CN108836302B - Intelligent electrocardiogram analysis method and system based on deep neural network - Google Patents
Intelligent electrocardiogram analysis method and system based on deep neural network Download PDFInfo
- Publication number
- CN108836302B CN108836302B CN201810224387.7A CN201810224387A CN108836302B CN 108836302 B CN108836302 B CN 108836302B CN 201810224387 A CN201810224387 A CN 201810224387A CN 108836302 B CN108836302 B CN 108836302B
- Authority
- CN
- China
- Prior art keywords
- electrocardiogram
- neural network
- transverse
- data
- longitudinal
- 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
- 238000004458 analytical method Methods 0.000 title claims abstract description 46
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 31
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 66
- 230000003068 static effect Effects 0.000 claims abstract description 35
- 238000012549 training Methods 0.000 claims abstract description 28
- 238000000034 method Methods 0.000 claims abstract description 19
- 238000001514 detection method Methods 0.000 claims abstract description 13
- 238000007781 pre-processing Methods 0.000 claims description 16
- 238000002372 labelling Methods 0.000 claims description 15
- 230000002159 abnormal effect Effects 0.000 claims description 14
- 230000004927 fusion Effects 0.000 claims description 10
- 230000000747 cardiac effect Effects 0.000 claims description 7
- 238000013500 data storage Methods 0.000 claims description 6
- 238000005070 sampling Methods 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 5
- 230000000694 effects Effects 0.000 abstract description 9
- 210000002569 neuron Anatomy 0.000 description 11
- 238000003745 diagnosis Methods 0.000 description 10
- 238000000605 extraction Methods 0.000 description 10
- 238000013507 mapping Methods 0.000 description 10
- 230000005856 abnormality Effects 0.000 description 6
- 238000004422 calculation algorithm Methods 0.000 description 5
- 206010003119 arrhythmia Diseases 0.000 description 4
- 230000006793 arrhythmia Effects 0.000 description 4
- 208000004301 Sinus Arrhythmia Diseases 0.000 description 3
- 230000008901 benefit Effects 0.000 description 3
- 230000034994 death Effects 0.000 description 3
- 231100000517 death Toxicity 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 230000008722 morphological abnormality Effects 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 239000013598 vector Substances 0.000 description 3
- 206010003130 Arrhythmia supraventricular Diseases 0.000 description 2
- 208000024172 Cardiovascular disease Diseases 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 2
- 210000004027 cell Anatomy 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000000877 morphologic effect Effects 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 238000000718 qrs complex Methods 0.000 description 2
- 230000001020 rhythmical effect Effects 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 206010003671 Atrioventricular Block Diseases 0.000 description 1
- 238000012935 Averaging Methods 0.000 description 1
- 206010012118 Defect conduction intraventricular Diseases 0.000 description 1
- 208000010496 Heart Arrest Diseases 0.000 description 1
- 206010020880 Hypertrophy Diseases 0.000 description 1
- 206010047281 Ventricular arrhythmia Diseases 0.000 description 1
- 206010049761 Ventricular pre-excitation Diseases 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000001746 atrial effect Effects 0.000 description 1
- 238000013529 biological neural network Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000006073 displacement reaction Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 208000010125 myocardial infarction Diseases 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000001902 propagating effect Effects 0.000 description 1
- 230000001360 synchronised effect Effects 0.000 description 1
- 230000005945 translocation Effects 0.000 description 1
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—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
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
- A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B2576/00—Medical imaging apparatus involving image processing or analysis
- A61B2576/02—Medical imaging apparatus involving image processing or analysis specially adapted for a particular organ or body part
- A61B2576/023—Medical imaging apparatus involving image processing or analysis specially adapted for a particular organ or body part for the heart
Abstract
The invention discloses an electrocardiogram intelligent analysis method and system based on a deep neural network, wherein the method comprises a training stage and a detection stage; in the training stage, marking a transverse layer label and a longitudinal layer label on each acquired image information of the N-lead static electrocardiogram; training to obtain N convolutional neural network models for identifying the transverse layers and N convolutional neural network models for identifying the longitudinal layers; and in the detection stage, N characteristic sequences of the acquired N-lead static electrocardiogram to be detected and characteristic sequences segmented by the heartbeat cycle are used as the input of N convolutional neural network models for identifying the transverse layer and the input of the convolutional neural network models for identifying the transverse layer, so that transverse identification anomaly analysis and longitudinal identification anomaly analysis are obtained. The invention uses the convolution neural network to respectively learn and judge all leads, and experimental results show that the method has better identification effect. The invention has stronger operability, better network generalization capability and higher correct identification rate of the electrocardiogram.
Description
Technical Field
The invention relates to the technical field of medical artificial intelligence, in particular to an electrocardiogram intelligent analysis method and system based on a deep neural network.
Background
According to authoritative surveys, cardiovascular diseases have become one of the leading causes of death in the world population. The number of deaths from cardiovascular disease accounts for one third of the total number of deaths each year.
The electrocardiogram is a common medical examination means for observing the heart electrical activity of a human body, and an electrocardiograph extracts the electrical signals of the heart electrical activity into digital signals and displays the digital signals in the form of the electrocardiogram. With the development of the field of artificial intelligence, particularly the deep learning technology, the technology for analyzing digital electrocardiosignals is more and more abundant and mature. However, the manual analysis of the electrocardiogram not only requires a professional trained physician to perform a large amount of labor, but also has strong subjectivity, so that the automatic analysis of the electrocardiogram is urgently needed. Therefore, the electrocardiogram intelligent diagnosis method and system which are low in cost and have performance reaching medical standards are designed, and have significant social and economic benefits.
At present, the support vector machine algorithm is usually adopted to carry out the intelligent judgment of the electrocardiogram. The electrocardiogram diagnosis of the SVM algorithm is mainly divided into feature extraction and pattern recognition, and the quality of the feature extraction and the performance of a pattern classifier directly influence the performance of the whole recognition system. Because the requirement of the feature engineering on the understanding of the electrocardio by technicians is high, the features generally selected are few, and the effect of the electrocardiogram intelligent diagnosis method based on the support vector machine is not ideal.
Disclosure of Invention
The invention provides an electrocardiogram intelligent analysis method and system based on a deep neural network, aiming at the technical problems of difficult extraction of static electrocardiogram characteristics and low judgment and identification rate in the prior art, wherein a convolutional neural network is adopted to realize accurate judgment of 12 or 18-lead electrocardiogram of a patient, and the electrocardiogram of the patient is subjected to rhythm and morphological identification through 12 or 18-lead overall transverse layer judgment and single cardiac cycle longitudinal layer judgment.
The invention provides an electrocardiogram intelligent analysis method based on a deep neural network, which is characterized by comprising a training stage and a detection stage; the training phase comprises the following steps:
A1) collecting a plurality of pieces of image information of N-lead static electrocardiograms, wherein N is a natural number more than 4;
A2) labeling a transverse layer label and a longitudinal layer label on the image information of each N-lead static electrocardiogram;
A3) preprocessing image information of each N-lead static electrocardiogram, and obtaining N characteristic sequence sample data with transverse layer labels and N characteristic sequence sample data with longitudinal layer labels from each image information;
A4) taking N characteristic sequence sample data with transverse layer labels as input of N convolutional neural networks for identifying transverse layers, and training to obtain N convolutional neural network models for identifying transverse layers;
A5) taking each piece of characteristic sequence sample data with the longitudinal layer label as the input of the convolutional neural network for identifying the longitudinal layer, and training to obtain a convolutional neural network model for identifying the longitudinal layer;
the detection stage comprises the following steps:
B1) collecting image information of an N-lead static electrocardiogram to be detected;
B2) preprocessing the image information of the N-lead static electrocardiogram to be detected to obtain N characteristic sequences and characteristic sequences segmented by a heartbeat cycle;
B3) the N characteristic sequences are used as the input of N convolutional neural network models for identifying the transverse layers, and N outputs with transverse layer labels and probability values are obtained, namely transverse identification anomaly analysis is carried out;
B4) and taking the characteristic sequence segmented by the heartbeat cycle as the input of a convolutional neural network model for identifying the longitudinal layer to obtain a longitudinal layer label and a probability value, namely longitudinal identification anomaly analysis.
Preferably, the source of the N-lead static electrocardiogram signal data may be PDF, CSV, ecg, text, memory data stream, or XML, and the like, the horizontal layer label and the vertical layer label are text information, or other source label data in PDF format, and N is 12 or 18.
Preferably, the preprocessing in step A3) is to extract image information, text information of the horizontal and vertical layer labels, and convert the extracted text information into a CSV file or other electrocardiogram data storage format file with the horizontal and vertical layer labels, that is, unify the data acquired in step a2) into a certain specified format.
Preferably, the feature sequences segmented by the heartbeat cycles in the step B4) are obtained according to the N outputs with the transverse level labels and the probability values, the heartbeat cycles with the transverse level labels in the N leads are removed, and the feature sequences of the complete heartbeat cycle without any transverse level label are selected and input into the convolutional neural network model for identifying the longitudinal level.
Preferably, the transverse bedding label is the text information of transverse abnormal image characteristics labeled on each lead individual electrocardiogram data, and the transverse abnormal image characteristics reflect the arrhythmia type.
Preferably, the longitudinal bedding label is the text information of the longitudinal abnormal image characteristics labeled on the electrocardiogram data of the N-leads at the same time point corresponding to each heartbeat, and the longitudinal abnormal image characteristics reflect the type of the morphological abnormality.
Preferably, the image information of the N-lead static electrocardiogram in the step A1) is not less than 10000 parts.
Based on the intelligent electrocardiogram analysis method based on the deep neural network, the invention also provides a system of the intelligent electrocardiogram analysis method based on the deep neural network, which is characterized by comprising an electrocardiogram data sampling module, a data labeling module, a data preprocessing module, a transverse layer judgment module, a longitudinal layer judgment module and an analysis fusion module;
the electrocardio data sampling module comprises: the device is used for acquiring image information of the N-lead static electrocardiogram;
the data labeling module: the system is used for labeling the image information of the N-lead static electrocardiogram with a transverse layer label and a longitudinal layer label;
the data preprocessing module: extracting image information for the N-lead static electrocardiogram into electrocardiogram data, wherein the electrocardiogram data is a CSV file, a memory data stream or a file with other electrocardiogram data storage formats;
the transverse bedding surface judging module: the system comprises a convolutional neural network model, a transverse layer label and a probability value, wherein the convolutional neural network model is used for inputting preprocessed N-lead electrocardiogram data into corresponding N convolutional neural network models for identifying transverse layers, and respectively judging to obtain N outputs with the transverse layer labels and the probability value; each convolutional neural network model identifying a transverse layer corresponds to the electrocardiographic data of one lead.
The longitudinal bedding surface judging module: the system is used for dividing the preprocessed N-lead electrocardio data into electric signal data according to the cardiac cycle, inputting the electric signal data into a convolutional neural network model for identifying a longitudinal layer, judging, and outputting a label with the longitudinal layer and a probability value; each convolution neural network model for identifying the longitudinal layer corresponds to the electrocardio data of one lead.
The analysis fusion module: and the fusion module is used for fusing the N outputs with the transverse layer labels and the probability values and the outputs with the longitudinal layer labels and the probability values to obtain an electrocardiogram analysis report.
In machine learning, a Convolutional Neural Network (CNN) is a feed-forward Neural Network whose artificial neurons can respond to a portion of the coverage of surrounding cells. Convolutional neural networks are an efficient identification method that has been developed in recent years and has attracted extensive attention. In general, the basic structure of CNN includes two stages, one of which is a feature extraction layer, and the input of each neuron is connected to a local acceptance domain of the previous layer and extracts the feature of the local. Once the local feature is extracted, the position relation between the local feature and other features is determined; the other is a feature mapping layer, each calculation layer of the network is composed of a plurality of feature mappings, each feature mapping is a plane, and the weights of all neurons on the plane are equal. Because the feature detection layer of CNN learns from training data, when using CNN, explicit feature extraction is avoided, but learning is done implicitly from training data; moreover, because the neuron weights on the same feature mapping surface are the same, the network can learn in parallel.
Compared with the traditional judgment method, the electrocardiogram intelligent analysis method and system based on the deep neural network adopt the convolutional neural network aiming at the intelligent diagnosis method of the 12-channel or 18-channel static electrocardiogram, realize the accurate judgment of the 12-channel or 18-channel electrocardiogram of the patient, and carry out the rhythm and morphological identification on the electrocardiogram of the patient through the judgment of the whole 12 or 18 channels and the single cardiac cycle. The invention uses the convolution neural network to respectively learn and judge all leads, thereby avoiding the difficulty of characteristic engineering of the traditional machine learning method. Experimental results show that the method has a better identification effect. The invention has stronger operability, better generalization capability of the network and higher accuracy of electrocardiogram recognition.
Drawings
FIG. 1 is a block diagram of the system of the intelligent electrocardiogram analysis method based on the deep neural network.
FIG. 2 is a flow chart of the intelligent electrocardiogram analysis method based on the deep neural network.
FIG. 3 is a network structure of deep learning of the intelligent electrocardiogram analysis method based on the deep neural network.
FIG. 4 is a specific classification of the intelligent analysis method of electrocardiogram based on deep neural network of the present invention.
Detailed Description
The invention is described in further detail below with reference to the following figures and examples, which should not be construed as limiting the invention.
As shown in fig. 1, the system of the intelligent electrocardiogram analysis method based on the deep neural network provided by the invention comprises an electrocardiogram data sampling module, a data labeling module, a data preprocessing module, a transverse layer judgment module, a longitudinal layer judgment module and an analysis fusion module.
An electrocardiogram data sampling module: used for acquiring the image information of the N-lead static electrocardiogram.
A data labeling module: the method is used for labeling the image information of the N-lead static electrocardiogram with a transverse layer label and a longitudinal layer label. The data annotation module is realized by an electrocardiogram annotation tool. A marking person obtains a task by using an electrocardiogram marking tool, unmarked electrocardiogram data are directly transmitted to a client side from a background, when the marking person marks the whole picture, the marking person needs to select corresponding classification according to abnormal information of the electrocardiogram picture, wherein the classification belongs to transverse layer abnormity or longitudinal layer abnormity, and then selects corresponding secondary classification, and if no abnormal mark appears, the electrocardiogram is considered to be normal. When labeling work is carried out on each cycle, the system can identify each cardiac cycle, labeling work is carried out on each cardiac cycle one by labeling personnel, and data can be automatically synchronized to a server side after the current electrocardiogram finishes the labeling work.
A data preprocessing module: extracting image information for the N-lead static electrocardiogram into electrocardiogram data, wherein the electrocardiogram data is a CSV file, a memory data stream or a file with other electrocardiogram data storage formats;
the transverse bedding surface judging module: the system comprises a convolutional neural network model, a transverse layer label and a probability value, wherein the convolutional neural network model is used for inputting preprocessed N-lead electrocardiogram data into corresponding N convolutional neural network models for identifying transverse layers, and respectively judging to obtain N outputs with the transverse layer labels and the probability value; each convolutional neural network model identifying a transverse layer corresponds to the electrocardiographic data of one lead.
The longitudinal layer judging module: the system is used for dividing the preprocessed N-lead electrocardio data into electric signal data according to a cardiac cycle, inputting the electric signal data into a convolutional neural network model for identifying a longitudinal layer, judging, and outputting a label with the longitudinal layer and a probability value; each convolution neural network model for identifying the longitudinal layer corresponds to the electrocardio data of one lead.
An analysis fusion module: and the fusion module is used for fusing the N outputs with the transverse layer labels and the probability values and the outputs with the longitudinal layer labels and the probability values to obtain an electrocardiogram analysis report.
The electrocardiogram machine is directly connected with an electrocardiogram diagnosis system, the system of the electrocardiogram intelligent analysis method based on the deep neural network is arranged in the electrocardiogram diagnosis system, a doctor uploads electrocardiogram data of all patients, the electrocardiogram diagnosis system comprehensively analyzes the processed electrocardiogram data according to the name, age, sex and other data of the patients, carries out numbering and archiving on the processed electrocardiogram data, the system of the electrocardiogram intelligent analysis method based on the deep neural network carries out detection through a model and obtains two categories of results, wherein the four categories are classified through the transverse and longitudinal levels of the electrocardiogram, and are specifically shown in figure 4, and are divided into two categories of transverse identification and longitudinal identification by judging whether the electrocardiogram is a pacemaker electrocardiogram. Whether the transverse identification abnormality is judged to be sinus arrhythmia or atrial arrhythmia is specifically divided into five categories of arrhythmia, including sinus arrhythmia and sinus arrhythmia, atrial arrhythmia, borderline arrhythmia, ventricular arrhythmia and cardiac arrest. For the judgment of longitudinal abnormality, the QRS complex, P wave, T wave, J wave and U wave are judged, and the method is divided into twelve types of morphological abnormality such as P wave abnormality, QRS complex abnormality, translocation/electric axis, ST-T abnormality, J wave, T wave change, U wave change, QT interval, atrioventricular hypertrophy, myocardial infarction and other special phenomena. The abnormalities caused by conduction are classified into five categories, namely sinus block, atrial conduction, atrioventricular block, intraventricular block, ventricular pre-excitation and the like. The pacemaker electrocardiogram is divided into two categories of pacing mode and pacing perception. The electrocardiogram diagnosis system can also be used for a household electrocardiogram detector to obtain local data, a large amount of normal electrocardiogram data is filtered through preprocessing, the processed data is uploaded to a server, and classified detection is carried out by using a system of an electrocardiogram intelligent analysis method based on a deep neural network.
The intelligent electrocardiogram analysis method based on the deep neural network can be realized based on the system and can also be realized by other systems, as shown in fig. 2, taking 12 leads as an example, the processing methods of 6 leads and 18 leads are the same, and the method comprises a training phase and a detection phase.
The training phase comprises the following steps:
A1) acquiring nearly 100,000 image information of 12-lead static electrocardiograms;
A2) the image information of each 12-lead static electrocardiogram is labeled with a transverse layer label and a longitudinal layer label. The transverse layer is data of each lead single electrocardiogram, the transverse layer label is character information marked according to transverse abnormal image characteristics displayed by the electrocardiogram data, and the transverse abnormal image characteristics reflect arrhythmia types. The 12 leads correspond to 12 sets of transverse slice labels.
Each group of longitudinal layers is an electrocardiogram form of twelve leads at the same time point corresponding to each heart beat (namely a heart beat cycle), the longitudinal layer label is character information labeled according to longitudinal abnormal image characteristics displayed by electrocardiogram data, and the longitudinal abnormal image characteristics reflect the type of morphological abnormality. The 12 leads correspond to a set of longitudinal slice labels at one time point.
The image information of the 12-lead static electrocardiogram is PDF format image, and the transverse layer label and the longitudinal layer label are character information of PDF format.
A3) Preprocessing image information of each 12-lead static electrocardiogram, and obtaining 12 characteristic sequence sample data with a transverse layer label and 12 characteristic sequence sample data with a longitudinal layer label from each image information. The preprocessing process is to extract the image information, the character information of the transverse layer label and the longitudinal layer label and convert the image information, the character information of the transverse layer label and the character information of the longitudinal layer label into a CSV file with the transverse layer label and the longitudinal layer label. In this embodiment, a CSV format file is taken as an example, and the effect of the present invention can be achieved by a memory data stream or a file in another electrocardiographic data storage format.
A4) Taking 12 feature sequence sample data with transverse layer labels as the input of 12 convolutional neural networks for identifying transverse layers, and training to obtain 12 convolutional neural network models for identifying transverse layers;
A5) taking each piece of characteristic sequence sample data with the longitudinal layer label as the input of the convolutional neural network for identifying the longitudinal layer, and training to obtain a convolutional neural network model for identifying the longitudinal layer;
the detection stage comprises the following steps:
B1) acquiring image information of a 12-lead static electrocardiogram to be detected;
B2) preprocessing image information of a 12-lead static electrocardiogram to be detected to obtain N characteristic sequences and characteristic sequences segmented by a heartbeat cycle;
B3) taking the N characteristic sequences as the input of 12 convolutional neural network models for identifying the transverse layers to obtain 12 outputs with transverse layer labels and probability values, namely transverse identification abnormal analysis of the electrocardiogram;
B4) and taking the characteristic sequence segmented by the heartbeat cycle as the input of a convolutional neural network model for identifying the longitudinal layer to obtain a longitudinal layer label and a probability value, namely longitudinal identification abnormal analysis of the electrocardiogram.
And obtaining the characteristic sequence segmented by the heartbeat cycle according to 12 outputs with transverse layer labels and probability values, removing the heartbeat cycle with the transverse layer labels in 12 leads, and selecting the characteristic sequence of the complete heartbeat cycle without any transverse layer label to input the convolutional neural network model for identifying the longitudinal layer.
The learning algorithm based on the convolutional neural network is used in the multi-convolutional neural network fusion algorithm, and the network structure is shown in fig. 3. A Convolutional Neural Network (CNN) is a feed-forward Neural Network whose artificial neurons can respond to a portion of the coverage of surrounding cells, and performs well for large data processing. In general, the basic structure of a CNN comprises two stages, each stage comprising several layers. One is a feature extraction stage, in which the input of each neuron is connected with a local acceptance domain of the previous layer, and the local features are extracted. Once the local feature is extracted, the position relation between the local feature and other features is determined; the second is a feature mapping stage, each computation layer of the network is composed of a plurality of feature maps, each feature map is a plane, and the weights of all neurons on the plane are equal. The feature mapping structure adopts a sigmoid function with small influence function kernel as an activation function of the convolution network, so that the feature mapping has displacement invariance. In addition, since the neurons on one mapping surface share the weight, the number of free parameters of the network is reduced. Each convolutional layer in the convolutional neural network is followed by a computation layer for local averaging and quadratic extraction, which reduces the feature resolution.
Because the feature detection stage of CNN learns from the training data, when using CNN, explicit feature extraction is avoided, but learning is done implicitly from the training data; moreover, because the neuron weights on the same feature mapping surface are the same, the network can learn in parallel, which is also a great advantage of the convolution network compared with the full-connection network. The layout of the convolutional neural network is closer to that of an actual biological neural network, the complexity of the network is reduced by weight sharing, and particularly, the complexity of data reconstruction in the process of feature extraction and classification is reduced by the characteristic that multidimensional input vectors can be directly input into the network.
The learning steps of the convolutional neural network are as follows:
step1 input training sample set.
Step2, constructing a network and initializing, and randomly generating an input layer-to-hidden layer weight matrix W, a hidden layer threshold vector b and convolutional layer parameters.
Step3, calculating the input sample data according to the existing network to obtain a vector about the label information.
Step4, propagating from the classifier to the previous feature extractor.
Step5, updating the weight according to the weight modification strategy of the neural network.
Step6 repeat Step3 through Step5 for the number of iterations.
And respectively inputting sample data of the characteristic sequence of the 12-lead electrical signal data to train the convolutional neural network and calculate the parameters of all convolutional layers. Inputting training sample set to corresponding neural network for training according to the number of lead, finally calculating parameters of each convolution layer and pooling layer in 12 convolutional neural network models for identifying transverse layers and 1 convolutional neural network model for identifying longitudinal layers.
The training process of training the convolutional neural network is a process of solving all parameters in the network, and each parameter is solved, namely the training is completed.
And respectively inputting the data of the 12-lead static electrocardiogram characteristic sequence into corresponding convolutional neural networks for identification. The output value is compared with the label data obtained in advance, and whether the output value is correct or not can be judged.
Test and result analysis
The data selected by the invention is about 100000 parts of electrocardiogram data acquired from about 35000 patients, and the electrocardiogram experts make diagnosis for each period and print labels. The test set was 12000 collected data from 4000 patients who were independent of each other, and this portion of the electrocardiographic data was not involved in model training. The flow chart of the intelligent judgment of the electrocardiogram is shown in fig. 2.
Table 1 independent test statistics
The identification effect of the electrocardiogram intelligent analysis method based on the deep neural network is given in table 1, and we can see that the identification effect of the convolutional neural network algorithm is very excellent.
From the above experimental results, the following conclusions can be drawn:
the electrocardiogram intelligent diagnosis method based on the convolutional neural network is excellent in the recognition of 12-lead static electrocardiograms. This is because the feature detection layer of CNN learns from training data, so when using CNN, it avoids the feature extraction displayed, but implicitly learns from training data; moreover, because the weights of the neurons on the same feature mapping surface are the same, the network can learn in parallel, which is also a great advantage of the convolutional network relative to the network in which the neurons are connected with each other. Experimental results show that the method has a better identification effect. The invention has stronger operability, better generalization capability of the network and higher correct identification rate of the electrocardiogram.
Although the preferred embodiments of the present invention have been described above with reference to the accompanying drawings, the present invention is not limited to the above-described embodiments, which are merely illustrative and not restrictive, and those skilled in the art can make various changes and modifications within the spirit and scope of the present invention without departing from the spirit and scope of the appended claims.
Claims (6)
1. An electrocardiogram intelligent analysis method based on a deep neural network is characterized in that: the method comprises a training phase and a detection phase; the training phase comprises the following steps:
A1) collecting a plurality of pieces of image information of N-lead static electrocardiograms, wherein N is a natural number more than 4;
A2) labeling a transverse layer label and a longitudinal layer label on the image information of each N-lead static electrocardiogram; the transverse layer label is the character information of transverse abnormal image characteristics marked on each lead single electrocardiogram data; the longitudinal layer label is the character information marked according to the longitudinal abnormal image characteristics displayed by the electrocardiogram data; each group of longitudinal layers is an N-lead electrocardiogram form of the same time point corresponding to each heart beat;
A3) preprocessing image information of each N-lead static electrocardiogram, and obtaining N characteristic sequence sample data with transverse layer labels and N characteristic sequence sample data with longitudinal layer labels from each image information;
A4) taking N characteristic sequence sample data with transverse layer labels as input of N convolutional neural networks for identifying transverse layers, and training to obtain N convolutional neural network models for identifying transverse layers;
A5) taking each piece of characteristic sequence sample data with the longitudinal layer label as the input of the convolutional neural network for identifying the longitudinal layer, and training to obtain a convolutional neural network model for identifying the longitudinal layer;
the detection stage comprises the following steps:
B1) collecting image information of an N-lead static electrocardiogram to be detected;
B2) preprocessing the image information of the N-lead static electrocardiogram to be detected to obtain N characteristic sequences and characteristic sequences segmented by a heartbeat cycle;
B3) the N characteristic sequences are used as the input of N convolutional neural network models for identifying the transverse layers, and N outputs with transverse layer labels and probability values are obtained, namely transverse identification anomaly analysis is carried out;
B4) and taking the characteristic sequence segmented by the heartbeat cycle as the input of a convolutional neural network model for identifying the longitudinal layer to obtain a longitudinal layer label and a probability value, namely longitudinal identification anomaly analysis.
2. The intelligent analysis method for electrocardiogram based on deep neural network as claimed in claim 1, wherein: the image information of the N-lead static electrocardiogram is PDF, CSV, ecg, text, memory data stream or electrocardiogram data in an XML format, the transverse layer label and the longitudinal layer label are character information in a PDF format, text information or label data of other sources, and N is 12 or 18.
3. The intelligent analysis method for electrocardiogram based on deep neural network as claimed in claim 1, wherein: the preprocessing process in the step A3) is to extract image information, text information of the transverse layer label and the longitudinal layer label, and convert the extracted text information into a CSV file with the transverse layer label and the longitudinal layer label or a file with other electrocardio data storage formats.
4. The intelligent analysis method for electrocardiogram based on deep neural network as claimed in claim 1, wherein: and B4), obtaining the characteristic sequence segmented by the heartbeat cycle in the step B) according to the N outputs with the transverse layer labels and the probability values, removing the heartbeat cycle with the transverse layer labels in the N leads, and selecting the characteristic sequence of the complete heartbeat cycle without any transverse layer label to input the convolutional neural network model for identifying the longitudinal layer.
5. The intelligent analysis method for electrocardiogram based on deep neural network as claimed in claim 1, wherein: the image information of the N-lead static electrocardiogram in the step A1) is not less than 10000 parts.
6. The system of the deep neural network-based electrocardiogram intelligent analysis method according to any one of claims 1-5, wherein: the electrocardio-data acquisition and analysis system comprises an electrocardio-data sampling module, a data labeling module, a data preprocessing module, a transverse bedding surface judging module, a longitudinal bedding surface judging module and an analysis and fusion module;
the electrocardio data sampling module comprises: the device is used for acquiring image information of the N-lead static electrocardiogram;
the data labeling module: the system is used for labeling the image information of the N-lead static electrocardiogram with a transverse layer label and a longitudinal layer label;
the data preprocessing module: extracting image information for the N-lead static electrocardiogram into electrocardiogram data, wherein the electrocardiogram data is a CSV file, a memory data stream or a file with other electrocardiogram data storage formats;
the transverse bedding surface judging module: the system comprises a convolutional neural network model, a transverse layer label and a probability value, wherein the convolutional neural network model is used for inputting preprocessed N-lead electrocardiogram data into corresponding N convolutional neural network models for identifying transverse layers, and respectively judging to obtain N outputs with the transverse layer labels and the probability value;
the longitudinal bedding surface judging module: the system is used for dividing the preprocessed N-lead electrocardio data into electric signal data according to the cardiac cycle, inputting the electric signal data into a convolutional neural network model for identifying a longitudinal layer, judging, and outputting a label with the longitudinal layer and a probability value;
the analysis fusion module: and the fusion module is used for fusing the N outputs with the transverse layer labels and the probability values and the outputs with the longitudinal layer labels and the probability values to obtain an electrocardiogram analysis report.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810224387.7A CN108836302B (en) | 2018-03-19 | 2018-03-19 | Intelligent electrocardiogram analysis method and system based on deep neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810224387.7A CN108836302B (en) | 2018-03-19 | 2018-03-19 | Intelligent electrocardiogram analysis method and system based on deep neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108836302A CN108836302A (en) | 2018-11-20 |
| CN108836302B true CN108836302B (en) | 2021-06-04 |
Family
ID=64211843
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810224387.7A Active CN108836302B (en) | 2018-03-19 | 2018-03-19 | Intelligent electrocardiogram analysis method and system based on deep neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108836302B (en) |
Families Citing this family (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111261289A (en) * | 2018-11-30 | 2020-06-09 | 上海图灵医疗科技有限公司 | Heart disease detection method based on artificial intelligence model |
| CN109567789B (en) * | 2018-12-03 | 2021-08-31 | 东软集团股份有限公司 | Electrocardiogram data segmentation processing method and device and readable storage medium |
| CN109480824B (en) * | 2018-12-11 | 2021-10-08 | 武汉中旗生物医疗电子有限公司 | Method and device for processing electrocardio waveform data and server |
| CN110327034B (en) * | 2019-04-15 | 2022-07-15 | 浙江工业大学 | Tachycardia electrocardiogram screening method based on depth feature fusion network |
| CN109907753B (en) * | 2019-04-23 | 2022-07-26 | 杭州电子科技大学 | Multi-dimensional ECG signal intelligent diagnosis system |
| CN110236520B (en) * | 2019-05-20 | 2020-09-29 | 上海数创医疗科技有限公司 | Electrocardiogram type recognition device based on double convolution neural network |
| WO2021031155A1 (en) * | 2019-08-21 | 2021-02-25 | 中国科学院深圳先进技术研究院 | Method and device for multi-scale characteristic extraction based on ecg |
| CN110495872B (en) * | 2019-08-27 | 2022-03-15 | 中科麦迪人工智能研究院(苏州)有限公司 | Electrocardiogram analysis method, device, equipment and medium based on picture and heartbeat information |
| CN110704619B (en) * | 2019-09-24 | 2022-06-10 | 支付宝(杭州)信息技术有限公司 | Text classification method and device and electronic equipment |
| CN110731773B (en) * | 2019-10-28 | 2022-10-28 | 浙江工业大学 | Abnormal electrocardiogram screening method based on fusion of global and local depth features of electrocardiogram |
| CN111657915A (en) * | 2020-04-30 | 2020-09-15 | 上海数创医疗科技有限公司 | Electrocardiogram form recognition model based on deep learning and use method thereof |
| CN111657914A (en) * | 2020-04-30 | 2020-09-15 | 上海数创医疗科技有限公司 | Electrocardiogram form recognition method and device based on deep learning |
| CN111657935B (en) * | 2020-05-11 | 2021-10-01 | 浙江大学 | Epilepsia electroencephalogram recognition system based on hierarchical graph convolutional neural network, terminal and storage medium |
| US20220079491A1 (en) * | 2020-09-14 | 2022-03-17 | Biosense Webster (Israel) Ltd. | Local activation time analysis system |
| CN113116300A (en) * | 2021-03-12 | 2021-07-16 | 复旦大学 | Physiological signal classification method based on model fusion |
| CN113052229B (en) * | 2021-03-22 | 2023-08-29 | 武汉中旗生物医疗电子有限公司 | Heart condition classification method and device based on electrocardiographic data |
| CN116864140A (en) * | 2023-09-05 | 2023-10-10 | 天津市胸科医院 | Intracardiac branch of academic or vocational study postoperative care monitoring data processing method and system thereof |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101785670A (en) * | 2009-01-22 | 2010-07-28 | 陈跃军 | Intelligent blurry electrocardiogram on-line analyzer system |
| CN102697491A (en) * | 2012-06-26 | 2012-10-03 | 海信集团有限公司 | Identification method and system of characteristic waveform of electrocardiogram |
| CN105748063A (en) * | 2016-04-25 | 2016-07-13 | 山东大学齐鲁医院 | Intelligent arrhythmia diagnosis method based on multiple-lead and convolutional neural network |
| CN106214145A (en) * | 2016-07-20 | 2016-12-14 | 杨平 | A kind of electrocardiogram classification method based on degree of depth learning algorithm |
Family Cites Families (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030009078A1 (en) * | 1999-10-29 | 2003-01-09 | Elena A. Fedorovskaya | Management of physiological and psychological state of an individual using images congnitive analyzer |
| JP2016073373A (en) * | 2014-10-03 | 2016-05-12 | 日本光電工業株式会社 | Electrocardiogram display device |
| US10194821B2 (en) * | 2014-10-29 | 2019-02-05 | Khalifa University of Science and Technology | Medical device having automated ECG feature extraction |
| CN104799847B (en) * | 2015-05-13 | 2018-05-18 | 北京五维康科技有限公司 | A kind of 12 lead ECG detection devices and remote ecg monitoring diagnosis system |
| US20160374578A1 (en) * | 2015-06-25 | 2016-12-29 | Intel Corporation | Contextual heart health monitoring with integrated ecg (electrocardiogram) |
| CN105125208B (en) * | 2015-10-16 | 2018-04-27 | 东软熙康健康科技有限公司 | A kind of electrocardio drawing drawing method and electrocardiogram drawing tool |
| CN106073753A (en) * | 2016-04-29 | 2016-11-09 | 北京蓬阳丰业医疗设备有限公司 | 18 lead holographic dynamic and static ecg analysis method and system |
-
2018
- 2018-03-19 CN CN201810224387.7A patent/CN108836302B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101785670A (en) * | 2009-01-22 | 2010-07-28 | 陈跃军 | Intelligent blurry electrocardiogram on-line analyzer system |
| CN102697491A (en) * | 2012-06-26 | 2012-10-03 | 海信集团有限公司 | Identification method and system of characteristic waveform of electrocardiogram |
| CN105748063A (en) * | 2016-04-25 | 2016-07-13 | 山东大学齐鲁医院 | Intelligent arrhythmia diagnosis method based on multiple-lead and convolutional neural network |
| CN106214145A (en) * | 2016-07-20 | 2016-12-14 | 杨平 | A kind of electrocardiogram classification method based on degree of depth learning algorithm |
Non-Patent Citations (1)
| Title |
|---|
| "基于卷积神经网络的 ECG 信号个性化分类方法";王海龙等;《2017中国自动化大会》;20171022;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108836302A (en) | 2018-11-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108836302B (en) | Intelligent electrocardiogram analysis method and system based on deep neural network | |
| US11344243B2 (en) | Artificial intelligence self-learning-based static electrocardiography analysis method and apparatus | |
| Deng et al. | Extracting cardiac dynamics within ECG signal for human identification and cardiovascular diseases classification | |
| US20230143594A1 (en) | Systems and methods for reduced lead electrocardiogram diagnosis using deep neural networks and rule-based systems | |
| CN106725426A (en) | A kind of method and system of electrocardiosignal classification | |
| CN109303559B (en) | Dynamic electrocardiogram and heartbeat classification method based on gradient boosting decision tree | |
| CN109480825B (en) | Electrocardio data processing method and device | |
| CN107440679B (en) | Method for detecting abnormalities of an electrocardiogram and corresponding system | |
| CN109222963A (en) | A kind of anomalous ecg method for identifying and classifying based on convolutional neural networks | |
| CN108577831B (en) | Single-guide-core-paste data long-range monitoring and diagnosing system and processing method thereof | |
| CN112932433B (en) | Arrhythmia automatic classification method of electrocardiogram | |
| CN112906748A (en) | 12-lead ECG arrhythmia detection classification model construction method based on residual error network | |
| WO2021071646A1 (en) | Systems and methods for electrocardiogram diagnosis using deep neural networks and rule-based systems | |
| CN111557660A (en) | Arrhythmia identification method under sub-population deep learning framework | |
| Hadiyoso et al. | Image Based ECG Signal Classification Using Convolutional Neural Network. | |
| CN113080996B (en) | Electrocardiogram analysis method and device based on target detection | |
| JP2023104885A (en) | Electrocardiographic heart rate multi-type prediction method based on graph convolution | |
| CN111329467A (en) | Heart disease auxiliary detection method based on artificial intelligence | |
| Alagarsamy et al. | Performing the classification of pulsation cardiac beats automatically by using CNN with various dimensions of kernels | |
| Imtiaz et al. | Cross-Database and Cross-Channel ECG Arrhythmia Heartbeat Classification Based on Unsupervised Domain Adaptation | |
| CN111513702A (en) | Method for detecting electrocardiographic data and storage medium | |
| CN114595731B (en) | Semantic segmentation method of nonlinear medical sensor data based on continuous learning | |
| CN117421678B (en) | Single-lead atrial fibrillation recognition system based on knowledge distillation | |
| CN114098757B (en) | ECG signal monitoring method based on quantum particle swarm optimization | |
| Chao et al. | Research on arrhythmia classification based on domain adaptation |
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 | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20231019 Address after: Room A1006, A1007, A1008, Xinjin Industrial Park, No. 25 Jiangxing Road, Economic Development Zone, Jianghan District, Wuhan City, Hubei Province, 430000 Patentee after: Wenhao (Wuhan) Technology Co.,Ltd. Address before: 430000 room 4, 14 / F, building 8, oceanwide international SOHO city (phase I), Wuhan CBD, Jianghan District, Wuhan City, Hubei Province Patentee before: WUHAN HAIXINGTONG TECHNOLOGY CO.,LTD. |
|
| TR01 | Transfer of patent right |