CN113229825A - Deep neural network-based multi-label multi-lead electrocardiogram classification method - Google Patents
Deep neural network-based multi-label multi-lead electrocardiogram classification method Download PDFInfo
- Publication number
- CN113229825A CN113229825A CN202110690012.1A CN202110690012A CN113229825A CN 113229825 A CN113229825 A CN 113229825A CN 202110690012 A CN202110690012 A CN 202110690012A CN 113229825 A CN113229825 A CN 113229825A
- Authority
- CN
- China
- Prior art keywords
- data
- lead
- layer
- fusion part
- feature fusion
- 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 32
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 15
- 230000004927 fusion Effects 0.000 claims abstract description 42
- 230000007246 mechanism Effects 0.000 claims abstract description 41
- 238000011176 pooling Methods 0.000 claims abstract description 35
- 238000012545 processing Methods 0.000 claims abstract description 31
- 238000007781 pre-processing Methods 0.000 claims abstract description 5
- 238000010606 normalization Methods 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 230000007306 turnover Effects 0.000 claims description 3
- 230000002159 abnormal effect Effects 0.000 description 10
- 238000013135 deep learning Methods 0.000 description 8
- 208000024172 Cardiovascular disease Diseases 0.000 description 6
- 238000002474 experimental method Methods 0.000 description 6
- 201000010099 disease Diseases 0.000 description 4
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 4
- 238000010801 machine learning Methods 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000001514 detection method Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 241000283973 Oryctolagus cuniculus Species 0.000 description 1
- 230000006793 arrhythmia Effects 0.000 description 1
- 206010003119 arrhythmia Diseases 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000013210 evaluation model Methods 0.000 description 1
- 238000013401 experimental design Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
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
- A61B5/725—Details of waveform analysis using specific filters therefor, e.g. Kalman or adaptive filters
-
- 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
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Physics & Mathematics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physiology (AREA)
- Psychiatry (AREA)
- Signal Processing (AREA)
- Veterinary Medicine (AREA)
- Public Health (AREA)
- Pathology (AREA)
- General Health & Medical Sciences (AREA)
- Heart & Thoracic Surgery (AREA)
- Medical Informatics (AREA)
- Molecular Biology (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- Mathematical Physics (AREA)
- Evolutionary Computation (AREA)
- Fuzzy Systems (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The invention provides a deep neural network-based multi-label multi-lead electrocardiogram classification method, which comprises the steps of preprocessing original electrocardiosignals into 12-lead signals, inputting the 12-lead signals into a multi-scale residual error network model for processing and classification, wherein the multi-scale residual error network model comprises a feature fusion part and a multi-scale feature fusion part; the feature fusion part comprises convolution layers and a first maximum pooling layer, the multi-scale feature fusion part comprises 32 convolution network layers and full-connection layers, the full-connection layers comprise two second maximum pooling layers, and channel space attention mechanism modules are sequentially arranged between the convolution layers and the first maximum pooling layer and between the 32 convolution layers and the second maximum pooling layers of each branch; the multi-scale residual error network model in the multi-label multi-lead electrocardiogram classification method provided by the invention has good performance and generalization performance.
Description
Technical Field
The invention belongs to the field of electrocardiogram monitoring, and particularly relates to a deep neural network-based multi-label multi-lead electrocardiogram classification method.
Background
In recent years, the study of assisting doctors in diagnosing electrocardiogram by using an automatic electrocardio analysis technology at home and abroad becomes a more popular study direction. The work of electrocardiogram type identification can be roughly divided into two aspects: the problem of single-label electrocardiogram classification and the problem of multi-label electrocardiogram classification are respectively solved. According to different technical methods used in different periods, the method can be further divided into the following steps: conventional machine learning classifiers and deep neural network electrocardiogram classifications, such as machine learning-based single-label electrocardiogram classification, deep learning-based multi-label electrocardiogram classification.
The single-label electrocardiogram classification based on machine learning is to extract characteristic parameters from electrocardiogram signals and then send the characteristic parameters to different classifiers for classification; although this method works well, the choice of features has a large impact on the final result and requires manual feature extraction work, wasting time and effort on the huge database of ECG data.
The deep learning can omit the step of manual extraction in the traditional machine learning for automatic learning of a large amount of data because of the characteristic that the deep learning can automatically extract data features, and is widely applied to ECG classification research; with the proposal of Deep Learning (DL) theory, many researchers introduce a Deep learning algorithm into an electrocardiogram, and realize automatic detection of cardiovascular diseases by a Deep learning method, and the result shows that the diagnosis efficiency of arrhythmia can be greatly improved by applying a computer to the field of heart beat recognition.
In deep learning-based multi-label electrocardiogram classification, more than one signal classification result appears in electrocardiosignals within a certain time range, and a plurality of researchers also pay attention to the result; in the 2018 Chinese physiological signal data challenge match, competition data is 12-lead electrocardiogram data, and corresponding data labels are one or more of 9 disease types; in 2019, the fixed length data of 8 leads given by data in 'combined fat high new cup electrocardiogram man-machine intelligent competition-electrocardiogram abnormal event prediction' is one or more of 55 electrocardiogram abnormal event types corresponding to the disease types; it is therefore meaningful to develop multi-labeled electrocardiogram classification; with the development of multi-label electrocardiogram classification research work, multi-label classification of various electrocardiogram data is provided and a feasible analysis method is provided, but the methods are carried out on a single database or a private database and cannot reflect the application effect of a model in an actual clinical database.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a deep neural network-based multi-label multi-lead electrocardiogram classification method, which solves the problems in the background art.
A multi-label multi-lead electrocardiogram classification method based on a deep neural network comprises the steps that original electrocardiosignals are preprocessed to be 12-lead signals, the 12-lead signals are input into a multi-scale residual error network model for processing and classification, the multi-scale residual error network model comprises a feature fusion part and a multi-scale feature fusion part, the 12-lead signals are processed by the feature fusion part to form two branches comprising 32 convolutional layers and an average pool layer, the two branches are respectively endowed with convolutional kernels with different scales, the two branches are processed by the multi-scale feature fusion part to output 1024 nodes, and the 1024 nodes output multi-label electrocardio categories through an output layer; the feature fusion part comprises a convolution layer and a first maximum pooling layer, the multi-scale feature fusion part comprises 32 convolution network layers and a full-connection layer, the full-connection layer comprises two second maximum pooling layers, and a channel attention mechanism module and a space attention mechanism module are sequentially arranged between the convolution layer and the first maximum pooling layer and between the 32 convolution layers of each branch and the second maximum pooling layers.
Further, 64 convolution kernels with the scale of 15 are adopted by the convolution layer of the feature fusion part.
Further, the 30-layer convolution network layers of the two branches of the feature fusion part respectively adopt 64 convolution kernels with the scale of 3 and 64 convolution kernels with the scale of 7.
Further, the data processing process of the feature fusion part is that the 12-lead signal is processed by the convolutional layer to obtain a data set F, and the data set F is processed by the channel attention mechanism module to obtain a data set MC(F) Data set F and data set MC(F) The element-by-element multiplication is carried out to obtain a data set F ', and the data set F' is processed by a space attention mechanism module to obtain a data set Ms(F '), and comparing the data set F' with the data set Ms(F ') element by element multiplication to obtain a data set F ', and transmitting the data of the data set F ' to the first maximum pooling layer for processing.
Further, the data processing process of the multi-scale feature fusion part is that two branch data including 32 convolutional layers and an average pool layer, which are obtained after the processing of the first maximum pooling layer, are respectively transmitted to the second large pooling layer in the full connection layer after being processed by the same processing process from the convolutional layer to the first maximum pooling layer in the feature fusion part.
Further, the specific processing process of preprocessing the original electrocardiosignals into 12-lead signals is as follows: the extracted 8-lead original electrocardiogram data are converted into 12-lead data, the integral data of the 12-lead data are multiplied by 4.88 microvolts to obtain real electrocardiogram data, and the real electrocardiogram data are subjected to variable length data processing and data enhancement to obtain 12-lead signals.
Further, the variable length data are processed by taking 5000 points of real electrocardio data, omitting the initial 1.25s and taking the middle 10s data; and the data enhancement is to perform normalization processing on the acquired middle 10s data by using Gaussian distribution, and then vertically turn over and vertically translate the normalized data by 20 sampling points to increase data samples.
Compared with the prior art, the invention has the following beneficial effects:
according to the invention, the channel attention mechanism and the space attention mechanism are introduced into the multi-scale residual error network model, so that channel information and space information can be concerned, important data is filtered, more important data segments and abnormal data channels on abnormal electrocardiograms are concerned, and larger weight is distributed to the important data segments and the abnormal data channels, thereby improving the performance of the multi-scale residual error network model, and the multi-scale residual error network model introduced with the channel attention mechanism and the space attention mechanism has good generalization performance.
Drawings
FIG. 1 is a flow chart of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments.
A multi-label multi-lead electrocardiogram classification method based on a deep neural network comprises the steps that original electrocardiosignals are preprocessed to be 12-lead signals, the 12-lead signals are input into a multi-scale residual error network model for processing and classification, the multi-scale residual error network model comprises a feature fusion part and a multi-scale feature fusion part, the 12-lead signals are processed by the feature fusion part to form two branches comprising 32 convolutional layers and an average pool layer, the two branches are respectively endowed with convolutional kernels with different scales, the two branches are processed by the multi-scale feature fusion part to output 1024 nodes, and the 1024 nodes output multi-label electrocardio categories through an output layer; the feature fusion part comprises convolution layers and a first maximum pooling layer, the multi-scale feature fusion part comprises 32 convolution network layers and full-connection layers, the full-connection layers comprise two second maximum pooling layers, and a channel attention mechanism module and a space attention mechanism module are sequentially arranged between the convolution layers and the first maximum pooling layer and between the 32 convolution layers and the second maximum pooling layers of each branch; in the invention, a channel space attention mechanism is introduced at the end of the multi-scale residual error network model, namely the channel space attention mechanism is added after the first convolution layer of the multi-scale residual error network model and before the last pooling layer on two scales, and the channel space attention mechanism is defined as CSA-MResNte 37.
In the invention, 64 convolution kernels with the scale of 15 are adopted by the convolution layer of the feature fusion part; the 30-layer convolution network layers of the two branches of the feature fusion part respectively adopt 64 convolution kernels with the scale of 3 and 64 convolution kernels with the scale of 7; in the process of extracting the features by using the convolutional neural network, because filters used by convolutional kernels have different sizes and the receptive fields have different sizes, and finally convolutional features obtained after convolution operation are also different, the convolutional kernels with different sizes are adopted for the two branches in the multi-scale feature fusion part to obtain more comprehensive features, and further more comprehensive classification of the electrocardio-diseases is obtained.
In the invention, the data processing process of the characteristic fusion part is that the 12-lead signal is processed by the convolutional layer to obtain a data set F, and the data set F is processed by the channel attention mechanism module to obtain a data set MC(F) Data set F and data set MC(F) The element-by-element multiplication is carried out to obtain a data set F ', and the data set F' is processed by a space attention mechanism module to obtain a data set Ms(F '), and comparing the data set F' with the data set Ms(F ') element by element multiplication to obtain a data set F ', and transmitting the data of the data set F ' to the first maximum pooling layer for processing.
In the invention, the data processing process of the multi-scale feature fusion part is that two branch data which are obtained after the processing of the first maximum pooling layer and comprise 32 convolutional layers and an average pooling layer are respectively transmitted to the second large pooling layer in the full connection layer after the processing of the two branch data which are the same as the processing from the convolutional layers to the first maximum pooling layer in the feature fusion part.
In the invention, the specific processing process of preprocessing the original electrocardiosignals into 12-lead signals comprises the following steps: the extracted 8-lead original electrocardiogram data are converted into 12-lead data, the integral data of the 12-lead data are multiplied by 4.88 microvolts to obtain real electrocardiogram data, and the real electrocardiogram data are subjected to variable length data processing and data enhancement to obtain 12-lead signals.
In the invention, the variable length data is processed by taking 5000 points of real electrocardio data, discarding the initial 1.25s and taking the middle 10s data; and the data enhancement is to perform normalization processing on the acquired middle 10s data by using Gaussian distribution, and then vertically turn over and vertically translate the normalized data by 20 sampling points to increase data samples.
The multi-scale residual error network model provided by the invention belongs to electrocardiogram data of rabbit 12-lead signals, and then comprises 2 branches after passing through a convolutional layer and a maximum pooling layer. Each branch has 32 layers of convolution and an average pool layer, residual connection is added on the basis of convolution to prevent rapid decrease of accuracy rate caused by deepening after saturation, and a channel space attention mechanism is added before each pool layer to obtain more important weight. The model can be divided into an upper part and a lower part: feature fusion and multi-scale feature fusion. Feature fusion includes convolution, channel spatial attention mechanism, and pooling layers. In this section, the convolution is set to 15 and only the set of data features at that scale is extracted. Then filtering important information through a channel space attention mechanism, and finally fusing the information through a pooling layer. In the multi-scale feature fusion part, convolution kernels with different scales are respectively endowed on the two branches, and the convolution kernels with different scales can be extracted. In each branch, 32 layers of remaining network and channel space attention mechanisms are set. The features of the two different scale branches are then aggregated by the pooling layer. And finally, combining the information on the two scales, and sending the combined information to a full connection layer for 9 kinds of multi-label electrocardiogram classification.
The experimental analysis procedure is as follows:
1. experimental design, in order to prove that the added attention mechanism is effective and better compare the experimental results of different attention mechanisms, the experiment is carried out by using a Chinese cardiovascular disease database. Selecting the best performance model from the experimental results to carry out the generalization experiment, and testing by using other databases in Precision, Recall and F1Score and other index evaluation models. In order to better represent the model and other researches provided by the invention on the cardiovascular disease database in China, two classification experiments of normal and abnormal electrocardiograms are added, and AUC indexes are added on the indexes.
2. Comparing experiments with different attention mechanisms, and respectively comparing a multi-scale residual error network model SE-MResNet37 with a channel attention mechanism, an MS-CSA-ResNet37 with a channel space attention mechanism added into a multi-scale residual error network residual error block, a CSA-MResNet37 with a channel space attention mechanism added into the head end and the tail end of the multi-scale residual error network, and MResNet-37 with the multi-scale residual error network without the attention mechanism on the multi-label electrocardiogram classification performance of the Chinese cardiovascular disease database;
the above three types of multi-scale residual error network model structure pairs with attention mechanism are as follows:
where conv15 indicates the size of the convolution kernel is 15, 3@64 indicates that the layer has 64 convolution kernels and 3 blocks of that number, ca and sa indicate channel attention (channel attention) and spatial attention (spatial attention), respectively;
the structure of a multi-scale residual error network model without an attention mechanism is as follows:
the four multi-scale residual error network models are analyzed in the Chinese cardiovascular disease database according to the following experimental results:
as can be seen from the results in the table above, F for SE-MResNet37 is compared to MResNet-37 without the added attention mechanism1The total score is improved by 1.1%, all average indexes are improved, and the fact that the attention mechanism is added in the multi-scale network is proved to improve the multi-label electrocardiogram classification performance to a certain extent. The MS-CSA-ResNet37 with the channel space attention mechanism added to the multi-scale residual network structure has better performance than MResNet-37 without the attention mechanism. However, compared with the CSA-MResNet37 which adds the mechanism to the head and tail ends of the multi-scale residual network model, adding the channel space attention mechanism in the residual structure destroys the performance of the original identity connection to a certain extent, which results in that the model ignores the identity mapping information in learning and training, thereby reducing the learning ability of the model. Therefore, the CSA-MResNet37 model, which incorporates a channel space attention mechanism at both ends of the multi-scale residual network, performs best. Also, as can be seen from the above table, F on a single category1In point of view, the CSA-MResNet37 at the head and tail ends of the multi-scale residual network model with the channel space attention mechanism has obvious advantages.
3. Other database generalization validation
Directly applying a model CSA-MResNet37 trained on a Chinese cardiovascular disease database to 9 disease types on a combined-fat high-new cup electrocardio abnormal event detection test set for testing, and evaluating results according to corresponding indexes, wherein the results are as follows:
verifying generalized classification performance of proposed models on HF-challenge testA
The results in the above table show that the CSA-MResNet-37 model averages F over the SA, TWC and APB classes1The fraction is lower, less than 60 percent, and the generalization performance is lower. This may be because these ECG categories of data are more complex or because the amount of experimental data is small, indicating that the CCDD cannot contain information for all of these ECG categories. But average F over categories of SB and AF1The fractions were all greater than 90% and were comparable to the results tested on CCDD. Finally, on the average scale, both Acc and Rec are lower than the results on CCDD, and Spe and Pre are higher than the CCDD average results. Average F over the data set1A fraction of 85.8%, and the average F obtained on CCDD1The scores are not very different, which proves that the model proposed by the work has good generalization.
4. Results of two-classification of positive and abnormal experiments
And (3) performing experiments on the CSA-MResNet37 model with the best performance in the multi-label electrocardio abnormal event on positive abnormal two categories, modifying the output of the full connection layer to be the maximum threshold output, observing the final performance of the model in the five indexes, and increasing the AUC index to be used as an experimental evaluation standard of the two categories.
The final comprehensive performance on the two-classification questions is shown in the following table;
experimental results of two classification of CSA-MResNet model on CCDD
In conclusion, the channel space attention mechanism is added at the head end and the tail end of the multi-scale residual error network model, so that the performance of the multi-scale residual error network model is effectively improved, and the multi-scale residual error network model has good generalization performance.
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 person skilled in the art should be considered to be within the technical scope of the present invention, and the technical solutions and the inventive concepts thereof according to the present invention should be equivalent or changed within the scope of the present invention.
Claims (7)
1. A multi-label multi-lead electrocardiogram classification method based on a deep neural network is characterized in that: preprocessing an original electrocardiosignal into a 12-lead signal, inputting the 12-lead signal into a multi-scale residual error network model for processing and classification, wherein the multi-scale residual error network model comprises a feature fusion part and a multi-scale feature fusion part, the 12-lead signal forms two branches comprising 32 convolutional layers and an average pool layer after being processed by the feature fusion part, the two branches are respectively endowed with convolutional kernels with different scales, the two branches are output into 1024 nodes after being processed by the multi-scale feature fusion part, and the 1024 nodes output multi-label electrocardio categories through an output layer; the feature fusion part comprises a convolution layer and a first maximum pooling layer, the multi-scale feature fusion part comprises 32 convolution network layers and a full-connection layer, the full-connection layer comprises two second maximum pooling layers, and a channel attention mechanism module and a space attention mechanism module are sequentially arranged between the convolution layer and the first maximum pooling layer and between the 32 convolution layers of each branch and the second maximum pooling layers.
2. The deep neural network-based multi-label multi-lead electrocardiogram classification method of claim 1, which is characterized in that: the convolution layer of the feature fusion part adopts 64 convolution kernels with the scale of 15.
3. The deep neural network-based multi-label multi-lead electrocardiogram classification method of claim 1, which is characterized in that: the 30-layer convolution network layers of the two branches of the feature fusion part respectively adopt 64 convolution kernels with the scale of 3 and 64 convolution kernels with the scale of 7.
4. The deep neural network-based multi-label multi-lead electrocardiogram classification method of claim 1, which is characterized in that: the data processing process of the characteristic fusion part is that the 12-lead signal is processed by the convolutional layer to obtain a data set F, and the data set F is processed by the channel attention mechanism module to obtain a data set MC(F) Data set F and data set MC(F) The element-by-element multiplication is carried out to obtain a data set F ', and the data set F' is processed by a space attention mechanism module to obtain a data set Ms(F '), and comparing the data set F' with the data set Ms(F ') element by element multiplication to obtain a data set F ', and transmitting the data of the data set F ' to the first maximum pooling layer for processing.
5. The deep neural network-based multi-label multi-lead electrocardiogram classification method of claim 4, wherein: the data processing process of the multi-scale feature fusion part is that two branch data which are obtained after the processing of the first maximum pooling layer and comprise 32 convolutional layers and an average pooling layer are respectively transmitted to the second large pooling layer in the full connection layer after the processing of the two branch data which are the same as the processing from the convolutional layer to the first maximum pooling layer in the feature fusion part.
6. The deep neural network-based multi-label multi-lead electrocardiogram classification method according to any one of claims 1-5, wherein the specific processing procedure for preprocessing the original electrocardiogram signals into 12-lead signals is as follows: the extracted 8-lead original electrocardiogram data are converted into 12-lead data, the integral data of the 12-lead data are multiplied by 4.88 microvolts to obtain real electrocardiogram data, and the real electrocardiogram data are subjected to variable length data processing and data enhancement to obtain 12-lead signals.
7. The deep neural network-based multi-label multi-lead electrocardiogram classification method of claim 6, wherein: the variable length data processing is to take 5000 points of real electrocardio data, to omit the initial 1.25s and to take the middle 10s data; and the data enhancement is to perform normalization processing on the acquired middle 10s data by using Gaussian distribution, and then vertically turn over and vertically translate the normalized data by 20 sampling points to increase data samples.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110690012.1A CN113229825A (en) | 2021-06-22 | 2021-06-22 | Deep neural network-based multi-label multi-lead electrocardiogram classification method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110690012.1A CN113229825A (en) | 2021-06-22 | 2021-06-22 | Deep neural network-based multi-label multi-lead electrocardiogram classification method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113229825A true CN113229825A (en) | 2021-08-10 |
Family
ID=77140483
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110690012.1A Pending CN113229825A (en) | 2021-06-22 | 2021-06-22 | Deep neural network-based multi-label multi-lead electrocardiogram classification method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113229825A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113768515A (en) * | 2021-09-17 | 2021-12-10 | 重庆邮电大学 | Electrocardiosignal classification method based on deep convolutional neural network |
| CN114041780A (en) * | 2021-12-08 | 2022-02-15 | 中国科学院计算技术研究所 | Method for monitoring respiration based on data acquired by inertial sensor |
| CN114190952A (en) * | 2021-12-01 | 2022-03-18 | 山东省人工智能研究院 | 12-lead electrocardiosignal multi-label classification method based on lead grouping |
| CN115211866A (en) * | 2022-09-07 | 2022-10-21 | 西南民族大学 | Arrhythmia classification method and system and electronic equipment |
| CN116226763A (en) * | 2023-03-03 | 2023-06-06 | 山东省人工智能研究院 | Multi-label electrocardiosignal classification method based on improved attention mechanism |
| WO2023165005A1 (en) * | 2022-03-04 | 2023-09-07 | 平安科技（深圳）有限公司 | Multi-lead elctrocardiogram signal processing method, device, apparatus, and storage medium |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110251120A (en) * | 2019-06-06 | 2019-09-20 | 山东省计算中心（国家超级计算济南中心） | The electrocardiogram classification method of depth residual error neural network based on attention mechanism |
| CN111329469A (en) * | 2020-03-05 | 2020-06-26 | 广州天嵌计算机科技有限公司 | Arrhythmia prediction method |
| CN111449644A (en) * | 2020-03-19 | 2020-07-28 | 复旦大学 | Bioelectricity signal classification method based on time-frequency transformation and data enhancement technology |
| CN111667921A (en) * | 2020-06-09 | 2020-09-15 | 苏州米特希赛尔人工智能有限公司 | Artificial intelligence ECG atrial fibrillation and arrhythmia detection system |
| CN112450944A (en) * | 2020-12-11 | 2021-03-09 | 郑州大学 | Label correlation guide feature fusion electrocardiogram multi-classification prediction system and method |
| CN112603330A (en) * | 2020-12-28 | 2021-04-06 | 中国科学院微电子研究所 | Electrocardiosignal identification and classification method |
| CN112906748A (en) * | 2021-01-25 | 2021-06-04 | 浙江大学 | 12-lead ECG arrhythmia detection classification model construction method based on residual error network |
| CN112957052A (en) * | 2021-01-25 | 2021-06-15 | 北京工业大学 | Multi-lead electrocardiosignal classification method based on NLF-CNN lead fusion depth network |
-
2021
- 2021-06-22 CN CN202110690012.1A patent/CN113229825A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110251120A (en) * | 2019-06-06 | 2019-09-20 | 山东省计算中心（国家超级计算济南中心） | The electrocardiogram classification method of depth residual error neural network based on attention mechanism |
| CN111329469A (en) * | 2020-03-05 | 2020-06-26 | 广州天嵌计算机科技有限公司 | Arrhythmia prediction method |
| CN111449644A (en) * | 2020-03-19 | 2020-07-28 | 复旦大学 | Bioelectricity signal classification method based on time-frequency transformation and data enhancement technology |
| CN111667921A (en) * | 2020-06-09 | 2020-09-15 | 苏州米特希赛尔人工智能有限公司 | Artificial intelligence ECG atrial fibrillation and arrhythmia detection system |
| CN112450944A (en) * | 2020-12-11 | 2021-03-09 | 郑州大学 | Label correlation guide feature fusion electrocardiogram multi-classification prediction system and method |
| CN112603330A (en) * | 2020-12-28 | 2021-04-06 | 中国科学院微电子研究所 | Electrocardiosignal identification and classification method |
| CN112906748A (en) * | 2021-01-25 | 2021-06-04 | 浙江大学 | 12-lead ECG arrhythmia detection classification model construction method based on residual error network |
| CN112957052A (en) * | 2021-01-25 | 2021-06-15 | 北京工业大学 | Multi-lead electrocardiosignal classification method based on NLF-CNN lead fusion depth network |
Non-Patent Citations (1)
| Title |
|---|
| SHUHONG WANG ET AL.: ""Multiscale Residual Network Based on Channel Spatial Attention Mechanism for Multilabel ECG Classification"", 《JOURNAL OF HEALTHCARE ENGINEERING》 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113768515A (en) * | 2021-09-17 | 2021-12-10 | 重庆邮电大学 | Electrocardiosignal classification method based on deep convolutional neural network |
| CN114190952A (en) * | 2021-12-01 | 2022-03-18 | 山东省人工智能研究院 | 12-lead electrocardiosignal multi-label classification method based on lead grouping |
| CN114041780A (en) * | 2021-12-08 | 2022-02-15 | 中国科学院计算技术研究所 | Method for monitoring respiration based on data acquired by inertial sensor |
| CN114041780B (en) * | 2021-12-08 | 2022-08-30 | 中国科学院计算技术研究所 | Method for monitoring respiration based on data acquired by inertial sensor |
| WO2023165005A1 (en) * | 2022-03-04 | 2023-09-07 | 平安科技（深圳）有限公司 | Multi-lead elctrocardiogram signal processing method, device, apparatus, and storage medium |
| CN115211866A (en) * | 2022-09-07 | 2022-10-21 | 西南民族大学 | Arrhythmia classification method and system and electronic equipment |
| CN115211866B (en) * | 2022-09-07 | 2022-12-27 | 西南民族大学 | Arrhythmia classification method and system and electronic equipment |
| CN116226763A (en) * | 2023-03-03 | 2023-06-06 | 山东省人工智能研究院 | Multi-label electrocardiosignal classification method based on improved attention mechanism |
| CN116226763B (en) * | 2023-03-03 | 2023-10-03 | 山东省人工智能研究院 | Multi-label electrocardiosignal classification method based on improved attention mechanism |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113229825A (en) | Deep neural network-based multi-label multi-lead electrocardiogram classification method | |
| CN111657926B (en) | Arrhythmia classification method based on multi-lead information fusion | |
| CN107822622B (en) | Electrocardiogram diagnosis method and system based on deep convolutional neural network | |
| CN110890155B (en) | Multi-class arrhythmia detection method based on lead attention mechanism | |
| CN108511055B (en) | Ventricular premature beat recognition system and method based on classifier fusion and diagnosis rules | |
| Cao et al. | Atrial fibrillation detection using an improved multi-scale decomposition enhanced residual convolutional neural network | |
| CN108836302A (en) | Electrocardiogram intelligent analysis method and system based on deep neural network | |
| CN109009102B (en) | Electroencephalogram deep learning-based auxiliary diagnosis method and system | |
| CN108090509B (en) | Data length self-adaptive electrocardiogram classification method | |
| CN113080994A (en) | Multi-lead electrocardiosignal classification method based on convolutional neural network | |
| CN109645983A (en) | A kind of uneven beat classification method based on multimode neural network | |
| CN112932433B (en) | Arrhythmia automatic classification method of electrocardiogram | |
| CN111956208B (en) | ECG signal classification method based on ultra-lightweight convolutional neural network | |
| CN112733774A (en) | Light-weight ECG classification method based on combination of BiLSTM and serial-parallel multi-scale CNN | |
| CN110638430A (en) | Multi-task cascade neural network ECG signal arrhythmia disease classification model and method | |
| CN114970637A (en) | Lightweight arrhythmia classification method based on deep learning | |
| CN113288157A (en) | Arrhythmia classification method based on depth separable convolution and improved loss function | |
| CN111261283B (en) | Electrocardiosignal deep neural network modeling method based on pyramid convolution layer | |
| CN112957052B (en) | Multi-lead electrocardiosignal classification method based on NLF-CNN lead fusion depth network | |
| CN114041800A (en) | Electrocardiosignal real-time classification method and device and readable storage medium | |
| Wang et al. | MS-MLP: Multi-scale sampling MLP for ECG classification | |
| CN114767123B (en) | Electrocardiogram scatter diagram anomaly identification and classification method based on multi-modal neural network | |
| CN115500807A (en) | Arrhythmia classification detection method and system based on small convolutional neural network | |
| CN113647954A (en) | Cardiovascular disease identification method, device and medium of two-channel hybrid network model | |
| CN114098757B (en) | ECG signal monitoring method based on quantum particle swarm optimization |
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 |
Application publication date: 20210810 |
|
| RJ01 | Rejection of invention patent application after publication |