CN113171105A - Myocardial ischemia identification and classification method based on integrated CNN - Google Patents
Myocardial ischemia identification and classification method based on integrated CNN Download PDFInfo
- Publication number
- CN113171105A CN113171105A CN202110449036.8A CN202110449036A CN113171105A CN 113171105 A CN113171105 A CN 113171105A CN 202110449036 A CN202110449036 A CN 202110449036A CN 113171105 A CN113171105 A CN 113171105A
- Authority
- CN
- China
- Prior art keywords
- training
- myocardial ischemia
- cnn
- classification method
- integrated
- 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
Images
Abstract
The invention relates to the technical field of electrocardiogram abnormity detection, in particular to a cardiac ischemia identification and classification method based on integrated CNN. The identification and classification method of myocardial ischemia based on integrated CNN comprises the following steps: step 1: two training databases were obtained: an abnormal training database and a normal training database; step 2: preprocessing data in the two training databases to generate an electrocardiosignal sample with a two-dimensional structure; and step 3: constructing an integrated convolutional neural network model; and 4, step 4: constructing a convolutional neural network model; and 5: acquiring a feature set; step 6: training a random forest classifier according to the feature set; and 7: the automatic identification and test are carried out on the sample, the identification accuracy of myocardial ischemia and the discovery efficiency of the myocardial ischemia hiding period are effectively improved, the risk degree of the myocardial ischemia is classified, and the treatment of doctors is assisted.
Description
Technical Field
The invention relates to the technical field of electrocardiogram abnormity detection, in particular to a cardiac ischemia identification and classification method based on integrated CNN.
Background
The electrocardiogram is a conventional and efficient technical means for doctors to diagnose heart diseases clinically, myocardial ischemia belongs to a disease type with high morbidity, and particularly in the middle-aged and the elderly, the myocardial ischemia is not found on time to possibly cause serious arrhythmia and sudden death of heart diseases of patients. The risk stratification and timely discovery of myocardial ischemia have great significance for diagnosis and treatment of diseases, and are closely related to human health. In clinical practice, myocardial ischemia in some patients is manifested as asymptomatic myocardial ischemia, and is easily overlooked by doctors and patients, thereby developing into one of the important risk factors of myocardial infarction and sudden death.
The image examination cannot be widely used due to the price of the patient and psychological factors of the patient. In contrast, the non-invasive electrocardiographic detection technology, especially the conventional 12-lead electrocardiographic technology, is simple and practical, has wider audiences, and plays an important role in myocardial ischemia diagnosis and dangerous stratification as an important supplementary part of medical imaging. The reasonable and orderly selection of the noninvasive electrocardiographic examination can provide electrocardiographic information different from image information, realize the early detection of myocardial ischemia, complement the advantages of the imaging examination and further improve the accuracy and the detectable rate of clinical diagnosis. The existing myocardial ischemia analysis system is not enough to meet the accuracy requirement of clinical application and the problem of risk stratification diagnosis.
Disclosure of Invention
The technical problem to be solved by the invention is as follows: the defects of the prior art are overcome, and a recognition and classification method of myocardial ischemia based on integrated CNN is provided.
The technical scheme adopted by the invention for solving the technical problem is as follows: the CNN-integrated myocardial ischemia identification and classification method comprises the following steps:
step 1: two training databases were obtained: the abnormal training database is a database known to suffer from cardiac ischemia, and the normal training database is a comparison healthy population electrocardiosignal database;
step 2: preprocessing data in the two training databases, reducing noise, intercepting, and performing extraction analysis to generate a two-dimensional structural electrocardiosignal sample;
and step 3: constructing an integrated convolutional neural network model;
and 4, step 4: constructing a convolutional neural network model;
and 5: acquiring a feature set;
step 6: training a random forest classifier according to the feature set;
and 7: the samples are automatically identified and tested.
And 2, reading 12-lead electrocardiosignal data, performing low-pass filtering processing on the data by using a Butterworth filter carried by an MATLAB tool box aiming at an ECG signal sample, removing the influence of noise on an experimental result, uniformly intercepting P points, unifying each lead into a vector with the length of Q dimension by using an extraction and analysis processing method, and converting each electrocardiosignal into a 12X Q two-dimensional matrix as a sample after processing.
Step 3 includes using the two acquired training databases (i.e. the abnormal training database and the normal training database) as training data sets, and setting the training data sets according to a ratio of 3: dividing 1 into a training set and a testing set, randomly extracting 3/4 sample numbers in the training set, extracting k times to obtain k sub-training data sets, training to obtain k CNN classifiers which are CNN1-CNNk respectively and solidifying model parameters.
The structure of the convolutional neural network constructed in the step 4 is five-layer convolutional pooling operation, abstract features of samples are extracted and reduced in dimension, the convolutional neural network is converted into 3 output points through three layers of full connection layers, in a CNN model, average pooling is selected by a pooling layer method, a Softmax classifier is added at the last of the network, the number of nodes is set to be 3, the step length of a convolutional kernel is set to be 1 in each direction, non-zero-filling measures are adopted in boundary processing, a modified linear unit (Relu) is selected as an activation function, average pooling is selected by a pooling method, a Cross entropy (Cross-entropy) is adopted as a network loss function, and self-adaptive momentum estimation is adopted as a model training algorithm. The method comprises the following substeps:
4-1: using the extracted two-dimensional structural sample of the ECG signal as a network input,
4-2: performing five-layer convolution pooling operation, and extracting and reducing dimensional sample abstract features;
4-3: and transforming to 3 output points through three full connection layers, and connecting a Softmax classifier to obtain a final result. Specifically, each stage consists of 3 cascaded layers: convolutional layers, active layers, and pooling layers. According to the templates C1(Size) -S1-C2(Size) -S2-C3(Size) -S3-C4(Size) -S4-C5(Size) -S5-H1-H2-O, C1 to C5 are the number of filters in the first stage to the fifth stage, respectively, Size represents the Size of the convolution kernel, S1 and S2 are sub-sampling factors, H1, H2 and O represent the number of cells in the full connection layer and the output layer. The CNN network structure parameters designed herein are: 32(5) -4-64(5) -4-64(5) -4-128(5) -4-128(5) -4-2000-256-3. Here, the convolution kernels are all one-dimensional convolutions.
In the step 5, the Softmax probability output results of the training data set data are extracted in parallel by using k CNN solidification models and are used as feature vectors, and the extracted feature vectors are fused and used as new feature vectors to obtain feature sets.
In step 6, a CART algorithm is adopted to generate a decision tree, a Gini index is used as an index to select features, and the Gini index of the attribute a is defined as:
selecting the attribute which makes the divided Keyny index minimum in the candidate attribute feature set A as the most divided attribute to be selected, namely:
a*＝argmina∈AGini_index(D,a)。
the candidate attribute feature set A refers to a feature vector X extracted in parallel by k CNN curing models1To XkThe formed set.
And 7, carrying out electrocardiosignal discrimination on the test set by using the trained random forest classifier, identifying normal electrocardiosignals and electrocardiosignals with myocardial ischemia, and carrying out danger hierarchical classification on the electrocardiosignals with myocardial ischemia.
Compared with the prior art, the invention has the following beneficial effects:
according to the method, the CNN is used for extracting the sampling characteristics of the electrocardiosignal samples, the processes of dimension reduction and artificial characteristic selection are omitted, then an integrated convolutional neural network structure is designed, the ECG signal classification problem is modeled based on an electrocardiogram, the strong correlation among the electrocardiosignals is deeply mined, correct identification and danger stratification are carried out on myocardial ischemia, the prediction of myocardial ischemia to a certain degree is completed, the identification accuracy of the myocardial ischemia and the discovery efficiency of the hidden period of the myocardial ischemia are effectively improved, the risk degree of the myocardial ischemia is classified, and the treatment of doctors is assisted;
the electrocardiosignals are learned and classified by utilizing an integrated convolutional neural network and a random forest so as to better identify myocardial ischemia and stratify risks.
Drawings
FIG. 1 is a diagram of the structure of the convolutional neural network of the present invention.
FIG. 2 is a schematic diagram of the integration strategy of the present invention.
Detailed Description
Embodiments of the invention are further described below with reference to the accompanying drawings:
examples
As shown in fig. 1 to 2, the method comprises the following steps:
step 1: two training databases were obtained: the abnormal training database is a database known to suffer from cardiac ischemia, and the normal training database is a comparison healthy population electrocardiosignal database;
step 2: preprocessing data in the two training databases, reducing noise, intercepting, and performing extraction analysis to generate a two-dimensional structural electrocardiosignal sample; specifically, 12-lead electrocardiograph signal data are read, low-pass filtering processing is performed on the data by using a Butterworth filter (Butterworth) carried by an MATLAB tool box for ECG signal samples, the influence of noise on an experimental result is removed, P points are uniformly intercepted, each lead is uniformly formed into a vector with the length being Q dimension by using an analytical processing method, and after the processing, each electrocardiograph signal is converted into a 12-Q two-dimensional matrix to be used as a sample.
And step 3: constructing an integrated convolutional neural network model; specifically, two training databases (i.e., an abnormal training database and a normal training database) are obtained as a training data set, and the training data set is divided into 3: dividing 1 into a training set and a testing set, randomly extracting 3/4 sample numbers in the training set, extracting k times to obtain k sub-training data sets, training to obtain k CNN classifiers which are CNN1-CNNk respectively and solidifying model parameters.
And 4, step 4: constructing a convolutional neural network model; specifically, the constructed convolutional neural network is structured by five-layer convolutional pooling operation, abstract features of extraction and dimensionality reduction samples are converted to 3 output points through three layers of full connection layers, in a CNN model, average pooling is selected by a pooling layer method, a Softmax classifier is added at the end of the network, the number of nodes is set to be 3, the step length of a convolutional kernel is set to be 1 in each direction, non-zero-filling measures are adopted in boundary processing, a modified linear unit (Relu) is selected as an activation function, average pooling is selected by a pooling method, a Cross entropy (Cross-entropy) is adopted as a network loss function, and self-adaptive momentum estimation is adopted as a model training algorithm. The method comprises the following substeps:
4-1: using the extracted two-dimensional structural sample of the ECG signal as a network input,
4-2: performing five-layer convolution pooling operation, and extracting and reducing dimensional sample abstract features;
4-3: and transforming to 3 output points through three full connection layers, and connecting a Softmax classifier to obtain a final result. Specifically, each stage consists of 3 cascaded layers: convolutional layers, active layers, and pooling layers. According to the templates C1(Size) -S1-C2(Size) -S2-C3(Size) -S3-C4(Size) -S4-C5(Size) -S5-H1-H2-O, C1 to C5 are the number of filters in the first stage to the fifth stage, respectively, Size represents the Size of the convolution kernel, S1 and S2 are sub-sampling factors, H1, H2 and O represent the number of cells in the full connection layer and the output layer. The CNN network structure parameters designed herein are: 32(5) -4-64(5) -4-64(5) -4-128(5) -4-128(5) -4-2000-256-3. Here, the convolution kernels are all one-dimensional convolutions.
And 5: acquiring a feature set; specifically, a Softmax probability output result of training data set data is extracted in parallel by using k CNN solidification models to serve as a feature vector, and the extracted feature vector is fused to serve as a new feature vector to obtain a feature set.
Step 6: training a random forest classifier according to the feature set; specifically, a CART algorithm is adopted to generate a decision tree, a Gini index is used as an index to select features, and the Gini index of the attribute a is defined as:
selecting the attribute which makes the divided Keyny index minimum in the candidate attribute feature set A as the most divided attribute to be selected, namely:
a*＝argmina∈AGini_index(D,a)。
the candidate attribute feature set A refers to a feature vector X extracted in parallel by k CNN curing models1To XkThe formed set.
And 7: the samples are automatically identified and tested. Specifically, a trained random forest classifier is used for discriminating electrocardiosignals of a test set, identifying the electrocardiosignals of normal and myocardial ischemia, and classifying the electrocardiosignals of myocardial ischemia in a risk layering way.
Claims (7)
1. A recognition and classification method for myocardial ischemia based on integrated CNN is characterized by comprising the following steps:
step 1: two training databases were obtained: an abnormal training database and a normal training database;
step 2: preprocessing data in the two training databases to generate an electrocardiosignal sample with a two-dimensional structure;
and step 3: constructing an integrated convolutional neural network model;
and 4, step 4: constructing a convolutional neural network model;
and 5: acquiring a feature set;
step 6: training a random forest classifier according to the feature set;
and 7: the samples are automatically identified and tested.
2. The integrated CNN-based myocardial ischemia identification and classification method as claimed in claim 1, wherein the data of 12-lead cardiac electric signals are read in step 2, low-pass filtering processing is performed on the data by using a butterworth filter provided in an MATLAB kit for ECG signal samples, P points are uniformly extracted, each lead is uniformly formed into a vector with a length of Q dimension by using an extraction processing method, and after the processing, each cardiac electric signal is converted into a two-dimensional matrix of 12 × Q as a sample.
3. The integrated CNN-based myocardial ischemia recognition and classification method according to claim 2, wherein the step 3 comprises using the two obtained training databases as training data sets, and the training data sets are classified according to 3: dividing 1 into a training set and a testing set, randomly extracting 3/4 sample numbers in the training set, extracting k times to obtain k sub-training data sets, training to obtain k CNN classifiers which are CNN1-CNNk respectively and solidifying model parameters.
4. The integrated CNN-based myocardial ischemia identification and classification method as claimed in claim 3, wherein the structure of the convolutional neural network constructed in step 4 is a five-layer convolutional pooling operation, comprising the following sub-steps:
4-1: using the extracted two-dimensional structural sample of the ECG signal as a network input,
4-2: performing five-layer convolution pooling operation, and extracting and reducing dimensional sample abstract features;
4-3: and transforming to 3 output points through three full connection layers, and connecting a Softmax classifier to obtain a final result.
5. The integrated CNN-based myocardial ischemia identification and classification method according to claim 4, wherein the Softmax probability output result of the training data set data is extracted in parallel by using k CNN solidification models as feature vectors in the step 5, and the extracted feature vectors are fused as new feature vectors to obtain a feature set.
6. The integrated CNN-based myocardial ischemia identification and classification method according to claim 5, wherein in step 6, a CART algorithm is used to generate a decision tree, and a Gini index is used as an index for feature selection, wherein the Gini index of the attribute a is defined as:
feature vector X to be extracted in parallel by k CNN curing models1To XkThe formed set is used as a candidate attribute feature set A, and the attribute which enables the divided Gini index to be minimum is selected from the candidate attribute feature set A as the most divided attribute to be selected, namely:
a*＝arg mina∈AGini_index(D,a)。
7. the integrated CNN-based myocardial ischemia identification and classification method as claimed in claim 6, wherein in step 7, the trained random forest classifier is used to discriminate cardiac electric signals of the test set, identify cardiac electric signals of normal and myocardial ischemia, and classify cardiac electric signals of myocardial ischemia in risk classification.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110449036.8A CN113171105A (en) | 2021-04-25 | 2021-04-25 | Myocardial ischemia identification and classification method based on integrated CNN |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110449036.8A CN113171105A (en) | 2021-04-25 | 2021-04-25 | Myocardial ischemia identification and classification method based on integrated CNN |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113171105A true CN113171105A (en) | 2021-07-27 |
Family
ID=76925550
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110449036.8A Pending CN113171105A (en) | 2021-04-25 | 2021-04-25 | Myocardial ischemia identification and classification method based on integrated CNN |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113171105A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115985505A (en) * | 2023-01-19 | 2023-04-18 | 北京未磁科技有限公司 | Multidimensional fusion myocardial ischemia auxiliary diagnosis model and construction method thereof |
-
2021
- 2021-04-25 CN CN202110449036.8A patent/CN113171105A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115985505A (en) * | 2023-01-19 | 2023-04-18 | 北京未磁科技有限公司 | Multidimensional fusion myocardial ischemia auxiliary diagnosis model and construction method thereof |
| CN115985505B (en) * | 2023-01-19 | 2023-12-12 | 北京未磁科技有限公司 | Multidimensional fusion myocardial ischemia auxiliary diagnosis model and construction method thereof |
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 | |
| Jiang et al. | A novel multi-module neural network system for imbalanced heartbeats classification | |
| Pourbabaee et al. | Feature leaning with deep convolutional neural networks for screening patients with paroxysmal atrial fibrillation | |
| CN108090509B (en) | Data length self-adaptive electrocardiogram classification method | |
| CN111261277A (en) | Heart disease detection method based on deep learning model | |
| Ma et al. | Multi-classification of arrhythmias using ResNet with CBAM on CWGAN-GP augmented ECG Gramian Angular Summation Field | |
| Islam et al. | New hybrid deep learning approach using BiGRU-BiLSTM and multilayered dilated CNN to detect arrhythmia | |
| Yang et al. | Proegan-ms: A progressive growing generative adversarial networks for electrocardiogram generation | |
| Gajendran et al. | Ecg classification using deep transfer learning | |
| CN108520283B (en) | Medical image classification method constructed based on brain region indirect relation network | |
| Yao et al. | Interpretation of electrocardiogram heartbeat by CNN and GRU | |
| CN111261278A (en) | Heart disease detection method based on deep learning model of three-dimensional image | |
| Yang et al. | A novel method for automated congestive heart failure and coronary artery disease recognition using THC-Net | |
| Li et al. | Automatic heartbeat classification using S-shaped reconstruction and a squeeze-and-excitation residual network | |
| CN113408603B (en) | Coronary artery stenosis degree identification method based on multi-classifier fusion | |
| CN113171105A (en) | Myocardial ischemia identification and classification method based on integrated CNN | |
| CN112932498A (en) | T wave morphology classification system with strong generalization capability based on deep learning | |
| CN116664956A (en) | Image recognition method and system based on multi-task automatic encoder | |
| Sane et al. | Detection of myocardial infarction from 12 lead ECG images | |
| CN114041800B (en) | Electrocardiosignal real-time classification method and device and readable storage medium | |
| Ren et al. | A contrastive predictive coding-based classification framework for healthcare sensor data | |
| CN111759302A (en) | Wearable individual electrocardiogram detection method | |
| Nasim et al. | An Evolutionary-Neural Mechanism for Arrhythmia Classification With Optimum Features Using Single-Lead Electrocardiogram | |
| Li et al. | Efficient ECG classification method for arrhythmia using MODWPT and adaptive incremental broad learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210727 |