior art.
The invention provides an electrocardiosignal classification method based on deep learning, which comprises the following steps:
collecting original electrocardiosignals, carrying out R wave detection on the original electrocardiosignals to obtain R point coordinates, segmenting the original electrocardiosignals according to the R point coordinate graph and the total heart beat number in the original electrocardiosignals to obtain electrocardiosignal data segments, and forming a data set according to the electrocardiosignal data segments;
constructing a neural network model, and training the neural network model by using the data set to obtain the neural network model for electrocardiosignal classification;
and re-collecting the original electrocardiosignals, and classifying the original electrocardiosignals by the neural network model for classifying the electrocardiosignals to obtain the types of the original electrocardiosignals.
Further, segmenting the original electrocardiosignal according to the R point coordinate and the total heart beat number in the original electrocardiosignal to obtain an electrocardiosignal data segment, which specifically comprises:
and segmenting the original electrocardiosignal between two adjacent R point coordinates to obtain N sections of electrocardiosignal data sections, wherein N is S/T and is rounded up, S is the total heart beat number in the original electrocardiosignal, and T is the heart beat number in the electrocardiosignal data sections.
Further, forming a data set according to the electrocardiographic signal data segment, specifically comprising:
acquiring the maximum data length of all the electrocardiosignal data segments, carrying out zero filling on the electrocardiosignal data segment codes with the data length not being the maximum data length to the maximum data length, and forming a data set according to the electrocardiosignal data segment codes after the zero filling of the codes and the electrocardiosignal data segments without the zero filling of the codes.
Further, forming a data set according to the encoded and zero-padded cardiac signal data segment and the uncoded and zero-padded cardiac signal data segment, specifically comprising:
the electrocardio type is used as a label, whether disease characteristics are contained is used as marks of an electrocardiosignal data segment after coding and zero padding and an electrocardiosignal data segment without coding and zero padding, the maximum data length is used as a characteristic, and the electrocardiosignal data segment after coding and zero padding and the electrocardiosignal data segment without coding and zero padding are used as data in a data set to form a data set.
Further, constructing a neural network model specifically includes:
and performing feature extraction on the electrocardiosignal data segment after zero padding by coding and the electrocardiosignal data segment without zero padding by coding through a main network, adding a feature attention structure on a feature module of a one-dimensional convolution network Res-Net, performing dimension fusion on the maximum data length and the features extracted by the main network by taking the maximum data length as the number of nodes of a full connection layer, and classifying the fusion features through a Softmax layer to construct a neural network model.
Further, the method for extracting features of the coded and zero-padded cardiac signal data segment and the uncoded and zero-padded cardiac signal data segment by using the main network as input specifically comprises the following steps:
the convolution module divides input into two channels to extract features, one channel increases feature map dimension through a one-dimensional convolution layer and a batch normalization layer, the other channel extracts features through two times of convolution-batch normalization-activation layers under the condition of not changing the dimension, and the feature map dimension is improved through one time of convolution-batch normalization; the dimensions of the two channels are the same, and the results of the two channel feature maps are added and then are passed through the ReLu activation layer to obtain the result of the convolution module.
Further, the feature attention structure is specifically that the input is connected with the jump of the lower channel, the first half part of the lower channel extracts features through two convolutions, batch normalization, activation layer and one convolution, and batch normalization layer, and the shape of the features is the same as the input of the backbone network.
Further, training a neural network model by using the data set specifically includes: the data set is randomly disordered and divided into a training verification set and a test set, the training verification set is divided into the training set and the verification set by using cross verification, iterative training is carried out by using the training set and the test set, and verification is carried out by using the verification set.
The invention also provides an electrocardiosignal classification device based on deep learning, which comprises a processor and a memory, wherein the memory is stored with a computer program, and when the computer program is executed by the processor, the electrocardiosignal classification device based on deep learning realizes the electrocardiosignal classification method based on deep learning according to any technical scheme.
The invention also provides a computer readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the method for classifying electrocardiosignals based on deep learning according to any one of the above technical solutions is realized.
Compared with the prior art, the invention has the beneficial effects that: carrying out R wave detection on an original electrocardiosignal by collecting the original electrocardiosignal to obtain an R point coordinate, segmenting the original electrocardiosignal according to the R point coordinate graph and the total heart beat number in the original electrocardiosignal to obtain an electrocardiosignal data segment, and forming a data set according to the electrocardiosignal data segment; constructing a neural network model, and training the neural network model by using the data set to obtain the neural network model for electrocardiosignal classification; re-collecting original electrocardiosignals, and classifying the original electrocardiosignals by the neural network model for classifying the electrocardiosignals to obtain the types of the original electrocardiosignals; the classification precision of the electrocardiosignals is improved.
Drawings
FIG. 1 is a schematic flow chart of a deep learning-based electrocardiosignal classification method according to the present invention;
FIG. 2 is a schematic diagram of the segmentation of an original ECG signal provided by the present invention;
FIG. 3 is a schematic structural diagram of a neural network provided in the present invention;
FIG. 4 is a schematic structural diagram of a convolution module provided in the present invention;
fig. 5 is a schematic structural diagram of a feature attention module provided in the present invention.
Detailed Description
The accompanying drawings, which are incorporated in and constitute a part of this application, illustrate preferred embodiments of the invention and together with the description, serve to explain the principles of the invention and not to limit the scope of the invention.
Example 1
The embodiment of the invention provides an electrocardiosignal classification method based on deep learning, which has a flow schematic diagram, as shown in figure 1, and comprises the following steps:
s1, collecting original electrocardiosignals, carrying out R wave detection on the original electrocardiosignals to obtain R point coordinates, segmenting the original electrocardiosignals according to the R point coordinate diagram and the total heart beat number in the original electrocardiosignals to obtain electrocardiosignal data segments, and forming a data set according to the electrocardiosignal data segments;
s2, constructing a neural network model, and training the neural network model by using the data set to obtain the neural network model for electrocardiosignal classification;
and S3, re-collecting the original electrocardiosignals, and classifying the original electrocardiosignals by the neural network model for classifying the electrocardiosignals to obtain the types of the original electrocardiosignals.
According to the technical scheme, original electrocardiosignals are collected, R wave detection is carried out on the original electrocardiosignals to obtain R point coordinates, the original electrocardiosignals are segmented according to an R point coordinate graph and the total heart beat number in the original electrocardiosignals to obtain electrocardiosignal data segments, data sets are formed according to the electrocardiosignal data segments, and a neural network model obtained through training of the data sets can improve the identification precision of the electrocardiosignals, namely the classification precision of the electrocardiosignals.
It should be noted that each original electrocardiographic signal has its label, for example, for electrocardiographic signal data labeled as disease a, if this data segment contains the characteristics of disease a, it is labeled as 1, and if not, it is labeled as 0; at the moment, each piece of electrocardiosignal data can form One-Hot coding characteristics with the length corresponding to the number of data segments;
preferably, segmenting the original electrocardiographic signal according to the R point coordinate and the total heart beat number in the original electrocardiographic signal to obtain an electrocardiographic signal data segment, and specifically includes:
and segmenting the original electrocardiosignal between two adjacent R point coordinates to obtain N sections of electrocardiosignal data sections, wherein N is S/T and is rounded up, S is the total heart beat number in the original electrocardiosignal, and T is the heart beat number in the electrocardiosignal data sections.
In a specific embodiment, after R-wave detection is performed on an original electrocardiographic signal, an R-point coordinate graph including R-point coordinates is obtained, the total number of heartbeats included in original electrocardiographic signal data D is set as S, the number of heartbeats included in a data segment is set as T, the original electrocardiographic signal data D is segmented according to the number of heartbeats, so that the length of a data segment One-Hot of the original electrocardiographic signal data is rounded up for S/T, the obtained One-Hot is a feature D of the original electrocardiographic signal, and a schematic diagram of original electrocardiographic signal segmentation is shown in fig. 2.
The operation is carried out on all original electrocardiosignal data of the data set, the obtained characteristic d is different due to different lengths of the data, zero filling is carried out on all other original data to the maximum data length at the moment, zero filling is carried out on One-Hot codes of the characteristic d to the maximum characteristic, and a data set containing new data after the zero filling of the original data, a label (waveform type) and the maximum characteristic is formed.
Preferably, forming a data set according to the electrocardiographic signal data segment specifically includes:
acquiring the maximum data length of all the electrocardiosignal data segments, carrying out zero filling on the electrocardiosignal data segment codes with the data length not being the maximum data length to the maximum data length, and forming a data set according to the electrocardiosignal data segment codes after the zero filling of the codes and the electrocardiosignal data segments without the zero filling of the codes.
It should be noted that the maximum data length is the maximum characteristic.
Preferably, the forming a data set according to the encoded and zero-padded cardiac signal data segment and the uncoded and zero-padded cardiac signal data segment specifically includes:
the electrocardio type is used as a label, whether disease characteristics are contained is used as marks of an electrocardiosignal data segment after coding and zero padding and an electrocardiosignal data segment without coding and zero padding, the maximum data length is used as a characteristic, and the electrocardiosignal data segment after coding and zero padding and the electrocardiosignal data segment without coding and zero padding are used as data in a data set to form a data set.
Preferably, the constructing of the neural network model specifically includes:
and performing feature extraction on the electrocardiosignal data segment after zero padding by coding and the electrocardiosignal data segment without zero padding by coding through a main network, adding a feature attention structure on a feature module of a one-dimensional convolution network Res-Net, performing dimension fusion on the maximum data length and the features extracted by the main network by taking the maximum data length as the number of nodes of a full connection layer, and classifying the fusion features through a Softmax layer to construct a neural network model.
In a specific embodiment, as shown in fig. 3, a schematic structural diagram of a neural network is obtained by inputting coded zero-padded electrocardiographic signal data segments and uncoded zero-padded electrocardiographic signal data segments through a trunk network to perform feature extraction, setting the trunk network as a 50-layer one-dimensional convolution network Res-Net, modifying the number of nodes of a final full-connection layer to be the maximum data length (maximum feature), then performing dimension fusion operation on the maximum feature and the features extracted from the trunk network, and finally classifying the fusion features through a Softmax layer.
The backbone network adds a feature attention structure in an original Res-Net feature module (index Block); the structure of the convolution Block (Conv Block) is shown in fig. 4, and the structure of the feature attention Block (index Block) is shown in fig. 5.
Preferably, the method for extracting features of the encoded and zero-padded cardiac signal data segment and the uncoded and zero-padded cardiac signal data segment by using the main network includes:
the convolution module divides input into two channels to extract features, one channel increases feature map dimension through a one-dimensional convolution layer and a batch normalization layer, the other channel extracts features through two times of convolution-batch normalization-activation layers under the condition of not changing the dimension, and the feature map dimension is improved through one time of convolution-batch normalization; the dimensions of the two channels are the same, and the results of the two channel feature maps are added and then are passed through the ReLu activation layer to obtain the result of the convolution module.
In a specific embodiment, the convolution module divides input into two channels to extract features, the dimension of a feature map is increased by an upper channel through a one-dimensional convolution layer and a batch normalization layer, the feature is extracted by a lower channel through two times of convolution-batch normalization-activation layers under the condition of not changing the dimension, and the dimension of the feature map is improved through one time of convolution-batch normalization; the dimensionality of the upper channel and the dimensionality of the lower channel are the same, the results of the convolution module are obtained through the ReLu activation layer after the results of the upper channel and the lower channel are added, and the convolution module is arranged to improve the feature extraction of the dimensionality of the feature map.
Preferably, the feature attention structure is specifically that the input is connected with the jump of the lower channel, the first half part of the lower channel extracts the feature through two convolutions, batch normalization, activation layer and one convolution, and batch normalization layer, and the shape of the feature is the same as the input of the backbone network.
In a specific embodiment, because the feature graph has multiple channel dimensions, but each channel has different importance, in order to enable a network to notice a key feature, a weight of each channel of the feature graph at the time needs to be obtained, a specific process for obtaining the weight of each channel of the feature graph is to assume that a y shape of the feature graph obtained from the first half of a lower channel is lxc, a shape of 1 × C is obtained through global pooling, a shape of 1 × r is obtained through a full connection layer with the number of nodes r, the shape obtained through the full connection layer with the number of nodes C is reduced to 1 × C through an activation layer, and finally, weights of C channels are obtained through a sigmoid activation layer. And correspondingly multiplying the weight value by y, wherein the operation is Scale operation in the graph, and the weighted final characteristic z of the lower channel is obtained. And adding the characteristic z and the corresponding element of the module input x, and obtaining the final output of the module through a ReLu activation layer.
The module input and the final output are the same in shape, so that the module input and the final output can be connected in series in the network, and the series connection times of the module input and the module output at different positions in the backbone network are respectively 2, 3, 5 and 2. The convolution module in the backbone network mainly has the functions of feature extraction and feature dimension increasing to obtain more useful features; the feature attention module acts as a feature extraction and weights each channel of the feature, allowing the network to notice important channels and ignore unimportant channels.
Preferably, the training of the neural network model by using the data set specifically includes: the data set is randomly disordered and divided into a training verification set and a test set, the training verification set is divided into the training set and the verification set by using cross verification, iterative training is carried out by using the training set and the test set, and verification is carried out by using the verification set.
In one embodiment, the 9-class data set disclosed in CPSC2018 is used as the data set source, and the data set comprises Normal (Normal, 918 strips), atrial fibrillation (AF, 1098 strips), one-degree atrioventricular block (I-AVB, 704 strips), left bundle branch block (LBBB, 207 strips), right bundle branch block (RBBB, 1695 strips), atrial premature beat (PAC, 556 strips), ventricular premature beat (PVC, 672 strips), ST-segment elevation (STE, 202 strips), ST-segment reduction (STD, 825 strips) totaling 6877 strips of 9 types of data, the data sampling rate is 500hz, the length is between 5s and 140s, and the number of leads is 12; randomly disorganizing the data, dividing the data into a training verification set containing 6400 strips and a test set containing 477 strips, dividing the training test set into 10 parts by using 10-time cross verification, and training 10 models by using 9 parts of the training test set and 1 part of the training test set as the verification set each time; the size of the trained batch is 64, the number of iterations is 5000, and the used deep learning frame is Pythrch; the final evaluation index of the network is the average value of F1 scores of each class, and the probability average value of 10 test results of 10 models to the test set obtained by 10 times of cross validation is used as the final result.
In order to compare the superiority of the neural network model (attention model) for classifying electrocardiosignals according to the embodiment of the invention, the neural network model for classifying electrocardiosignals is compared with the three models; the three models are model A, which does not contain the characteristic d as an input model; the model B is a model of the attention structure of the rear half part which is not contained in the attention characteristic module; model C, both structures do not contain; other experimental processes of the three models are the same as those of the neural network model for classifying the electrocardiosignals; the results of F1 values for the attention model and the other three models were obtained experimentally, as shown in table 1.
TABLE 1
| F1 value | Attention model | Model A | Model B | Model C |
| Normal | 0.7527 | 0.7444 | 0.7503 | 0.7401 |
| AF | 0.7887 | 0.7801 | 0.7861 | 0.7781 |
| I-AVB | 0.7620 | 0.7551 | 0.7534 | 0.7446 |
| LBBB | 0.7125 | 0.6997 | 0.7002 | 0.6926 |
| RBBB | 0.8412 | 0.8310 | 0.8398 | 0.8237 |
| PAC | 0.6003 | 0.5948 | 0.5901 | 0.5885 |
| PVC | 0.8121 | 0.8001 | 0.8106 | 0.7931 |
| STD | 0.6752 | 0.6731 | 0.6543 | 0.6502 |
| STE | 0.3141 | 0.3221 | 0.3007 | 0.2814 |
| Average | 0.6954 | 0.6889 | 0.6873 | 0.6769 |
It should be noted that, in the following description,
p is the accuracy, namely the proportion of the correct case to the predicted case data; r is the recall rate, namely the proportion of the data predicted to be the regular case to the actual regular case data; the results of each model in the above table are the average of ten-fold cross-validation models, and it can be seen from table 1 that the predicted results of the attention model in each classThe method is higher than other models, and the effectiveness and the accuracy of the attention model are reflected.
Example 2
The embodiment of the invention provides an electrocardiosignal classification device based on deep learning, which comprises a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the electrocardiosignal classification device based on deep learning in the embodiment 1 is realized.
Example 3
An embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored, where the computer program, when executed by a processor, implements the deep learning-based electrocardiographic signal classification method according to embodiment 1.
The invention discloses an electrocardiosignal classification method and device based on deep learning and a computer readable storage medium, wherein original electrocardiosignals are collected and subjected to R wave detection to obtain R point coordinates, the original electrocardiosignals are segmented according to an R point coordinate graph and the total heart beat number in the original electrocardiosignals to obtain electrocardiosignal data segments, and a data set is formed according to the electrocardiosignal data segments; constructing a neural network model, and training the neural network model by using the data set to obtain the neural network model for electrocardiosignal classification; re-collecting original electrocardiosignals, and classifying the original electrocardiosignals by the neural network model for classifying the electrocardiosignals to obtain the types of the original electrocardiosignals; the classification precision of the electrocardiosignals is improved.
The technical scheme of the invention divides the original electrocardiosignal data by fixed heart beat, marks each divided data segment, and gives prior information to the network, namely the data segment comprises a data segment with disease characteristics (One-Hot is 1) and a data segment without disease characteristics (One-Hot is 0); after the network extracts the final characteristics, the network is combined with the prior information to allow the network to pay attention to more important data segments, so that the identification precision of the neural network is improved; by means of the modified Res-Net characteristic module, a channel attention mechanism is added in the characteristic module, the weight of each channel is learned along with network training, the network is made to pay attention to important channel parts in the characteristics, the recognition accuracy of the neural network is further improved, and therefore the classification accuracy of the electrocardiosignals is improved.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention.
Claims (10)
1. An electrocardiosignal classification method based on deep learning is characterized by comprising the following steps:
collecting original electrocardiosignals, carrying out R wave detection on the original electrocardiosignals to obtain R point coordinates, segmenting the original electrocardiosignals according to the R point coordinate graph and the total heart beat number in the original electrocardiosignals to obtain electrocardiosignal data segments, and forming a data set according to the electrocardiosignal data segments;
constructing a neural network model, and training the neural network model by using the data set to obtain the neural network model for electrocardiosignal classification;
and re-collecting the original electrocardiosignals, and classifying the original electrocardiosignals by the neural network model for classifying the electrocardiosignals to obtain the types of the original electrocardiosignals.
2. The deep learning-based electrocardiosignal classification method according to claim 1, wherein the step of segmenting the original electrocardiosignal according to the R point coordinates and the total heart beat number in the original electrocardiosignal to obtain an electrocardiosignal data segment specifically comprises the steps of:
and segmenting the original electrocardiosignal between two adjacent R point coordinates to obtain N sections of electrocardiosignal data sections, wherein N is S/T and is rounded up, S is the total heart beat number in the original electrocardiosignal, and T is the heart beat number in the electrocardiosignal data sections.
3. The deep learning-based electrocardiographic signal classification method according to claim 2, wherein forming a data set from the electrocardiographic signal data segments specifically comprises:
acquiring the maximum data length of all the electrocardiosignal data segments, carrying out zero filling on the electrocardiosignal data segment codes with the data length not being the maximum data length to the maximum data length, and forming a data set according to the electrocardiosignal data segment codes after the zero filling of the codes and the electrocardiosignal data segments without the zero filling of the codes.
4. The deep learning-based electrocardiosignal classification method according to claim 3 is characterized in that a data set is formed according to the coded and zero-padded electrocardiosignal data segment and the uncoded and zero-padded electrocardiosignal data segment, and specifically comprises the following steps:
the electrocardio type is used as a label, whether disease characteristics are contained is used as marks of an electrocardiosignal data segment after coding and zero padding and an electrocardiosignal data segment without coding and zero padding, the maximum data length is used as a characteristic, and the electrocardiosignal data segment after coding and zero padding and the electrocardiosignal data segment without coding and zero padding are used as data in a data set to form a data set.
5. The deep learning-based electrocardiosignal classification method according to claim 4, wherein the neural network model is constructed, and specifically comprises the following steps:
and performing feature extraction on the electrocardiosignal data segment after zero padding by coding and the electrocardiosignal data segment without zero padding by coding through a main network, adding a feature attention structure on a feature module of a one-dimensional convolution network Res-Net, performing dimension fusion on the maximum data length and the features extracted by the main network by taking the maximum data length as the number of nodes of a full connection layer, and classifying the fusion features through a Softmax layer to construct a neural network model.
6. The deep learning-based electrocardiosignal classification method according to claim 5, wherein the characteristics of the coded and zero-padded electrocardiosignal data segment and the uncoded and zero-padded electrocardiosignal data segment are extracted through a backbone network by taking the coded and zero-padded electrocardiosignal data segment and the uncoded and zero-padded electrocardiosignal data segment as input, and the method specifically comprises the following steps:
the convolution module divides input into two channels to extract features, one channel increases feature map dimension through a one-dimensional convolution layer and a batch normalization layer, the other channel extracts features through two times of convolution-batch normalization-activation layers under the condition of not changing the dimension, and the feature map dimension is improved through one time of convolution-batch normalization; the dimensions of the two channels are the same, and the results of the two channel feature maps are added and then are passed through the ReLu activation layer to obtain the result of the convolution module.
7. The deep learning-based electrocardiosignal classification method according to claim 5, wherein the feature attention structure is specifically a jump connection of an input and a lower channel, the first half of the lower channel is subjected to feature extraction through two convolutions, batch normalization, an activation layer, one convolution and a batch normalization layer, and the shape of the feature is the same as the input of a trunk network.
8. The deep learning-based electrocardiosignal classification method according to claim 1, wherein the training of the neural network model by using the data set specifically comprises: the data set is randomly disordered and divided into a training verification set and a test set, the training verification set is divided into the training set and the verification set by using cross verification, iterative training is carried out by using the training set and the test set, and verification is carried out by using the verification set.
9. An apparatus for classifying electrocardiographic signals based on deep learning, comprising a processor and a memory, wherein the memory stores a computer program, and the computer program, when executed by the processor, implements the method for classifying electrocardiographic signals based on deep learning according to any one of claims 1 to 8.
10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out a deep learning based classification method for cardiac electrical signals according to any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110076938.1A CN112883803B (en) | 2021-01-20 | 2021-01-20 | Electrocardiogram signal classification method, device and storage medium based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110076938.1A CN112883803B (en) | 2021-01-20 | 2021-01-20 | Electrocardiogram signal classification method, device and storage medium based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112883803A true CN112883803A (en) | 2021-06-01 |
| CN112883803B CN112883803B (en) | 2023-09-01 |
Family
ID=76051044
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110076938.1A Active CN112883803B (en) | 2021-01-20 | 2021-01-20 | Electrocardiogram signal classification method, device and storage medium based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112883803B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113288163A (en) * | 2021-06-04 | 2021-08-24 | 浙江理工大学 | Multi-feature fusion electrocardiosignal classification model modeling method based on attention mechanism |
| CN116584951A (en) * | 2023-04-23 | 2023-08-15 | 山东省人工智能研究院 | Electrocardiosignal detection and positioning method based on weak supervision learning |
Citations (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103815897A (en) * | 2014-02-28 | 2014-05-28 | 吉林大学 | Electrocardiogram characteristic extraction method |
| CN106725428A (en) * | 2016-12-19 | 2017-05-31 | 中国科学院深圳先进技术研究院 | A kind of electrocardiosignal sorting technique and device |
| CN107122788A (en) * | 2017-03-09 | 2017-09-01 | 吉林大学 | A kind of personal identification method and device based on electrocardiosignal |
| CN107516075A (en) * | 2017-08-03 | 2017-12-26 | 安徽华米信息科技有限公司 | Detection method, device and the electronic equipment of electrocardiosignal |
| CN108596142A (en) * | 2018-05-09 | 2018-09-28 | 吉林大学 | A kind of cardioelectric characteristic extracting process based on PCANet |
| CN109620205A (en) * | 2018-12-26 | 2019-04-16 | 上海联影智能医疗科技有限公司 | Electrocardiogram (ECG) data classification method, device, computer equipment and storage medium |
| CN109645983A (en) * | 2019-01-09 | 2019-04-19 | 南京航空航天大学 | A kind of uneven beat classification method based on multimode neural network |
| CN110477906A (en) * | 2019-08-26 | 2019-11-22 | 电子科技大学 | A kind of electrocardiosignal QRS complex start-stop independent positioning method |
| CN110974211A (en) * | 2019-12-09 | 2020-04-10 | 上海数创医疗科技有限公司 | ST-segment classification neural network of high-order polynomial activation function and application thereof |
| US20200121255A1 (en) * | 2017-11-27 | 2020-04-23 | Lepu Medical Technology (Bejing) Co., Ltd. | Artificial intelligence-based interference recognition method for electrocardiogram |
| CN111110228A (en) * | 2020-01-17 | 2020-05-08 | 武汉中旗生物医疗电子有限公司 | Electrocardiosignal R wave detection method and device |
| US20200237246A1 (en) * | 2017-11-27 | 2020-07-30 | Lepu Medical Technology (Bejing) Co., Ltd. | Automatic recognition and classification method for electrocardiogram heartbeat based on artificial intelligence |
| CN111603170A (en) * | 2020-06-01 | 2020-09-01 | 河南华南医电科技有限公司 | Human body position detection method based on vector cardiogram |
| CN111657914A (en) * | 2020-04-30 | 2020-09-15 | 上海数创医疗科技有限公司 | Electrocardiogram form recognition method and device based on deep learning |
| US20200289010A1 (en) * | 2019-03-14 | 2020-09-17 | University Of Seoul Industry Cooperation Foundation | Method for classifying type of heartbeat and apparatus using the same |
-
2021
- 2021-01-20 CN CN202110076938.1A patent/CN112883803B/en active Active
Patent Citations (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103815897A (en) * | 2014-02-28 | 2014-05-28 | 吉林大学 | Electrocardiogram characteristic extraction method |
| CN106725428A (en) * | 2016-12-19 | 2017-05-31 | 中国科学院深圳先进技术研究院 | A kind of electrocardiosignal sorting technique and device |
| CN107122788A (en) * | 2017-03-09 | 2017-09-01 | 吉林大学 | A kind of personal identification method and device based on electrocardiosignal |
| CN107516075A (en) * | 2017-08-03 | 2017-12-26 | 安徽华米信息科技有限公司 | Detection method, device and the electronic equipment of electrocardiosignal |
| US20200121255A1 (en) * | 2017-11-27 | 2020-04-23 | Lepu Medical Technology (Bejing) Co., Ltd. | Artificial intelligence-based interference recognition method for electrocardiogram |
| US20200237246A1 (en) * | 2017-11-27 | 2020-07-30 | Lepu Medical Technology (Bejing) Co., Ltd. | Automatic recognition and classification method for electrocardiogram heartbeat based on artificial intelligence |
| CN108596142A (en) * | 2018-05-09 | 2018-09-28 | 吉林大学 | A kind of cardioelectric characteristic extracting process based on PCANet |
| CN109620205A (en) * | 2018-12-26 | 2019-04-16 | 上海联影智能医疗科技有限公司 | Electrocardiogram (ECG) data classification method, device, computer equipment and storage medium |
| CN109645983A (en) * | 2019-01-09 | 2019-04-19 | 南京航空航天大学 | A kind of uneven beat classification method based on multimode neural network |
| US20200289010A1 (en) * | 2019-03-14 | 2020-09-17 | University Of Seoul Industry Cooperation Foundation | Method for classifying type of heartbeat and apparatus using the same |
| CN110477906A (en) * | 2019-08-26 | 2019-11-22 | 电子科技大学 | A kind of electrocardiosignal QRS complex start-stop independent positioning method |
| CN110974211A (en) * | 2019-12-09 | 2020-04-10 | 上海数创医疗科技有限公司 | ST-segment classification neural network of high-order polynomial activation function and application thereof |
| CN111110228A (en) * | 2020-01-17 | 2020-05-08 | 武汉中旗生物医疗电子有限公司 | Electrocardiosignal R wave detection method and device |
| CN111657914A (en) * | 2020-04-30 | 2020-09-15 | 上海数创医疗科技有限公司 | Electrocardiogram form recognition method and device based on deep learning |
| CN111603170A (en) * | 2020-06-01 | 2020-09-01 | 河南华南医电科技有限公司 | Human body position detection method based on vector cardiogram |
Non-Patent Citations (3)
| Title |
|---|
| 庞彦伟，等: "基于卷积神经网络的第一导联心电图心拍分类", 天津大学学报(自然科学与工程技术版), no. 10, pages 5 - 12 * |
| 潘辉，等: "基于改进残差网络对心电信号的识别", 数据采集与处理, no. 04, pages 96 - 106 * |
| 王文, 等: "基于小波变换的心电图QRS波群检测方法研究", 北京生物医学工程, no. 04, pages 3 - 5 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113288163A (en) * | 2021-06-04 | 2021-08-24 | 浙江理工大学 | Multi-feature fusion electrocardiosignal classification model modeling method based on attention mechanism |
| CN116584951A (en) * | 2023-04-23 | 2023-08-15 | 山东省人工智能研究院 | Electrocardiosignal detection and positioning method based on weak supervision learning |
| CN116584951B (en) * | 2023-04-23 | 2023-12-12 | 山东省人工智能研究院 | Electrocardiosignal detection and positioning method based on weak supervision learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112883803B (en) | 2023-09-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109886273B (en) | CMR image segmentation and classification system | |
| CN108446730B (en) | CT pulmonary nodule detection device based on deep learning | |
| CN107203692A (en) | The implementation method of atrial fibrillation detection based on depth convolutional neural networks | |
| CN111110228B (en) | Electrocardiosignal R wave detection method and device | |
| CN105139004A (en) | Face expression identification method based on video sequences | |
| CN112883803B (en) | Electrocardiogram signal classification method, device and storage medium based on deep learning | |
| CN110288028B (en) | Electrocardio detection method, system, equipment and computer readable storage medium | |
| CN112426160A (en) | Electrocardiosignal type identification method and device | |
| CN113274031B (en) | Arrhythmia classification method based on depth convolution residual error network | |
| CN111803059A (en) | Electrocardiosignal classification method and device based on time domain convolution network | |
| CN110688942A (en) | Electrocardiosignal joint identification module and method based on InResNet network | |
| CN115115620B (en) | Pneumonia lesion simulation method and system based on deep learning | |
| CN115530788A (en) | Arrhythmia classification method based on self-attention mechanism | |
| CN114781441B (en) | EEG motor imagery classification method and multi-space convolution neural network model | |
| CN113486752B (en) | Emotion recognition method and system based on electrocardiosignal | |
| CN116012568A (en) | System for acquiring cardiac rhythm information through photographing electrocardiogram | |
| CN115363599A (en) | Electrocardiosignal processing method and system for atrial fibrillation recognition | |
| CN112957052B (en) | Multi-lead electrocardiosignal classification method based on NLF-CNN lead fusion depth network | |
| CN113344933A (en) | Glandular cell segmentation method based on multi-level feature fusion network | |
| CN107092895A (en) | A kind of multi-modal emotion identification method based on depth belief network | |
| CN116958705A (en) | Medical image classifying system based on graph neural network | |
| CN116206160A (en) | Automatic identification network model and automatic sketching network model construction method for nasopharyngeal carcinoma lesion tissues based on convolutional neural network model | |
| CN112560784B (en) | Electrocardiogram classification method based on dynamic multi-scale convolutional neural network | |
| CN114366116A (en) | Parameter acquisition method based on Mask R-CNN network and electrocardiogram | |
| CN114190884A (en) | Longitudinal analysis method, system and device for brain disease data |
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 |