CN111613321A - Electrocardiogram stroke auxiliary diagnosis method based on dense convolutional neural network - Google Patents
Electrocardiogram stroke auxiliary diagnosis method based on dense convolutional neural network Download PDFInfo
- Publication number
- CN111613321A CN111613321A CN202010299702.XA CN202010299702A CN111613321A CN 111613321 A CN111613321 A CN 111613321A CN 202010299702 A CN202010299702 A CN 202010299702A CN 111613321 A CN111613321 A CN 111613321A
- Authority
- CN
- China
- Prior art keywords
- dense
- electrocardiogram
- stroke
- network
- neural network
- 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
- 238000003745 diagnosis Methods 0.000 title claims abstract description 31
- 238000000034 method Methods 0.000 title claims abstract description 31
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 29
- 208000006011 Stroke Diseases 0.000 claims abstract description 47
- 238000012549 training Methods 0.000 claims abstract description 34
- 238000012360 testing method Methods 0.000 claims abstract description 19
- 238000010606 normalization Methods 0.000 claims description 19
- 238000011176 pooling Methods 0.000 claims description 15
- 238000010586 diagram Methods 0.000 claims description 6
- 238000012937 correction Methods 0.000 claims description 4
- 230000006870 function Effects 0.000 claims description 4
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000012986 modification Methods 0.000 claims description 3
- 230000004048 modification Effects 0.000 claims description 3
- 230000008569 process Effects 0.000 claims description 2
- 206010008190 Cerebrovascular accident Diseases 0.000 claims 2
- 238000011835 investigation Methods 0.000 claims 1
- 238000013528 artificial neural network Methods 0.000 abstract description 4
- 238000000605 extraction Methods 0.000 description 5
- 238000001514 detection method Methods 0.000 description 4
- 238000004195 computer-aided diagnosis Methods 0.000 description 3
- 201000010099 disease Diseases 0.000 description 3
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 3
- 208000019622 heart disease Diseases 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000012706 support-vector machine Methods 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 2
- 210000004556 brain Anatomy 0.000 description 2
- 230000000747 cardiac effect Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 210000000056 organ Anatomy 0.000 description 2
- 238000000513 principal component analysis Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 206010003658 Atrial Fibrillation Diseases 0.000 description 1
- 208000014644 Brain disease Diseases 0.000 description 1
- 206010012289 Dementia Diseases 0.000 description 1
- WQZGKKKJIJFFOK-GASJEMHNSA-N Glucose Natural products OC[C@H]1OC(O)[C@H](O)[C@@H](O)[C@@H]1O WQZGKKKJIJFFOK-GASJEMHNSA-N 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 206010003119 arrhythmia Diseases 0.000 description 1
- 230000006793 arrhythmia Effects 0.000 description 1
- 230000001174 ascending effect Effects 0.000 description 1
- QVGXLLKOCUKJST-UHFFFAOYSA-N atomic oxygen Chemical compound [O] QVGXLLKOCUKJST-UHFFFAOYSA-N 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000008280 blood Substances 0.000 description 1
- 210000004369 blood Anatomy 0.000 description 1
- 230000036770 blood supply Effects 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 230000004064 dysfunction Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 239000008103 glucose Substances 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000014759 maintenance of location Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000000877 morphologic effect Effects 0.000 description 1
- 229910052760 oxygen Inorganic materials 0.000 description 1
- 239000001301 oxygen Substances 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000000306 qrs interval Methods 0.000 description 1
- 238000012552 review Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000001629 suppression Effects 0.000 description 1
- 235000012167 tiramisu Nutrition 0.000 description 1
- 230000002861 ventricular Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Public Health (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Medical Informatics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Databases & Information Systems (AREA)
- Evolutionary Biology (AREA)
- Pathology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Epidemiology (AREA)
- Primary Health Care (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The invention provides an electrocardiogram stroke auxiliary diagnosis method based on a dense convolutional neural network. The method comprises the steps of firstly establishing an initial database, and distributing a training set and a test set; inputting the training set serving as original training data into a dense convolutional neural network for parameter training to obtain a basic electrocardiogram stroke diagnosis model; and finally, inputting the test set serving as input data into the obtained electrocardiogram stroke diagnosis model for recognition to obtain a recognition result. The invention provides a deep neural network which eliminates artificial constraints and reduces subjective factors, so that a model can automatically extract features and explore the relation between stroke and electrocardiogram, not only can automatically extract special features related to stroke classification, but also can well save computer resources, and can meet the requirements on precision.
Description
Technical Field
The invention relates to the field of electrocardiogram signal research and stroke auxiliary diagnosis, in particular to an electrocardiogram stroke auxiliary diagnosis method based on a dense convolutional neural network.
Background
Stroke is the second leading cause of death and accounts for 11.1% of the total mortality in the world. There are a number of mechanisms that lead to stroke, such as microvascular disease, atrial fibrillation and heart disease. Since the brain relies on oxygen and blood glucose delivered through the circulatory system, interruption of blood supply will lead to microvascular disease, resulting in stroke and dementia. From a system perspective, human organs are functionally related to each other. The brain and heart are two important organs, and dysfunction of either of them affects the function of the other. Heart diseases may respond to brain diseases, and electrocardiograms are a common method of diagnosing many heart diseases, so that the data set relating to electrocardiograms is sufficient. A large data set is virtually impossible for a physician to review macroscopically at one time, and thus there is great potential for study in the data information of an electrocardiogram. Therefore, there is a need to design a computer-aided diagnosis (CAD) system to automatically detect abnormal heartbeats.
To date, most CAD used for classification of cardiac electrical signals relies solely on manual extraction of features from the cardiac electrical signals, including Heart Rate Variability (HRV) indicators, morphological features (P-wave absence, QRS interval prolongation). Traditional machine learning-based methods such as linear discriminant analysis and support vector machine are used to extract specific patterns for disease diagnosis, which can save time for doctors and reduce health cost. These modalities, however, rely on the experience of the physician and the individual's diagnostic philosophy. The study divides the electrocardiogram into individual heartbeats and ignores the connection between heartbeats, assuming that each heartbeat contains only one possibility of arrhythmia. Among these manual feature extraction methods, Support Vector Machines (SVMs) and Hidden Markov Models (HMMs) exhibit better performance. However, the accuracy of these methods is affected by the quality of the manually extracted features. The characteristics of the electrocardiosignals have strong subject dependence, and deeper domain knowledge and professional knowledge are needed for extracting effective characteristics. Therefore, it is necessary to extract the most representative and relevant features for electrocardiographic detection.
In recent years, deep learning neural networks have generated a great deal of interest in multi-dimensional signal processing because they can automatically learn features such as classification and time-series data analysis in computer vision. Deep learning helps to extract hierarchical features from large data in the biomedical field. For example, a multi-channel Electromyogram (EMG) pattern is identified using a depth belief network enhancement algorithm, and a non-stationary electroencephalogram (EEG) signal is identified using a superposition self-encoder (SAE) and Principal Component Analysis (PCA). Residual networks based on Convolutional Neural Networks (CNN) use the MIT ECG database to classify 30s single-lead electrocardiograms into 14 classes, where only the raw data is needed to learn the representative features iteratively. The previous work focuses on semantic information of electrocardiographic data, such as feature extraction or heartbeat type detection (normal, ventricular or supraventricular ectopy, fusion, etc.).
Disclosure of Invention
The invention aims to provide an electrocardiogram stroke auxiliary diagnosis method based on a dense convolutional neural network. The method can utilize the electrocardiogram to play a role in auxiliary diagnosis of the stroke to a certain extent, and obtains the result with the training precision of 99.99 percent and the prediction precision of 85.82 percent through fine adjustment of the model. Compared with the traditional algorithm which needs manual division of features, the method not only can automatically extract the special features related to stroke classification, but also can well save computer resources, and can meet the requirements on precision.
In order to achieve the purpose, the invention adopts the technical scheme that:
an electrocardiogram stroke auxiliary diagnosis method based on a dense convolutional neural network comprises the following steps:
step (1), establishing an initial database, and distributing a training set and a testing set.
The method comprises the following steps of performing initial offset correction on an electrocardiogram by arranging and adding modification to relevant 12-lead electrocardiograms provided by a hospital, specifically deleting pictures with unobvious electrocardiogram waveforms; and according to the existence of stroke, marking 12-lead electrocardiogram with corresponding labels, wherein the label with stroke phenomenon is 1, the label without stroke phenomenon is 0, integrating 12-lead electrocardiogram data and the corresponding labels to obtain an initial database, and dividing the established data set in a random division manner to obtain a training set and a test set;
and (2) inputting the training set serving as original training data into a dense convolutional neural network for parameter training to obtain a basic electrocardiogram stroke diagnosis model, wherein the specific method comprises the following steps:
inputting the training set as original data into a dense convolutional neural network, and performing parameter training by the dense convolutional neural network in a supervised learning mode;
firstly, unifying the image size of related 12-lead electrocardiogram in the training set into 256 × 256, inputting the processed training set data into the initial convolution layer to obtain the characteristic diagram output x0Then x is added0As input to the first dense network, an output x of the first dense network is obtained1Said first dense network comprising 5 dense blocks, then x1After the average pooling operation is carried out, the x after the average pooling operation is carried out1As input to the second dense network, an output x of the second dense network is obtained2Said second dense network comprising 5 dense blocks, then x2After the average pooling operation is carried out, the x after the average pooling operation is carried out2As input to the third dense network, an output x of the third dense network is obtained3Said third dense network comprising 4 dense blocks, the output x of the third dense network to be obtained3Performing batch normalization operation, extracting sample characteristics through a correction linear unit, performing average pooling operation, performing size normalization operation, inputting the normalized sample characteristics into a full-connection layer, and obtaining an output result by using a sigmoid function to obtain a basic electrocardiogram stroke diagnosis model;
in order to prevent overfitting caused by small data quantity, the Dense network Dense Net adopts a characteristic reuse method, wherein each Dense network comprises 4-5 Dense blocks Dense Block; the operation of each dense block comprises Batch Normalization, modified linear unit ReLU, 3 × 3convolution layer and regularization operation Dropout, and the feature output by each dense block is used as a part of feature of each layer for feature merging;
the Batch Normalization enables the extracted feature mean value to be 0 and the variance to be 1, so that the subsequent calculation is facilitated; the modified linear unit ReLU is used for extracting sample features;
the initial convolutional layer is used for carrying out convolution on input original data to obtain an original characteristic diagram; the dense blocks are used for reusing features, and the utilization efficiency of information and gradient in a network is improved;
and (3) inputting the test set serving as input data into the obtained electrocardiogram stroke diagnosis model for recognition to obtain a recognition result.
Firstly, unifying the size of the related 12-lead electrocardiogram images in the test set into 256 × 256, and inputting the processed test set data into an electrocardiogram stroke diagnosis model to obtain a recognition result.
The establishment of the initial database refers to the establishment of a corresponding data set according to requirements on the basis of relevant research on electrocardiogram stroke data provided by a hospital;
the internal structure of the Dense network Dense Net is realized as follows:
at input y0In the case of (2), first, y is0Inputting the data into a first dense block, performing Batch Normalization operation through Batch Normalization, extracting sample characteristics by using a modified linear unit ReLU, and finally performing 3 × 3convolution and regularization dropout operation to obtain output y1', will y0And y1' superposition to get input y of the next dense block1The input of each subsequent dense block is the superposition of the input and the output of the previous dense block.
Compared with the prior art, the invention has the following beneficial effects:
firstly, the traditional stroke diagnosis needs electroencephalogram related research and a professional doctor needs to mark for a long time in the aspect of feature extraction to be identified by a traditional algorithm, and different from the traditional feature extraction algorithm, the invention provides a deep neural network which eliminates artificial constraint and reduces subjective factors, so that a model can automatically extract features and explore the relationship between stroke and electrocardiogram.
Secondly, the invention realizes the classification precision of 99.99% on the training set and the classification precision of 85.82% on the testing set. The detection performance can be further improved as the amount of data increases in the future. And the present invention uses the entire electrocardiogram data image as input to ensure correlation between detected heartbeats. The network architecture uses dense blocks to extract features and shows significant performance on a limited set of data. The depth of the network may be appropriately deepened as the amount of data increases. The 85.82% accuracy of stroke detection means that the electrocardiogram can be used as an auxiliary means for stroke diagnosis to some extent.
Thirdly, the invention adopts a Dense convolutional neural network and a Dense Block operation, which means that each layer can be directly connected with input information and gradient information, and all layers are directly connected on the premise of ensuring maximum information transmission between the layers in the network. Therefore, gradient disappearance is reduced, the situation that features are insufficient due to small data input is avoided, and input information is utilized more effectively. Meanwhile, the connection method enables the transmission of the characteristic information and the gradient information to be more effective, the network is easier to train, and the computer resources can be reasonably utilized. In addition to this, the presence of a Connection such as Dense Connection has a regularizing effect, which, in addition to the above-mentioned reduction in the number of parameters, has a very good suppression of the over-fitting phenomenon. Meanwhile, the feature reuse operation used in the operation is different from the jump link adopted in other segmentation networks, such as Tiramisu networks, other jump links mainly focus on connecting the features obtained by downsampling in the process of upsampling, specifically, the features obtained by upsampling can be integrated corresponding to the features obtained by downsampling, and in DenseNet, on one hand, the operation of upsampling is not performed, on the other hand, the connection used is mainly to perform feature retention after each layer of convolution, namely, the last layer retains the features of all downsampling layers before.
Drawings
FIG. 1 is a flow chart of the method of the present invention;
FIG. 2 is a general block diagram of the dense convolutional neural network of the present invention;
fig. 3 is a schematic diagram of the internal structure of the Dense Net according to an embodiment of the present invention.
Detailed Description
In order to make the aforementioned and other features and advantages of the invention more comprehensible, embodiments accompanied with figures are described in detail below.
The invention provides an electrocardiogram stroke auxiliary diagnosis method based on a dense convolutional neural network, which comprises a training phase and an identification phase, and is shown in figure 1; the training phase comprises the following steps:
firstly, building a required data set. In the embodiment, 98 related 12-lead electrocardiograms are obtained to serve as an initial database of an electrocardiogram stroke auxiliary diagnosis task, wherein the data ratio of stroke to stroke is 1: 1; and finding out the label information corresponding to the original information matching.
And secondly, inputting 90 pictures into the dense convolutional neural network as a training set to train network parameters. The dense network of the dense neural network is mainly an addition operation using features, in particular according to XL=HL([x0,x1,x2..xL-1]). Wherein HLIncluding Batch Normalization, ReLU, 3 × 3 Normalization and Dropout, x0,x1,x2..xL-1For the raw data, XLIs passing through HLThe latter output information, the dense network is shown in FIG. 3, the whole convolutional neural network structure is shown in FIG. 2, and a 12-lead electrocardiogram is used as inputUnifying the size of the input image into 256 x 256, and obtaining a feature map output x through an initial convolution layer0Then x is added0As input to the first dense network, an output x of the first dense network is obtained1Said first dense network comprising 5 dense blocks, then x1After the average pooling operation is carried out, the x after the average pooling operation is carried out1As input to the second dense network, an output x of the second dense network is obtained2Said second dense network comprising 5 dense blocks, then x2After the average pooling operation is carried out, the x after the average pooling operation is carried out2As input to the third dense network, an output x of the third dense network is obtained3Said third dense network comprising 4 dense blocks, the output x of the third dense network to be obtained3And performing batch normalization operation, extracting sample characteristics through a modified linear unit, performing average pooling operation, performing size normalization operation, inputting the normalized sample characteristics into a full-connection layer, and obtaining an output result by using a sigmoid function.
Wherein the Batch Normalization enables the extracted feature mean value to be 0 and the variance to be 1, so that the subsequent calculation is facilitated; 3. the constraint operation of the 3 x 3 reduces the number of input feature maps, thereby reducing the calculated amount in dimension reduction and fusing the features extracted by each channel; the ReLU is used to extract sample features.
Please refer to fig. 3, which is a schematic diagram of an internal structure of a Dense network density Net according to an embodiment of the present invention. The input is y0In the case of (2), first, y is0Inputting the data into a first dense block, performing Batch Normalization operation through Batch Normalization, extracting sample features by using a modified linear unit ReLU, and finally performing 3-by-3 convolution operation and regularization Dropout to obtain an output y1', will y0And y1' performing channel superposition, i.e. changing the number of channels instead of adding weights without changing the number of channels, results in the input y of the next dense block1The input of each subsequent dense block is the superposition of the input and the output of the previous dense block.
Training dense convolutional nervesWhen the network is in use, the network is trained by adopting a supervised learning mode of ascending from bottom to top and randomly initializing weights, and default parameters β are used10.9 and β2Adam optimizer 0.999, batch size 9. The learning rate is set to 0.001. Firstly, taking preprocessed pictures and data as input to train a first hidden layer (namely inputting the preprocessed pictures and the data into an initial convolutional layer), and firstly learning parameters of a dense convolutional neural network during training; furthermore, due to the limitation of the network, sparsity constraint and constraint of prior conditions, the network structure obtains the characteristic with more representation capability than the data per se; after learning to obtain the n-1 th layer, the output of the n-1 th layer is used as the input of the n-1 th layer, and the n-th layer is trained, thereby obtaining the parameters of each layer.
And saving the adjusted parameters to obtain a relevant model.
Step one, test data (8 pictures) obtained randomly before are used as testing to test the accuracy of the model obtained in the training stage, and the result is stored;
and secondly, inputting the preprocessed data or the test data into the obtained model for recognition to obtain a recognition result corresponding to the test set.
The above embodiments are only for illustrating the technical solution of the present invention and not for limiting the same, and a person skilled in the art can make modifications or equivalent substitutions to the technical solution of the present invention without departing from the spirit and scope of the present invention, and the scope of the present invention should be determined by the claims.
Claims (8)
1. An electrocardiogram stroke auxiliary diagnosis method based on a dense convolutional neural network is characterized by comprising the following steps:
step (1), establishing an initial database, and distributing a training set and a test set;
inputting the training set serving as original training data into a dense convolutional neural network for parameter training to obtain a basic electrocardiogram stroke diagnosis model;
step (3), inputting the test set as input data into the obtained electrocardiogram stroke diagnosis model for recognition to obtain a recognition result;
firstly, unifying the size of the related 12-lead electrocardiogram images in the test set into 256 × 256, and inputting the processed test set data into an electrocardiogram stroke diagnosis model to obtain a recognition result.
2. The electrocardiogram stroke auxiliary diagnosis method based on the dense convolutional neural network as claimed in claim 1, wherein the step (1) of establishing an initial database, and distributing a training set and a testing set comprises the following specific steps:
the method comprises the following steps of performing initial offset correction on an electrocardiogram by arranging and adding modification to relevant 12-lead electrocardiograms provided by a hospital, specifically deleting pictures with unobvious electrocardiogram waveforms; and according to the fact that the 12-lead electrocardiogram is marked with the corresponding label in the process of stroke, wherein the label with the stroke phenomenon is 1, the label without the stroke phenomenon is 0, the 12-lead electrocardiogram data and the corresponding label are integrated to obtain an initial database, and the established data set is divided in a random dividing mode to obtain a training set and a testing set.
3. The electrocardiogram apoplexy auxiliary diagnosis method based on the dense convolutional neural network as claimed in claim 2, wherein the step (2) is to input the training set as the original training data into the dense convolutional neural network for parameter training to obtain the basic electrocardiogram apoplexy diagnosis model, and the specific method is as follows:
inputting the training set as original data into a dense convolutional neural network, and performing parameter training by the dense convolutional neural network in a supervised learning mode;
firstly, unifying the image size of related 12-lead electrocardiogram in the training set into 256 × 256, inputting the processed training set data into the initial convolution layer to obtain the characteristic diagram output x0Then x is added0As input to the first dense network, an output x of the first dense network is obtained1Said first dense network comprising 5 dense blocks, then x1After the average pooling operation is carried out, the x after the average pooling operation is carried out1As input to the second dense network, an output x of the second dense network is obtained2Said second dense network comprising 5 dense blocks, then x2After the average pooling operation is carried out, the x after the average pooling operation is carried out2As input to the third dense network, an output x of the third dense network is obtained3Said third dense network comprising 4 dense blocks, the output x of the third dense network to be obtained3And performing batch normalization operation, then extracting sample characteristics through a correction linear unit, performing average pooling operation, inputting the sample characteristics into a full-connection layer after performing size normalization operation, and obtaining an output result by using a sigmoid function to obtain a basic electrocardiogram stroke diagnosis model.
4. The electrocardiogram stroke auxiliary diagnosis method based on the Dense convolutional neural network as claimed in claim 3, wherein the Dense network Dense Net adopts a feature reuse method to prevent overfitting caused by small data volume, wherein each Dense network comprises 4-5 Dense blocks Dense Block; the operation of each dense block comprises Batch Normalization, modified linear unit ReLU, 3 × 3convolution layer and regularization operation Dropout, and the feature output by each dense block is combined as a part of the feature of each layer.
5. The intensive convolutional neural network-based electrocardiogram stroke auxiliary diagnosis method as claimed in claim 4, wherein the Batch Normalization makes the extracted feature mean 0 and variance 1, which is convenient for the subsequent calculation; the modified linear unit ReLU is used for extracting sample features.
6. The electrocardiogram stroke auxiliary diagnosis method based on the dense convolutional neural network as claimed in claim 5, wherein the initial convolutional layer is used for performing convolution on the input raw data to obtain a raw feature map; the dense blocks are used for feature reuse, and utilization efficiency of information and gradients in a network is improved.
7. The method as claimed in claim 6, wherein the initial database is created by constructing a corresponding data set according to requirements based on the investigation on the stroke data provided by hospitals.
8. The electrocardiogram stroke auxiliary diagnosis method based on the Dense convolutional neural network as claimed in claim 7, wherein the internal structure of the Dense network Dense Net is realized as follows:
at input y0In the case of (2), first, y is0Inputting the data into a first dense block, performing Batch Normalization operation through Batch Normalization, extracting sample characteristics by using a modified linear unit ReLU, and finally performing 3 × 3convolution and regularization dropout operation to obtain output y1', will y0And y1' superposition to get input y of the next dense block1The input of each subsequent dense block is the superposition of the input and the output of the previous dense block.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010299702.XA CN111613321A (en) | 2020-04-16 | 2020-04-16 | Electrocardiogram stroke auxiliary diagnosis method based on dense convolutional neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010299702.XA CN111613321A (en) | 2020-04-16 | 2020-04-16 | Electrocardiogram stroke auxiliary diagnosis method based on dense convolutional neural network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN111613321A true CN111613321A (en) | 2020-09-01 |
Family
ID=72199576
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010299702.XA Pending CN111613321A (en) | 2020-04-16 | 2020-04-16 | Electrocardiogram stroke auxiliary diagnosis method based on dense convolutional neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111613321A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112600772A (en) * | 2020-12-09 | 2021-04-02 | 齐鲁工业大学 | OFDM channel estimation and signal detection method based on data-driven neural network |
CN112836729A (en) * | 2021-01-19 | 2021-05-25 | 中南大学 | Construction method of image classification model and image classification method |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106073755A (en) * | 2016-05-27 | 2016-11-09 | 成都信汇聚源科技有限公司 | The implementation method that in a kind of miniature holter devices, atrial fibrillation identifies automatically |
CN107145702A (en) * | 2016-03-01 | 2017-09-08 | 西门子保健有限责任公司 | Cardiogenic embolic type of stroke Risk Forecast System and method based on medical image |
CN107887010A (en) * | 2017-11-29 | 2018-04-06 | 天津中科爱乐芙医疗科技有限公司 | A kind of angiocardiopathy data acquisition examines platform with dividing |
CN109658399A (en) * | 2018-12-13 | 2019-04-19 | 深圳先进技术研究院 | A kind of neck patch image-recognizing method and device |
US20190246904A1 (en) * | 2016-10-20 | 2019-08-15 | Jlk Inspection | Stroke diagnosis and prognosis prediction method and system |
CN110491506A (en) * | 2019-05-16 | 2019-11-22 | 中国医药大学附设医院 | Auricular fibrillation prediction model and its forecasting system |
-
2020
- 2020-04-16 CN CN202010299702.XA patent/CN111613321A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107145702A (en) * | 2016-03-01 | 2017-09-08 | 西门子保健有限责任公司 | Cardiogenic embolic type of stroke Risk Forecast System and method based on medical image |
CN106073755A (en) * | 2016-05-27 | 2016-11-09 | 成都信汇聚源科技有限公司 | The implementation method that in a kind of miniature holter devices, atrial fibrillation identifies automatically |
US20190246904A1 (en) * | 2016-10-20 | 2019-08-15 | Jlk Inspection | Stroke diagnosis and prognosis prediction method and system |
CN107887010A (en) * | 2017-11-29 | 2018-04-06 | 天津中科爱乐芙医疗科技有限公司 | A kind of angiocardiopathy data acquisition examines platform with dividing |
CN109658399A (en) * | 2018-12-13 | 2019-04-19 | 深圳先进技术研究院 | A kind of neck patch image-recognizing method and device |
CN110491506A (en) * | 2019-05-16 | 2019-11-22 | 中国医药大学附设医院 | Auricular fibrillation prediction model and its forecasting system |
Non-Patent Citations (1)
Title |
---|
熊鹏 等: "基于密集连接卷积神经网络的下壁心肌梗死检测" * |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112600772A (en) * | 2020-12-09 | 2021-04-02 | 齐鲁工业大学 | OFDM channel estimation and signal detection method based on data-driven neural network |
CN112600772B (en) * | 2020-12-09 | 2022-05-17 | 齐鲁工业大学 | OFDM channel estimation and signal detection method based on data-driven neural network |
CN112836729A (en) * | 2021-01-19 | 2021-05-25 | 中南大学 | Construction method of image classification model and image classification method |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Oh et al. | Automated diagnosis of arrhythmia using combination of CNN and LSTM techniques with variable length heart beats | |
CN111449645B (en) | Intelligent classification and identification method for electrocardiogram and heartbeat | |
CN111990989A (en) | Electrocardiosignal identification method based on generation countermeasure and convolution cyclic network | |
CN110890155A (en) | Multi-class arrhythmia detection method based on lead attention mechanism | |
CN110120055B (en) | Fundus fluorography image non-perfusion area automatic segmentation method based on deep learning | |
CN112906748A (en) | 12-lead ECG arrhythmia detection classification model construction method based on residual error network | |
Ma et al. | Multi-classification of arrhythmias using ResNet with CBAM on CWGAN-GP augmented ECG Gramian Angular Summation Field | |
CN113180688B (en) | Coronary heart disease electrocardiogram screening system and method based on residual error neural network | |
Yoon et al. | Discovering hidden information in biosignals from patients using artificial intelligence | |
Wang et al. | Automated classification model with OTSU and CNN method for premature ventricular contraction detection | |
Liu et al. | A simple and effective method for detecting myocardial infarction based on deep convolutional neural network | |
CN111613321A (en) | Electrocardiogram stroke auxiliary diagnosis method based on dense convolutional neural network | |
CN115024725A (en) | Tumor treatment aid decision-making system integrating psychological state multi-parameter detection | |
Daydulo et al. | Cardiac arrhythmia detection using deep learning approach and time frequency representation of ECG signals | |
CN114190950B (en) | Electrocardiogram intelligent analysis method for noise-containing label and electrocardiograph | |
CN111557660A (en) | Arrhythmia identification method under sub-population deep learning framework | |
CN115337018A (en) | Electrocardiosignal classification method and system based on overall dynamic characteristics | |
Xu et al. | Arrhythmia detection using gated recurrent unit network with ECG signals | |
Neifar et al. | Deep Generative Models for Physiological Signals: A Systematic Literature Review | |
CN116269426A (en) | Twelve-lead ECG-assisted heart disease multi-mode fusion screening method | |
Dissanayake et al. | DConv-LSTM-Net: A Novel Architecture for Single and 12-Lead ECG Anomaly Detection. | |
WO2021031155A1 (en) | Method and device for multi-scale characteristic extraction based on ecg | |
Wang et al. | Pay attention and watch temporal correlation: a novel 1-D convolutional neural network for ECG record classification | |
CN110786847B (en) | Electrocardiogram signal library building method and analysis method | |
Zhang et al. | An atrial fibrillation classification method based on an outlier data filtering strategy and modified residual block of the feature pyramid network |
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: 20200901 |