CN114176600B - Electrocardiogram ST segment abnormality discrimination system based on causal analysis - Google Patents
Electrocardiogram ST segment abnormality discrimination system based on causal analysis Download PDFInfo
- Publication number
- CN114176600B CN114176600B CN202111631456.4A CN202111631456A CN114176600B CN 114176600 B CN114176600 B CN 114176600B CN 202111631456 A CN202111631456 A CN 202111631456A CN 114176600 B CN114176600 B CN 114176600B
- Authority
- CN
- China
- Prior art keywords
- data
- causal
- electrocardiogram
- module
- segment
- 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.)
- Active
Links
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/316—Modalities, i.e. specific diagnostic methods
- A61B5/318—Heart-related electrical modalities, e.g. electrocardiography [ECG]
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/316—Modalities, i.e. specific diagnostic methods
- A61B5/318—Heart-related electrical modalities, e.g. electrocardiography [ECG]
- A61B5/346—Analysis of electrocardiograms
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/316—Modalities, i.e. specific diagnostic methods
- A61B5/318—Heart-related electrical modalities, e.g. electrocardiography [ECG]
- A61B5/346—Analysis of electrocardiograms
- A61B5/349—Detecting specific parameters of the electrocardiograph cycle
- A61B5/352—Detecting R peaks, e.g. for synchronising diagnostic apparatus; Estimating R-R interval
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/316—Modalities, i.e. specific diagnostic methods
- A61B5/318—Heart-related electrical modalities, e.g. electrocardiography [ECG]
- A61B5/346—Analysis of electrocardiograms
- A61B5/349—Detecting specific parameters of the electrocardiograph cycle
- A61B5/358—Detecting ST segments
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7203—Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/20—Ensemble learning
-
- 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
- Health & Medical Sciences (AREA)
- Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Physics & Mathematics (AREA)
- Cardiology (AREA)
- Molecular Biology (AREA)
- Theoretical Computer Science (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Surgery (AREA)
- Public Health (AREA)
- Pathology (AREA)
- Veterinary Medicine (AREA)
- Heart & Thoracic Surgery (AREA)
- Artificial Intelligence (AREA)
- Animal Behavior & Ethology (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Signal Processing (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Psychiatry (AREA)
- Physiology (AREA)
- Computational Linguistics (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The application provides an electrocardiogram ST segment abnormality discrimination system based on causal analysis, which comprises: extracting and preprocessing the extracardiac physical sign factor data, generating a weighted adjacent matrix among data variables based on the preprocessed extracardiac physical sign factor data, and generating a Bayesian network G by extracting non-zero weights from the weighted adjacent matrix 0 The method comprises the steps of carrying out a first treatment on the surface of the Computing Bayesian network G 0 Causal effect estimators of each path, and generating causal network G based on causal effect estimators by adjusting network structure 1 The method comprises the steps of carrying out a first treatment on the surface of the Extracting 12-lead data from an electrocardiogram, and preprocessing the 12-lead data to obtain preprocessed 12-lead data; acquiring 10-dimensional electrocardiographic features based on the 12-lead data; preprocessing the preprocessed 12-lead data and 10-dimensional electrocardio characteristics, and extracting depth through a convolution residual neutral networkA degree feature; combining the depth characteristic with causal mechanism variable data, and inputting a decision tree to obtain the prediction probability of the occurrence of the abnormal electrocardio characteristic st segment in the electrocardiogram.
Description
Technical Field
The application relates to the technical field of artificial intelligence, in particular to an electrocardiogram ST-segment abnormality judging system based on causal analysis, and more particularly relates to a judging system for judging whether an electrocardiogram has ST-segment abnormality and abnormality type through causal analysis and a tree classifier.
Background
Coronary atherosclerotic heart disease, abbreviated as coronary heart disease, refers to ischemic and anoxic heart disease caused by coronary atherosclerosis and coronary stenosis, and arrhythmia is a common complication thereof. Coronary heart disease is one of the important causes of death for the elderly, and the incidence rate and the age are in a proportional relationship. Clinically, angina pectoris, myocardial infarction and other conditions are often manifested, and even death is caused by arrhythmia and heart failure. The gold standard for diagnosing coronary heart disease is coronary angiography at present, but the gold standard is not popular because of high cost and certain risk. At present, twelve-lead electrocardiography is used as an important auxiliary examination means for detecting coronary heart disease, and at present electrocardiography screening is mainly carried out by a doctor observing electrocardiography and then giving out electrocardiography related diagnosis, and judging whether coronary heart disease is possible or not according to the diagnosis.
The composition of each wave and wave band of the electrocardiogram is P wave, PR interval, QRS wave group, J point, ST segment, T wave, U wave and QT interval in sequence. Coronary heart disease is mainly divided into chronic myocardial ischemia, which is mainly manifested by elevation of the ST segment, depression of the ST segment, low level of the T wave, inversion of the T wave, and acute myocardial ischemia, which is mainly manifested by elevation of the ST segment, pathological Q wave, inversion of the T wave, or obvious high tip.
The ST segment represents a period of time during which complete repolarization of the ventricular muscle has not yet begun. At this time, the ventricular muscles of each part are in a depolarization state, and there is no potential difference between cells, so the ST segment should be on an equipotential line under normal conditions. When ischemia or necrosis of cardiac muscle occurs in a certain part, the ventricular chamber still has potential difference after the depolarization is finished, and the myocardial chamber is represented as ST-segment deviation on an electrocardiogram. The latter T wave represents the repolarization of the ventricle, the lead in the QRS main wave direction, which should be the same as the QRS main wave direction. The change in T-waves on an electrocardiogram is affected by a number of factors. The Q wave is formed by the vector in the right front of the room-space depolarization, and normally the time limit is not more than 0.03s (except III and avR leads), and the depth is not more than 1/4 of the R wave of the same lead.
However, most of the electrocardiographic studies currently performed use internationally recognized standard arrhythmia databases, including MIT-BIH databases provided by the american college of bureau of technology, AHA databases provided by the american heart society, and eutectoid CSE databases, which are concentrated on arrhythmia classification, and relatively, myocardial ischemia and myocardial infarction are less studied, and mainly data are more difficult to obtain. But myocardial ischemia and myocardial infarction have high incidence rate, represent the coronary heart disease and are more diseases with extremely high mortality rate, and have high research value. Moreover, the artificial intelligence developed at the present stage can only be called weak artificial intelligence in the strict sense, and is difficult to be applied to the high-risk fields such as medical treatment. Most of these are due to the lack of interpretability, potential instability, and black box problems of neural networks. Deep learning is a statistical model based on correlation that easily learns "pseudo-relationships (spurious relation)" in the data, rather than causal relationships, thereby reducing generalization and resistance to attacks. For example, in the medical field, systems for prognosis of patients with pneumonia that appear to be very accurate are highly dependent on spurious correlations in the data set. Deep learning model systems predict that patients with a history of asthma have a lower risk of dying from pneumonia, but because asthmatics get a faster, better focus to have their mortality rate lower.
Coronary heart disease incidence rate and mortality rate are high, the patient's twelve electrocardiograph is often represented by ST-T segment change and pathological Q wave, and at present, doctors mainly depend on identifying related abnormalities of the electrocardiograph to screen out patients possibly suffering from coronary heart disease. However, in underdeveloped areas, limited to medical levels, erroneous or missed decisions may be possible. Electrocardiographs also give electrocardiographic diagnostic results, but are limited to the complexity of electrocardiographic data and the accuracy of electrocardiographic diagnostic results is low. In recent years, there have been many researchers applying deep learning techniques to electrocardiographic diagnosis.
Patent document CN112006678A (application number: 202010944862.5) discloses an electrocardiogram abnormal recognition method and system based on the combination of AlexNet and transfer learning, which belong to the technical field of feature extraction, classification and prediction, and the technical problem to be solved by the application is how to accurately and efficiently complete electrocardiogram abnormal recognition by combining an AlexNet deep convolutional neural network and transfer learning, so that the dependence on sample data capacity can be eliminated, and the characteristics of data samples can be automatically learned, and the technical scheme is as follows: the method comprises the following steps: pretreatment: converting each electrocardiogram signal in the data set into an electrocardiogram image, and cutting the abnormal electrocardiogram in different directions; feature extraction: the image after data enhancement is put into a pre-trained model for training, the pre-trained electrocardiogram image is used as input of an AlexNet model, the characteristics are automatically extracted, and the AlexNet model pre-trained by an image Net data set is utilized for transfer learning; classification prediction: and (3) putting the high features obtained by the pre-trained AlexNet deep convolutional neural network model into a support vector machine for electrocardiographic classification.
Disclosure of Invention
Aiming at the defects in the prior art, the application aims to provide an electrocardiogram ST segment abnormality discrimination system based on causal analysis.
The application provides an electrocardiogram ST segment abnormality discrimination system based on causal analysis, which comprises:
a bayesian network generation module: extracting and preprocessing the extracardiac physical sign factor data to obtain preprocessed extracardiac physical sign factor data, generating a weighted adjacency matrix among data variables based on the preprocessed extracardiac physical sign factor data, and extracting non-zero weight values by the weighted adjacency matrix to generate a Bayesian network G 0 Establishing a causal link mechanism between the abnormal electrocardiographic st section and the extracardiac physical sign factor data;
and the directional adjustment module is used for: computing Bayesian network G 0 Causal effect estimators of each path, and generating causal network G based on causal effect estimators by adjusting network structure 1 Screening out a preset causal mechanism variable related to the abnormality of the st section of the electrocardiogram according to a preset requirement;
an electrocardiogram signal processing module: extracting 12-lead data from an electrocardiogram, and preprocessing the 12-lead data to obtain preprocessed 12-lead data; acquiring 10-dimensional electrocardio characteristics based on the preprocessed 12-lead data;
and a judging module: preprocessing the preprocessed 12-lead data and 10-dimensional electrocardio features, and extracting depth features through a convolution residual neural network; combining the depth characteristic with causal mechanism variable data, and inputting a decision tree to obtain the prediction probability of the occurrence of the abnormal electrocardio characteristic st segment in the electrocardiogram.
Preferably, in the bayesian network generating module,
module M1.1: extracting the extracardiac physical sign factor data, and discretizing the extracted extracardiac physical sign factor data to obtain discretized extracardiac physical sign factor data;
module M1.2: generating a weighted adjacency matrix between data variables by using a NOTEARS structure learning algorithm and a K2 scoring algorithm from the discretized extracardiac sign factor data;
module M1.3: the weighted adjacency matrix carries out secondary processing on matrix weights through a hard threshold rule, extracts non-zero weights and generates a Bayesian network G 0 。
Preferably, the module M1.3 employs: the weighted adjacent matrix carries out secondary processing on matrix weights through a hard threshold omega=0.1, non-zero weights are extracted, and a Bayesian network G is generated 0 。
Preferably, in the directional adjustment module,
module M2.1: calculating a Bayesian network G by adopting a front door adjustment formula and a back door adjustment formula 0 A causal effect estimator for each path;
module M2.2: intervention operation is performed by using do operation based on causal effect estimation, network structure is adjusted, and causal network G is generated 1 。
Preferably, in said module M2.1, the front door adjustment formula, the rear door adjustment formula and the directed separation rule are used to infer different causal relationships between the motion variables and the st segment anomalies, including smoking, drinking.
Preferably, in the electrocardiogram signal processing module,
module M3.1: extracting 12-lead sequence data from an XML file generated by an electrocardiograph;
module M3.2: carrying out 7-layer decomposition denoising treatment on the electrocardiosignal by using symlets4 wavelet on the 12-lead sequence data to obtain denoised 12-lead sequence data;
module M3.3: and searching the position of R wave in the electrocardio sequence data based on the denoised 12-lead sequence data, and performing heart beat segmentation by taking the R wave as a reference to obtain 10-dimensional electrocardio characteristics.
Preferably, in the discriminating module,
module M4.1: carrying out data enhancement processing of translation and scaling on the denoised 12-lead sequence data and the 10-dimensional electrocardio characteristics to obtain processed data;
module M4.2: extracting depth features from the processed data by using a ResneXt50+SE depth residual error depth neural network;
module M4.3: and combining the depth characteristic with causal mechanism variable data, and inputting a trained xgBoost tree model to obtain the prediction probability of the occurrence of the abnormal electrocardio characteristic st section in the electrocardiogram.
Preferably, the resnex50+se convolution residual depth neural network employs: the processed data enter a ResneXt50+SE depth residual error depth neural network, enter a bottleneck layer with repeated preset quantity after passing through a 1-dimensional convolution layer and a maximum pooling layer, and finally are converted into one-dimensional data after passing through a global pooling layer and a global maximum pooling layer, and then a three-dimensional predicted value is obtained through a full-connection layer and a Softmax function layer.
Preferably, in the bottleneck layer: the data dimension is changed by using the operation of the transformation dimension with the convolution kernel of 1, then the data dimension is changed by using the grouping convolution, and finally the data with the corresponding dimension is obtained by using the operation of the transformation dimension with the convolution kernel of 1.
Preferably, the bottleneck layer further comprises: and the correlation among channels is learned through the SE module, so that the accuracy of the ResneXt50+SE convolution residual error deep neural network is improved.
Compared with the prior art, the application has the following beneficial effects:
1. the application has low installation and use cost, can automatically screen, and has higher accuracy than the diagnosis of an electrocardiograph. The method can be used by doctors, so that erroneous judgment or missed judgment is reduced, and the workload of the doctors is reduced;
2. compared with other deep network electrocardiographic researches, the method and the system have the advantages that the causal inference is combined, the interpretability of the model can be better improved, other beneficial characteristics are easy to be added into the diagnosis system, and the expansibility is good;
3. the method is simple, can effectively analyze the causal link mechanism between the st section abnormality and the extracardiac factor, improves the prediction performance of the model in st section abnormality classification, and can rapidly give out the diagnosis of whether the electrocardiograph st section is abnormal or not.
Drawings
Other features, objects and advantages of the present application will become more apparent upon reading of the detailed description of non-limiting embodiments, given with reference to the accompanying drawings in which:
fig. 1 is a schematic diagram of an electrocardiogram ST segment abnormality discrimination system based on causal analysis.
FIG. 2 is a causal extracardiac element screening flow.
FIG. 3 is a block diagram of a ResneXt50+XGBoost network.
Detailed Description
The present application will be described in detail with reference to specific examples. The following examples will assist those skilled in the art in further understanding the present application, but are not intended to limit the application in any way. It should be noted that variations and modifications could be made by those skilled in the art without departing from the inventive concept. These are all within the scope of the present application.
Example 1
According to the application, an electrocardiogram ST segment abnormality discrimination system based on causal analysis is provided, as shown in figures 1 to 3, and comprises:
a bayesian network generation module: extracting and preprocessing the extracardiac physical sign factor data to obtain preprocessed extracardiac physical sign factor data, generating a weighted adjacency matrix among data variables based on the preprocessed extracardiac physical sign factor data, and extracting non-zero weight values by the weighted adjacency matrix to generate a Bayesian network G 0 Jian (Chinese character of 'Jian')A causal link mechanism between the abnormal st segment of the vertical electrocardiogram and the data of the extracardiac physical sign factors; the application uses a large amount of clinical electrocardiogram data including original xml files, clinical diagnosis results of doctors on electrocardiogram and other sign data of interviewees, ten variables are screened from the original sign data based on priori knowledge, after the data are prepared, a weighted adjacent matrix between the data variables is generated by using a NOTEARS structure learning algorithm and a K2 scoring algorithm, a primary causal network model is obtained by adopting a mode of enabling the adjacent matrix to pass through a hard threshold omega=0.1 and a causal link mechanism between ST segment abnormality and sign variables of the electrocardiogram is established.
The extracardiac physical sign factor data includes an intracardiac factor and an extracardiac factor; the endocardium typically comprises an impaired sinus function and an acute myocardial injury; the extracardiac factors can be broadly classified as advanced age, vagal excitation, etc. The present application relates generally to extracardiac elements.
And the directional adjustment module is used for: computing Bayesian network G 0 Causal effect estimators of each path, and generating causal network G based on causal effect estimators by adjusting network structure 1 Screening out a preset causal mechanism variable related to the abnormality of the st section of the electrocardiogram according to a preset requirement; wherein, for the Bayesian network G generated by the Bayesian network generation module 0 Performing structure adjustment, and calculating Bayesian network G 0 The causal effect estimation quantity of each path adopts a front gate adjustment formula and a back gate adjustment formula to eliminate the existence condition of each path one by one, and the Bayesian network G is pruned and adjusted 0 And 8 causal mechanism variables related to the ST segment abnormality of the electrocardiogram are selected.
An electrocardiogram signal processing module: extracting 12-lead data from an electrocardiogram, and preprocessing the 12-lead data to obtain preprocessed 12-lead data; acquiring 10-dimensional electrocardio characteristics based on the preprocessed 12-lead data; the application uses a large amount of clinical electrocardiogram data comprising an original xml file and clinical diagnosis results of a doctor on an electrocardiogram, wherein the electrocardiogram data contains various noises including baseline drift, myoelectric noise, power frequency noise and the like. The method comprises the steps of removing various noises on an electrocardiogram by using wavelets, removing data with poor recording and poor quality by combining relevant information and electrocardiographic waveform information diagnosed by doctors, and then performing heart beat segmentation on the data to uniformly fill the data into fixed lengths. According to morphological characteristics of ST segment abnormality, ten types of characteristics are designed and extracted for subsequent training.
And a judging module: preprocessing the preprocessed 12-lead data and 10-dimensional electrocardio features, and extracting depth features through a convolution residual neural network; combining the depth characteristic with causal mechanism variable data, and inputting a decision tree to obtain the prediction probability of the occurrence of the abnormal electrocardio characteristic st segment in the electrocardiogram. The application relates to a method for acquiring data of an electrocardiogram, which comprises the steps of carrying out data enhancement on data generated by an electrocardiogram signal processing module, carrying out translation and scaling on original electrocardiogram sequence data, inputting the translated and scaled data into a RestneXt50 depth residual error network training comprising an SE module, and carrying out transposition one-dimensional convolution on all two-dimensional convolution operations in the ResneXt50 in order to adapt to the characteristics of the electrocardiogram data, wherein before a full connection layer, the SE module mainly learns the correlation between channels, thereby improving the network performance. The training data is input, 512-dimensional depth features are directly extracted by using an electrocardiogram signal processing module, then the depth features are spliced with electrocardiographic waveform features and causal mechanism features, an xgboost tree model is used for training, and xgboost relevant super-parameters are adjusted to learn and obtain a probability prediction model. When the method is used, only the electrocardiogram xml file is placed at a program appointed position, 512-dimensional characteristics extracted by the neural network are obtained through the neural network, then the characteristics are combined with extracardiac element information such as patient signs and the like, and an xgboost model is input, so that the probability and the abnormal type of the abnormal electrocardio characteristics of the ST segment of the detector can be obtained.
In particular, in the bayesian network generating module,
module M1.1: extracting the extracardiac physical sign factor data, and discretizing the extracted extracardiac physical sign factor data to obtain discretized extracardiac physical sign factor data;
module M1.2: generating a weighted adjacency matrix between data variables by using a NOTEARS structure learning algorithm and a K2 scoring algorithm from the discretized extracardiac sign factor data;
module M1.3: the weighted adjacency matrix carries out secondary processing on matrix weights through a hard threshold rule, extracts non-zero weights and generates a Bayesian network G 0 。
In particular, in the directional adjustment module,
module M2.1: calculating a Bayesian network G by adopting a front door adjustment formula and a back door adjustment formula 0 A causal effect estimator for each path;
module M2.2: intervention operation is performed by using do operation based on causal effect estimation, network structure is adjusted, and causal network G is generated 1 。
Specifically, in said module M2.1, different causal relationships between smoking, drinking and movement variables and st segment anomalies are inferred using front door adjustment formulas, rear door adjustment formulas and directed separation formulas.
In particular, in the electrocardiogram signal processing module,
module M3.1: extracting 12-lead sequence data from an XML file generated by an electrocardiograph;
module M3.2: carrying out 7-layer decomposition denoising treatment on the electrocardiosignal by using symlets4 wavelet on the 12-lead sequence data to obtain denoised 12-lead sequence data;
module M3.3: and searching the position of R wave in the electrocardio sequence data based on the denoised 12-lead sequence data, and performing heart beat segmentation by taking the R wave as a reference to obtain 10-dimensional electrocardio characteristics.
In particular, in the discriminating module,
module M4.1: carrying out data enhancement processing of translation and scaling on the denoised 12-lead sequence data and the 10-dimensional electrocardio characteristics to obtain processed data;
module M4.2: extracting depth features from the processed data by using a ResneXt50+SE depth residual error depth neural network;
module M4.3: and combining the depth characteristic with causal mechanism variable data, and inputting a trained xgBoost tree model to obtain the prediction probability of the occurrence of the abnormal electrocardio characteristic st section in the electrocardiogram.
Specifically, the resnex50+se convolution residual depth neural network employs: the processed data enter a ResneXt50+SE depth residual error depth neural network, enter a bottleneck layer with repeated preset quantity after passing through a 1-dimensional convolution layer and a maximum pooling layer, and finally are converted into one-dimensional data after passing through a global pooling layer and a global maximum pooling layer, and then a three-dimensional predicted value is obtained through a full-connection layer and a Softmax function layer. The Softmax function converts the output result into a probability
Specifically, in the bottleneck layer: changing the data dimension by using the operation of the transformation dimension with the convolution kernel of 1, then increasing the network width by using a grouping convolution mode, and finally obtaining the data with the corresponding size by using the operation of the transformation dimension with the convolution kernel of 1.
Specifically, the bottleneck layer further includes: and the correlation among channels is learned through the SE step, so that the accuracy of the ResneXt50+SE convolution residual error deep neural network is improved. The "squeeze" operation of the SE module functions to convert each channel's data into a value by a pooling operation, containing the channel's comprehensive information. Following the "decompression" operation, two fully connected layers are used to train learn the links between the various channels, resulting in a weight for each characteristic channel that characterizes the importance of the channel and the links between the channels.
The application provides an electrocardiographic ST segment abnormality judging method based on causal analysis, which comprises the following steps:
a Bayesian network generation step: extracting and preprocessing the extracardiac physical sign factor data to obtain preprocessed extracardiac physical sign factor data, generating a weighted adjacency matrix among data variables based on the preprocessed extracardiac physical sign factor data, and extracting non-zero weight values by the weighted adjacency matrix to generate a Bayesian network G 0 Establishing a causal link mechanism between the abnormal electrocardiographic st section and the extracardiac physical sign factor data; wherein the application uses a large amount of clinical electrocardiogram data including original xml file, clinical diagnosis result of a doctor on electrocardiogram and other sign data of an interviewee, and the application uses priori knowledge to extract the clinical electrocardiogram data from the original sign dataTen variables are screened out, after data are prepared, a NOTEARS structure learning algorithm and a K2 scoring algorithm are respectively used for generating a weighted adjacency matrix between the data variables, the adjacency matrix is subjected to secondary matrix weight through a hard threshold omega=0.1, a preliminary causal network model is obtained, and a causal link mechanism between the electrocardiogram ST segment abnormality and the physical sign variable is established.
Directional adjustment: computing Bayesian network G 0 Causal effect estimators of each path, and generating causal network G based on causal effect estimators by adjusting network structure 1 Screening out a preset causal mechanism variable related to the abnormality of the st section of the electrocardiogram according to a preset requirement; wherein, for the Bayesian network G generated in the Bayesian network generation step 0 Performing structure adjustment, and calculating Bayesian network G 0 The causal effect estimation quantity of each path adopts a front gate adjustment formula and a back gate adjustment formula to eliminate the existence condition of each path one by one, and the Bayesian network G is pruned and adjusted 0 And 8 causal mechanism variables related to the ST segment abnormality of the electrocardiogram are selected.
An electrocardiogram signal processing step: extracting 12-lead data from an electrocardiogram, and preprocessing the 12-lead data to obtain preprocessed 12-lead data; acquiring 10-dimensional electrocardio characteristics based on the preprocessed 12-lead data; the application uses a large amount of clinical electrocardiogram data comprising an original xml file and clinical diagnosis results of a doctor on an electrocardiogram, wherein the electrocardiogram data contains various noises including baseline drift, myoelectric noise, power frequency noise and the like. The method comprises the steps of removing various noises on an electrocardiogram by using wavelets, removing data with poor recording and poor quality by combining relevant information and electrocardiographic waveform information diagnosed by doctors, and then performing heart beat segmentation on the data to uniformly fill the data into fixed lengths. According to morphological characteristics of ST segment abnormality, ten types of characteristics are designed and extracted for subsequent training.
Judging: preprocessing the preprocessed 12-lead data and 10-dimensional electrocardio features, and extracting depth features through a convolution residual neural network; combining the depth characteristic with causal mechanism variable data, and inputting a decision tree to obtain the prediction probability of the occurrence of the abnormal electrocardio characteristic st segment in the electrocardiogram. The application relates to a method for processing data of an electrocardiogram signal, which comprises the steps of carrying out data enhancement on the data generated in the step of processing the electrocardiogram signal, carrying out translation and scaling on original electrocardiogram sequence data, inputting the data into a RestneXt50 depth residual error network training comprising an SE module, and in order to adapt to the characteristics of the electrocardiogram data, shifting one-dimensional convolution by all two-dimensional convolution operations in the ResneXt50, and before a full connection layer, the SE module mainly learns the correlation among channels, thereby improving the network performance. Inputting training data, directly extracting 512-dimensional depth features by using an electrocardiogram signal processing step, splicing the depth features with electrocardiographic waveform features and causal mechanism features, training by using an xgboost tree model, adjusting xgboost related super-parameters, and learning to obtain a probability prediction model. When the method is used, only the electrocardiogram xml file is placed at a program appointed position, 512-dimensional characteristics extracted by the neural network are obtained through the neural network, then the characteristics are combined with extracardiac element information such as patient signs and the like, and an xgboost model is input, so that the probability and the abnormal type of the abnormal electrocardio characteristics of the ST segment of the detector can be obtained.
Specifically, in the Bayesian network generation step,
step S1.1: extracting the extracardiac physical sign factor data, and discretizing the extracted extracardiac physical sign factor data to obtain discretized extracardiac physical sign factor data;
step S1.2: generating a weighted adjacency matrix between data variables by using a NOTEARS structure learning algorithm and a K2 scoring algorithm from the discretized extracardiac sign factor data;
step S1.3: the weighted adjacency matrix carries out secondary processing on matrix weights through a hard threshold rule, extracts non-zero weights and generates a Bayesian network G 0 。
In particular, in the directional adjustment step,
step S2.1: calculating a Bayesian network G by adopting a front door adjustment formula and a back door adjustment formula 0 A causal effect estimator for each path;
step S2.2: intervention operation is performed by using do operation based on causal effect estimation, network structure is adjusted, and causal network G is generated 1 。
Specifically, in said step S2.1, different causal relationships between smoking, drinking and movement variables and st segment anomalies are inferred using front door adjustment formulas, rear door adjustment formulas and directional separation formulas.
Specifically, in the electrocardiographic signal processing step,
step S3.1: extracting 12-lead sequence data from an XML file generated by an electrocardiograph;
step S3.2: carrying out 7-layer decomposition denoising treatment on the electrocardiosignal by using symlets4 wavelet on the 12-lead sequence data to obtain denoised 12-lead sequence data;
step S3.3: and searching the position of R wave in the electrocardio sequence data based on the denoised 12-lead sequence data, and performing heart beat segmentation by taking the R wave as a reference to obtain 10-dimensional electrocardio characteristics.
Specifically, in the discriminating step,
step S4.1: carrying out data enhancement processing of translation and scaling on the denoised 12-lead sequence data and the 10-dimensional electrocardio characteristics to obtain processed data;
step S4.2: extracting depth features from the processed data by using a ResneXt50+SE depth residual error depth neural network;
step S4.3: and combining the depth characteristic with causal mechanism variable data, and inputting a trained xgBoost tree model to obtain the prediction probability of the occurrence of the abnormal electrocardio characteristic st section in the electrocardiogram.
Specifically, the resnex50+se convolution residual depth neural network employs: the processed data enter a ResneXt50+SE depth residual error depth neural network, enter a bottleneck layer with repeated preset quantity after passing through a 1-dimensional convolution layer and a maximum pooling layer, and finally are converted into one-dimensional data after passing through a global pooling layer and a global maximum pooling layer, and then a three-dimensional predicted value is obtained through a full-connection layer and a Softmax function layer. The Softmax function converts the output result into a probability
Specifically, in the bottleneck layer: changing the data dimension by using the operation of the transformation dimension with the convolution kernel of 1, then increasing the network width by using a grouping convolution mode, and finally obtaining the data with the corresponding size by using the operation of the transformation dimension with the convolution kernel of 1.
Specifically, the bottleneck layer further includes: and the correlation among channels is learned through the SE step, so that the accuracy of the ResneXt50+SE convolution residual error deep neural network is improved. The "squeeze" operation of the SE module functions to convert each channel's data into a value by a pooling operation, containing the channel's comprehensive information. Following the "decompression" operation, two fully connected layers are used to train learn the links between the various channels, resulting in a weight for each characteristic channel that characterizes the importance of the channel and the links between the channels.
Example 2
Example 2 is a preferred example of example 1
The extracardiac variable data used in the application is a data set based on the investigation of the health condition of the aged population in the Shanghai region, the follow-up database is from the investigation of the health condition of the aged population above 60 years in all mud towns from 28 to 9 to 12 days of 2018, and the total investigation number is 12098. The original dataset contained 55 variables, ten of which were screened from the original data based on a priori knowledge.
The application provides an electrocardiographic ST segment abnormality judging method based on causal analysis, as shown in figure 1, comprising the following steps:
step 1: generating a weighted adjacent matrix between data variables by using a NOTEARS structure learning algorithm and a K2 scoring search structure learning algorithm, performing secondary processing on matrix weights by the weighted adjacent matrix through a hard threshold rule, extracting non-zero weights, and generating a Bayesian network G 0 The method comprises the steps of carrying out a first treatment on the surface of the Non-combined structure learning algorithm based on trace index and augmented Lagrangian method for calculating Bayesian network G 0 A loss function of (2);
adding l on the basis of the loss function 1 Regularization W 1 ＝||vec(W)|| 1 Obtaining a scoring function:
wherein X is E R n×d Is observed vector x= (X) from n independent co-distributed data 1 ，...，X d ) The formed data matrix has n matrix rows and d vector columns; w= [ W ] 1 ]，[w 2 ]...[w d ]By the formulaDefining a linear structural equation model; t is matrix transposition operation; z= (z) 1 ，...，z j ) Is a random noise vector; /> A least squares loss function; λ represents a random variable; vec (W) represents regularization;
establishing a causal relation graph, namely obtaining a partial function minimum value:
subjectto G(W)∈D (3)
wherein G (W) represents the directed acyclic graph DAG is composed of a set of points and a set of edges, D represents a discrete space of the directed acyclic graph generated based on n nodes, containing all possible graph structures; data matrix w= [ W 1 ]，[w 2 ]...[w d ]And d is a linear structural equation model representing d, and d is the number of rows and columns of the matrix.
The discrete constraint G (W) ∈d is converted into a smooth equality constraint h (W) =0.
h(W)＝tr(e W⊙W )-d＝0(4)
Wherein, the ". Aldrich represents Hadamard product, e A Is the matrix index of A; d is the number of rows and columns of the data matrix W; at present, is to be soughtThe problem of solving is that one equality constraint problem ECP (equality-constrained program) is as follows:
subjectto h(W)＝0 (6)
the ECP problem is solved using the augmented Lagrangian method, which generates a quadratic penalty term:
subjectto h(W)＝0 (8)
wherein ρ is a random variable; the dwell point is the minimum value (i.e. ) When the value of W is taken;
calculating to generate a 10×10 weighted adjacency matrix, and obtaining a standing point based on formula 8After that, given a fixed threshold ω > 0, +.>Weights less than the threshold ω are all set directly to 0. This strategy also has an important role in that it rounds up the augmented lagrangian value solution. Since the numerical accuracy is required to be considered, h (W ECP ) =0, but given a small error acceptance range e (e.g. ∈=10 -8 ) Require h (W ECP ) The amount of E is less than or equal to. This is also sufficient to exclude edges that would cause the ring to appear. Since the present application aims to recover the structure of the Bayesian network without concern for non-zero weights in the adjacency matrixSo that the weight in the adjacency matrix is binarized. And reserving zero values, setting all non-zero values to be 1 to obtain a binary matrix, and generating a primary causal network.
The arrow direction of the bayesian network reflects more of a relatively optimal structure derived from the conditional independence and scoring function between data, which is to be modified later by us based on knowledge accumulation in the field, where intervention tools are used to adjust the network model.
An intervention formula:
P(Y＝y|do(X＝x))＝∑ Z P′(Y＝y|X＝x，Z＝z)·P′(Z＝z) (9)
wherein X, Y, Z represents three variables, taking these three variables as examples we explore the relationship between the three variables through the intervention formula; x, y, z represent a particular value taken by variable X, Y, Z.
Given a directed acyclic graph G, and a pair of ordered variables X and Y in G, if none of the nodes in a set of variables Z are descendant nodes of X, and if Z is the condition that would block the back door path between all X and Y, then the variable Z satisfies the back door criterion for (X, Y). The back door adjusting formula is
P(Y＝y|do(X＝x))＝∑ Z P′(Y＝y|X＝x，Z＝z)·P′(Z＝z) (10)
The front door is also adjusted by the formula
P(Z＝z|do(X＝x))＝∑ x ′P(Z＝z|Y＝y，X＝x′)·P(X＝x′)∑ y P(Y＝y|x＝x) (11)
And stripping the network structure between each factor and the ST segment abnormality, checking the directions of arrows in the structure and whether paths exist one by one, and directly deleting the paths in the network.
The present application uses the fact that the electrocardiographic data is an xml format 12-lead electrocardiographic sequence data, which has been given the diagnostic results of electrocardiography by an experienced physician. The Symlet wavelet function is an approximately symmetrical wavelet function, has good regularity and can reduce phase distortion of signals during analysis and reconstruction to a certain extent. And (3) selecting a Symlets4 wavelet function to decompose the electrocardiosignals in 7 scales, regarding a low-frequency signal of wavelet decomposition as baseline drift of the electrocardiosignals, namely regarding a wavelet coefficient of a scale 1 as baseline drift, setting coefficients of the scales 1,6 and 7 as zero, and then obtaining the denoised electrocardiosignals through wavelet reconstruction.
After the denoising electrocardiosignal is obtained, the electrocardiograph is subjected to heart beat segmentation, and because the R wave difference is extremely large, the product of the maximum value of the electrocardiograph multiplied by 0.6 is taken as a threshold value, then the average value of RR intervals is taken as a period T, the point of the distance of 0.35 x T of the R wave front is taken as a starting point, the point of the distance of 0.65 x T of the R wave front is taken as an end point, heart beat is intercepted, and heart beat data is filled into sequence data with the length of 600 in a filling 0 mode.
The data is input into the ResneXt50 modular network for training. The convolution operation in the network is completely changed into one-dimensional convolution, and the discrete one-dimensional convolution calculation formula is as follows:
after entering a network, the data enters a repeated bottleneck layer after passing through a 1-dimensional convolution layer and a maximum pooling layer, and finally is converted into one-dimensional data after passing through a global pooling layer and a global maximum pooling layer, and a final three-dimensional predicted value is obtained after passing through a full-connection layer. In each Bottleneck, a transform dimension operation with a convolution kernel of 1 is performed first, then a grouping convolution is used, and finally a transform dimension operation with a kernel of 1 is used to obtain data with a corresponding size. It differs from Resnet50 in that packet convolution is used, which is equivalent to increasing the "width" of the network without increasing the number of parameters, resulting in a better effect than Resnet 50.
The training tree classifier xgboost is a lifting tree model, and is essentially that a plurality of tree models are integrated together, and a heuristic method is used for solving the problem of constructing a classification tree according to training characteristics and training data in a decision tree model, so as to judge the prediction result of each piece of data. The building tree uses gini index to calculate gain, namely, the feature selection of the building tree is carried out, the gini index formula is shown as formula (1), and the gini index calculation gain formula is shown as formula (2):
pk represents the probability of category K in dataset D, K represents the number of categories;
d denotes the whole dataset, D1 and D2 denote the dataset characterized by a and the dataset characterized by non-a in the dataset, respectively, gini (D1) denotes the Gini index of the dataset characterized by a.
When xgboost is trained, firstly, the full connection layer of the ResneXt50 model trained in the step 2 is removed, then training data is input, 512-dimensional depth characteristics for training a tree model can be obtained, then data of an 8-dimensional causal mechanism and the depth characteristics are transversely spliced to obtain 522-dimensional training data, the data is input into xgboost training, parameters such as a learning rate, a maximum depth, a column number ratio and the like are adjusted, and a final prediction and discrimination model is obtained.
Those skilled in the art will appreciate that the systems, apparatus, and their respective modules provided herein may be implemented entirely by logic programming of method steps such that the systems, apparatus, and their respective modules are implemented as logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers, etc., in addition to the systems, apparatus, and their respective modules being implemented as pure computer readable program code. Therefore, the system, the apparatus, and the respective modules thereof provided by the present application may be regarded as one hardware component, and the modules included therein for implementing various programs may also be regarded as structures within the hardware component; modules for implementing various functions may also be regarded as being either software programs for implementing the methods or structures within hardware components.
The foregoing describes specific embodiments of the present application. It is to be understood that the application is not limited to the particular embodiments described above, and that various changes or modifications may be made by those skilled in the art within the scope of the appended claims without affecting the spirit of the application. The embodiments of the application and the features of the embodiments may be combined with each other arbitrarily without conflict.
Claims (10)
1. An electrocardiogram ST segment abnormality discrimination system based on causal analysis, characterized by comprising:
a bayesian network generation module: extracting and preprocessing the extracardiac physical sign factor data to obtain preprocessed extracardiac physical sign factor data, generating a weighted adjacency matrix among data variables based on the preprocessed extracardiac physical sign factor data, and extracting non-zero weight values by the weighted adjacency matrix to generate a Bayesian network G 0 Establishing a causal link mechanism between the abnormal electrocardiographic st section and the extracardiac physical sign factor data;
and the directional adjustment module is used for: computing Bayesian network G 0 Causal effect estimators of each path, and generating causal network G based on causal effect estimators by adjusting network structure 1 Screening out a preset causal mechanism variable related to the abnormality of the st section of the electrocardiogram according to a preset requirement;
an electrocardiogram signal processing module: extracting 12-lead data from an electrocardiogram, and preprocessing the 12-lead data to obtain preprocessed 12-lead data; acquiring 10-dimensional electrocardio characteristics based on the preprocessed 12-lead data;
and a judging module: preprocessing the preprocessed 12-lead data and 10-dimensional electrocardio features, and extracting depth features through a convolution residual neural network; combining the depth characteristic with causal mechanism variable data, and inputting a decision tree to obtain the prediction probability of the occurrence of the abnormal electrocardio characteristic st segment in the electrocardiogram.
2. The causal analysis-based electrocardiogram ST segment abnormality discrimination system according to claim 1, wherein in the bayesian network generation module,
module M1.1: extracting the extracardiac physical sign factor data, and discretizing the extracted extracardiac physical sign factor data to obtain discretized extracardiac physical sign factor data;
module M1.2: generating a weighted adjacency matrix between data variables by using a NOTEARS structure learning algorithm and a K2 scoring algorithm from the discretized extracardiac sign factor data;
module M1.3: the weighted adjacency matrix carries out secondary processing on matrix weights through a hard threshold rule, extracts non-zero weights and generates a Bayesian network G 0 。
3. The system for discriminating ST-segment abnormalities of an electrocardiogram based on causal analysis according to claim 2 wherein said module M1.3 employs: the weighted adjacent matrix carries out secondary processing on matrix weights through a hard threshold omega=0.1, non-zero weights are extracted, and a Bayesian network G is generated 0 。
4. The system for discriminating an electrocardiogram ST segment abnormality based on a causal analysis according to claim 1 wherein, in said directional adjustment module,
module M2.1: calculating a Bayesian network G by adopting a front door adjustment formula and a back door adjustment formula 0 A causal effect estimator for each path;
module M2.2: intervention operation is performed by using do operation based on causal effect estimation, network structure is adjusted, and causal network G is generated 1 。
5. The causal analysis-based electrocardiogram ST-segment abnormality discrimination system according to claim 4, wherein in said module M2.1, a front gate adjustment formula, a back gate adjustment formula and a directional separation rule are employed to infer different causal relationships between smoking, drinking and motion variables and ST-segment abnormalities.
6. The system for discriminating an ST segment abnormality of an electrocardiogram based on a causal analysis according to claim 1 wherein, in said electrocardiogram signal processing module,
module M3.1: extracting 12-lead sequence data from an XML file generated by an electrocardiograph;
module M3.2: carrying out 7-layer decomposition denoising treatment on the electrocardiosignal by using symlets4 wavelet on the 12-lead sequence data to obtain denoised 12-lead sequence data;
module M3.3: and searching the position of R wave in the electrocardio sequence data based on the denoised 12-lead sequence data, and performing heart beat segmentation by taking the R wave as a reference to obtain 10-dimensional electrocardio characteristics.
7. The system for discriminating an electrocardiographic ST segment abnormality based on a causal analysis according to claim 1 wherein, in said discriminating module,
module M4.1: carrying out data enhancement processing of translation and scaling on the denoised 12-lead sequence data and the 10-dimensional electrocardio characteristics to obtain processed data;
module M4.2: extracting depth features from the processed data by using a ResneXt50+SE depth residual error depth neural network;
module M4.3: and combining the depth characteristic with causal mechanism variable data, and inputting a trained xgBoost tree model to obtain the prediction probability of the occurrence of the abnormal electrocardio characteristic st section in the electrocardiogram.
8. The causal analysis-based electrocardiogram ST segment abnormality discrimination system according to claim 7, wherein the resuxt50+se convolution residual depth neural network employs: the processed data enter a ResneXt50+SE depth residual error depth neural network, enter a bottleneck layer with repeated preset quantity after passing through a 1-dimensional convolution layer and a maximum pooling layer, and finally are converted into one-dimensional data after passing through a global pooling layer and a global maximum pooling layer, and then a three-dimensional predicted value is obtained through a full-connection layer and a Softmax function layer.
9. The causal analysis-based electrocardiogram ST segment abnormality discrimination system according to claim 8, wherein in the bottleneck layer: the data dimension is changed by using the operation of the transformation dimension with the convolution kernel of 1, then the data dimension is changed by using the grouping convolution, and finally the data with the corresponding dimension is obtained by using the operation of the transformation dimension with the convolution kernel of 1.
10. The causal analysis-based electrocardiogram ST segment abnormality discrimination system according to claim 9, wherein the bottleneck layer further comprises: and the correlation among channels is learned through the SE module, so that the accuracy of the ResneXt50+SE convolution residual error deep neural network is improved.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111631456.4A CN114176600B (en) | 2021-12-28 | 2021-12-28 | Electrocardiogram ST segment abnormality discrimination system based on causal analysis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111631456.4A CN114176600B (en) | 2021-12-28 | 2021-12-28 | Electrocardiogram ST segment abnormality discrimination system based on causal analysis |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114176600A CN114176600A (en) | 2022-03-15 |
| CN114176600B true CN114176600B (en) | 2023-10-20 |
Family
ID=80606325
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111631456.4A Active CN114176600B (en) | 2021-12-28 | 2021-12-28 | Electrocardiogram ST segment abnormality discrimination system based on causal analysis |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114176600B (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| RU2279246C2 (en) * | 2004-06-08 | 2006-07-10 | Анатолий Иванович Коряков | Method for predicting ischemic cardiac disease (icd) with the help of holter's electrocardiogram monitoring |
| WO2020141807A2 (en) * | 2019-01-03 | 2020-07-09 | 인하대학교 산학협력단 | Method for predicting paroxysmal atrial fibrillation in normal sinus rhythm electrocardiogram state by using deep learning |
| CN111860686A (en) * | 2020-07-31 | 2020-10-30 | 北京航空航天大学 | Causal relationship adjacency matrix characteristic extraction method for fault diagnosis |
| CN112201340A (en) * | 2020-07-15 | 2021-01-08 | 南京邮电大学 | Electrocardiogram disease determination method based on Bayesian network filtering |
| CN113128552A (en) * | 2021-03-02 | 2021-07-16 | 杭州电子科技大学 | Electroencephalogram emotion recognition method based on depth separable causal graph convolution network |
| WO2021237918A1 (en) * | 2020-05-25 | 2021-12-02 | 五邑大学 | Gpdc graph convolutional neural network-based fatigue detection method, apparatus, and storage medium |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060247544A1 (en) * | 2005-02-03 | 2006-11-02 | Maleeha Qazi | Characterization of cardiac motion with spatial relationship |
| US20190038148A1 (en) * | 2013-12-12 | 2019-02-07 | Alivecor, Inc. | Health with a mobile device |
| NZ730197A (en) * | 2014-09-11 | 2022-07-01 | Berg Llc | Bayesian causal relationship network models for healthcare diagnosis and treatment based on patient data |
| CN108113666B (en) * | 2017-12-19 | 2019-06-21 | 中国科学院深圳先进技术研究院 | Recognition methods, device and the equipment of atrial fibrillation signal |
| US20210375455A1 (en) * | 2020-05-27 | 2021-12-02 | MediSync, Inc. | Artificial intelligence systems that incorporate expert knowledge related to heart failure |
-
2021
- 2021-12-28 CN CN202111631456.4A patent/CN114176600B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| RU2279246C2 (en) * | 2004-06-08 | 2006-07-10 | Анатолий Иванович Коряков | Method for predicting ischemic cardiac disease (icd) with the help of holter's electrocardiogram monitoring |
| WO2020141807A2 (en) * | 2019-01-03 | 2020-07-09 | 인하대학교 산학협력단 | Method for predicting paroxysmal atrial fibrillation in normal sinus rhythm electrocardiogram state by using deep learning |
| WO2021237918A1 (en) * | 2020-05-25 | 2021-12-02 | 五邑大学 | Gpdc graph convolutional neural network-based fatigue detection method, apparatus, and storage medium |
| CN112201340A (en) * | 2020-07-15 | 2021-01-08 | 南京邮电大学 | Electrocardiogram disease determination method based on Bayesian network filtering |
| CN111860686A (en) * | 2020-07-31 | 2020-10-30 | 北京航空航天大学 | Causal relationship adjacency matrix characteristic extraction method for fault diagnosis |
| CN113128552A (en) * | 2021-03-02 | 2021-07-16 | 杭州电子科技大学 | Electroencephalogram emotion recognition method based on depth separable causal graph convolution network |
Non-Patent Citations (2)
| Title |
|---|
| An ECG Classification Method Based on Dilated Causal Convolution;Ma Hao，et al;Computational and Mathematical Methods in Medicine;第2021卷;全文 * |
| Nonliear Granger causality and its application in the analysis of epileptic EEG and ECG signal;Du P，et al;PROCEEDINGS OF THE 2015 6TH INTERNATIONAL CONFERENCE ON MANUFACTURING SCIENCE AND ENGINEERING;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114176600A (en) | 2022-03-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113040715B (en) | Human brain function network classification method based on convolutional neural network | |
| CN108714026B (en) | Fine-grained electrocardiosignal classification method based on deep convolutional neural network and online decision fusion | |
| Liu et al. | Classification of heart diseases based on ECG signals using long short-term memory | |
| CN113724880A (en) | Abnormal brain connection prediction system, method and device and readable storage medium | |
| CN111626114B (en) | Electrocardiosignal arrhythmia classification system based on convolutional neural network | |
| CN110522444B (en) | Electrocardiosignal identification and classification method based on Kernel-CNN | |
| CN108511055B (en) | Ventricular premature beat recognition system and method based on classifier fusion and diagnosis rules | |
| Merjulah et al. | Classification of myocardial ischemia in delayed contrast enhancement using machine learning | |
| Sumathi et al. | A wavelet transform based feature extraction and classification of cardiac disorder | |
| US20230225663A1 (en) | Method for predicting multi-type electrocardiogram heart rhythms based on graph convolution | |
| CN113274031B (en) | Arrhythmia classification method based on depth convolution residual error network | |
| Lv et al. | Heart-darts: Classification of heartbeats using differentiable architecture search | |
| Karthik et al. | Automated Deep Learning Based Cardiovascular Disease Diagnosis Using ECG Signals. | |
| Prakash et al. | A system for automatic cardiac arrhythmia recognition using electrocardiogram signal | |
| CN115984555A (en) | Coronary artery stenosis identification method based on depth self-encoder composition | |
| CN113180688B (en) | Coronary heart disease electrocardiogram screening system and method based on residual error neural network | |
| Mohebbanaaz et al. | A new transfer learning approach to detect cardiac arrhythmia from ECG signals | |
| CN110811591A (en) | Heart failure grading method based on heart rate variability | |
| CN113180685A (en) | Electrocardio abnormity discrimination system and method based on morphological filtering and wavelet threshold | |
| Zaorálek et al. | Patient-adapted and inter-patient ECG classification using neural network and gradient boosting | |
| Liu et al. | Automatic detection of ECG abnormalities by using an ensemble of deep residual networks with attention | |
| CN114176600B (en) | Electrocardiogram ST segment abnormality discrimination system based on causal analysis | |
| CN113768514B (en) | Arrhythmia classification method based on convolutional neural network and gating circulation unit | |
| Singh et al. | Automated atrial fibrillation classification based on denoising stacked autoencoder and optimized deep network | |
| CN114091530A (en) | Electrocardiogram classification method and domain-adaptive-based electrocardiogram classification model training method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |