WO2019100562A1 - 基于人工智能的心电图心搏自动识别分类方法 - Google Patents

基于人工智能的心电图心搏自动识别分类方法 Download PDF

Info

Publication number
WO2019100562A1
WO2019100562A1 PCT/CN2018/072350 CN2018072350W WO2019100562A1 WO 2019100562 A1 WO2019100562 A1 WO 2019100562A1 CN 2018072350 W CN2018072350 W CN 2018072350W WO 2019100562 A1 WO2019100562 A1 WO 2019100562A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
heartbeat
lead
analysis
classification
Prior art date
Application number
PCT/CN2018/072350
Other languages
English (en)
French (fr)
Inventor
胡传言
张雪
田亮
刘涛
曹君
刘畅
Original Assignee
乐普（北京）医疗器械股份有限公司
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 乐普（北京）医疗器械股份有限公司 filed Critical 乐普（北京）医疗器械股份有限公司
Priority to EP18880577.4A priority Critical patent/EP3692901A4/en
Priority to US16/755,105 priority patent/US11564612B2/en
Priority to JP2020518512A priority patent/JP7018133B2/ja
Publication of WO2019100562A1 publication Critical patent/WO2019100562A1/zh

Links

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]
    • A61B5/346—Analysis of electrocardiograms
    • A61B5/349—Detecting specific parameters of the electrocardiograph cycle
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • G06F18/24133—Distances to prototypes
    • G06F18/24143—Distances to neighbourhood prototypes, e.g. restricted Coulomb energy networks [RCEN]
    • 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
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
    • G06V10/449—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters
    • G06V10/451—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters with interaction between the filter responses, e.g. cortical complex cells
    • G06V10/454—Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN]
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H40/00—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices
    • G16H40/60—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the operation of medical equipment or devices
    • G16H40/63—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the operation of medical equipment or devices for local operation
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
    • 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
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/02—Detecting, measuring or recording pulse, heart rate, blood pressure or blood flow; Combined pulse/heart-rate/blood pressure determination; Evaluating a cardiovascular condition not otherwise provided for, e.g. using combinations of techniques provided for in this group with electrocardiography or electroauscultation; Heart catheters for measuring blood pressure
    • A61B5/024—Detecting, measuring or recording pulse rate or heart rate
    • A61B5/0245—Detecting, measuring or recording pulse rate or heart rate by using sensing means generating electric signals, i.e. ECG signals
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/15—Biometric patterns based on physiological signals, e.g. heartbeat, blood flow

Definitions

  • the invention relates to the technical field of artificial intelligence assisted data analysis and processing, in particular to an artificial intelligence based ECG heartbeat automatic recognition classification method.
  • Cardiovascular disease is one of the major diseases that threaten human health.
  • the use of effective means to detect cardiovascular diseases is an important topic of worldwide concern.
  • Electrocardiogram The main method for diagnosing cardiovascular disease in modern medicine. Using ECG to diagnose various cardiovascular diseases is essentially the process of extracting ECG's characteristic data to classify ECG.
  • the expert doctor needs to simultaneously compare the chronological changes of the signals of the lead (except the single-guided data), the correlation (spatial relationship) and variation between the leads, and then Can make a more accurate judgment. And this way of relying on the doctor's experience, the accuracy can not be guaranteed.
  • the object of the present invention is to provide an artificial intelligence-based ECG heartbeat automatic recognition classification method, which improves the defect of the classification error which is easily obtained by the traditional method only by relying on a single lead independent analysis, and greatly improves the heart.
  • the accuracy of the beat classification is to provide an artificial intelligence-based ECG heartbeat automatic recognition classification method, which improves the defect of the classification error which is easily obtained by the traditional method only by relying on a single lead independent analysis, and greatly improves the heart.
  • an artificial intelligence-based ECG heartbeat automatic recognition classification method including:
  • the lead heartbeat data is cut by setting the data amount to generate the lead heartbeat analysis data of the lead;
  • data dimension amplification conversion is performed to obtain four-dimensional tensor data
  • the four-dimensional tensor data is input to the trained LepuEcgCatNet heart rate classification model, and the heartbeat classification information corresponding to the heartbeat analysis data is obtained.
  • the heartbeat analysis data for cutting the lead heartbeat data by the set data amount according to the heartbeat time series data to generate the lead includes:
  • the lead heartbeat data is sampled to the two sides with the set data amount to obtain the heartbeat of the lead. analyze data.
  • the lead heartbeat analysis data is single-lead heartbeat analysis data, and the data of the lead heartbeat analysis data is combined to obtain a one-dimensional heartbeat analysis array, which specifically includes:
  • the single-lead heartbeat analysis data is combined into a one-dimensional heartbeat analysis array according to cardiac time series data.
  • the lead heartbeat analysis data is multi-lead heartbeat analysis data, and the data of the lead heartbeat analysis data is combined to obtain a one-dimensional heartbeat analysis array, which specifically includes:
  • the multi-lead beat analysis data is combined into a one-dimensional heartbeat analysis array according to the lead parameters and heartbeat time series data.
  • the one-dimensional heartbeat analysis array performs data dimension amplification conversion, and the four-dimensional tensor data is specifically included:
  • the four-dimensional tensor data has four factors, respectively height data , width data, channel data, and bulk data;
  • a four-dimensional tensor data is generated based on the batch data and the fused data; the batch data is the number of input samples.
  • the four-dimensional tensor data is input to the trained LepuEcgCatNet heart rate classification model, and the heartbeat classification information corresponding to the heartbeat analysis data is specifically included:
  • the input four-dimensional tensor data is subjected to a layer-by-layer convolution extraction feature, and the heartbeat classification information corresponding to the heartbeat analysis data is obtained through an inference operation.
  • the method further comprises: establishing and training the LepuEcgCatNet heart rate classification model.
  • establishing and training the LepuEcgCatNet heart rate classification model specifically includes:
  • a deep learning model of LepuEcgCatNet multilayer convolutional neural network for endo-end multi-label classification for ECG heartbeat classification is constructed.
  • the LepuEcgCatNet heartbeat classification model is trained to obtain model structure data and parameter data, and to store and encrypt model structure data and parameter data.
  • the training specifically includes:
  • Model structure data and parameter data are encrypted and protected.
  • the ECG heartbeat automatic recognition classification method based on the artificial intelligence self-learning obtains the heartbeat time series data and the lead heart rate data by processing the received digital signal of the original electrocardiogram; according to the heartbeat time series data, the guide The heart beat analysis data is used to cut and generate the heartbeat analysis data of the lead; the data of the lead heartbeat analysis data is combined to obtain an array of one-dimensional heart beat analysis; according to the one-dimensional heart beat analysis array, data dimension amplification conversion is performed, Four-dimensional tensor data is obtained; the four-dimensional tensor data is input to the trained LepuEcgCatNet heart rate classification model to obtain heartbeat classification information.
  • the artificial intelligence-based ECG heartbeat automatic recognition classification method of the invention improves the defect of the classification error which is easily obtained by the traditional method only by relying on the single lead independent analysis, and greatly improves the accuracy of the heartbeat classification. .
  • FIG. 1 is a flowchart of an ECG heartbeat automatic recognition classification method based on artificial intelligence self-learning according to an embodiment of the present invention
  • FIG. 2 is a schematic diagram of a LepuEcgCatNet heart rate classification model according to an embodiment of the present invention
  • FIG. 3 is a schematic diagram of a process of a single-lead heart beat data classification method according to an embodiment of the present invention
  • FIG. 4 is a schematic diagram of a process of a multi-lead voting decision classification method according to an embodiment of the present invention.
  • FIG. 5 is a schematic diagram of a process of a lead synchronization association classification method according to an embodiment of the present invention.
  • FIG. 1 The flow chart of the ECG heartbeat automatic recognition classification method based on the artificial intelligence self-learning provided by the embodiment of the present invention is shown in FIG. 1 , which mainly includes the following steps:
  • Step 110 processing the received original electrocardiogram digital signal, generating heartbeat time series data and lead heartbeat data;
  • the electrocardiogram monitoring device converts the electrocardiogram electrical analog signal into a digital signal output, or may be an electrocardiogram data obtained through a database or other file method, and the original data, that is, the original electrocardiogram digital signal is stored through the data storage transmission device, and It is transmitted to the analysis system hardware module through WIFI, Bluetooth, USB, 3G/4G/5G mobile communication network, Internet of Things, etc., and is input as an input signal to the analysis system execution module.
  • the electrocardiogram time series data generated in the lead circuit, the data encoding format, the gain, the precision are generated due to the difference in the collection of the analog circuit, the filter and the sampling rate of the electrocardiograph device of different equipment manufacturers.
  • the data length per second, the baseline position, etc. are all very different.
  • Data preprocessing must be performed, and all input ECG time series data are uniformly processed according to the requirements of the analysis process of the present invention.
  • the specific data processing receiving process includes the following steps:
  • the raw data is resampled according to a preset data sampling frequency to obtain time characterization data of each data point under a new sampling rate condition; wherein the time characterization data is used to indicate the time of each data point on the time axis of the electrocardiogram data signal information;
  • Step 120 according to the heartbeat time series data, cutting the lead heartbeat data by setting the data amount to generate the lead heartbeat analysis data of the lead;
  • the central sampling point can select the position of the P wave of the lead heartbeat data, the R wave position of the QRS complex, and the position of the T wave.
  • the selection of the central sampling point of the heartbeat data needs to be consistent with the selection of the central sampling point of the training sample of the heartbeat classification model.
  • the data of the lead heartbeat data is sampled to the two sides to obtain the lead heartbeat analysis of the lead. data.
  • the set data amount and the preset data sampling frequency are matched with the relevant parameters of the input data of the trained LepuEcgCatNet heart rate classification model.
  • Step 130 Combine the heartbeat analysis data of the lead to obtain a one-dimensional heartbeat analysis array
  • the cardiac analysis data of the lead may include single-lead heart beat analysis data and multi-lead heart beat analysis data, and the specific processing manners of the two are respectively as follows.
  • the single-lead heartbeat analysis data is combined into a one-dimensional heartbeat analysis array according to the heartbeat time series data.
  • the length of the one-dimensional heartbeat analysis array is set to the length x of the heartbeat time series
  • the content of the one-dimensional heartbeat analysis array is the time-series arrangement of the heartbeat analysis data.
  • Combining the data of the multi-lead heartbeat analysis data to obtain a one-dimensional heartbeat analysis array includes the following steps:
  • the above standard interference data may adopt all zero values, all one values, or other preset values
  • the above-mentioned arranged lead heartbeat analysis data is sequentially added to a one-dimensional heartbeat analysis array as a sample data;
  • Step 140 performing data dimension amplification conversion according to the one-dimensional heartbeat analysis array to obtain four-dimensional tensor data
  • the LepuEcgCatNet heart rate classification model requires a four-dimensional tensor data (b, h, w, d), where b is bulk data, h is height data, w is width data, and d is channel data;
  • the appropriate batch data value is determined according to the set data amount and computer resources used when cutting the lead heartbeat data; specifically, the computer resource refers to the memory size that the computer can use for deep learning, including the memory of the graphics card;
  • the lepuEcgCatNet heartbeat classification model also defines a lead sequence parameter for the specification processing of lead sequence sorting when inputting multi-lead sample data;
  • the one-dimensional heartbeat analysis array data whose number is the batch data value and the length is the set data amount x the lead number is sequentially extracted from the one-dimensional heartbeat analysis array, and then the data is converted into A two-dimensional tensor data, the value of the first dimension is the bulk data value, and the value of the second dimension is the set data amount x the number of leads. That is to say, the one-dimensional heartbeat analysis array combined in the previous step can be converted into a plurality of two-dimensional tensor data of the above size, and each row of the two-dimensional tensor data is a sample data;
  • Each row of the two-dimensional tensor data also needs to be converted into three dimensions of tensor data, the three dimensions are height, width and channel; that is, the two-dimensional tensor data is converted into a four-dimensional tensor data;
  • the conversion can be performed by means of high fusion, width fusion or channel fusion:
  • Each line of the two-dimensional data tensor can be converted into a three-dimensional data tensor by a high degree of fusion, specifically: each line of data of the set amount of data x is converted into a height value of the number of leads, and the width value is set.
  • the amount of data, the channel value is a three-dimensional tensor data; that is, the two-dimensional tensor data is converted into (bulk data value, number of leads, set data amount, 1) such a LepuEcgCatNet heart beat classification model Enter the four-dimensional tensor data required by the format;
  • Each row of the two-dimensional data tensor is converted into a three-dimensional data tensor by width fusion, specifically: each row of data of the set amount of data x is converted into a height value of 1, and the width value is a set data amount.
  • the number of x leads, the channel value is 1 3D tensor data; that is to say the 2D tensor data is converted into (bulk data value, 1, set the amount of data x lead number, 1) in accordance with LepuEcgCatNet heart beat classification Four-dimensional tensor data required by the model input format;
  • Each row of the two-dimensional data tensor is converted into a three-dimensional data tensor by channel fusion, specifically: each row of data of the set amount of data x is converted into a height value of 1, and the width value is a set data amount.
  • the channel value is the three-dimensional tensor data of the number of leads; that is, the two-dimensional tensor data is converted into the first dimension (bulk data value, 1, set data amount, number of leads) conforming to LepuEcgCatNet heart beat classification Four-dimensional tensor data required by the model input format;
  • Different fusion methods can make the original heartbeat analysis data get the information features on the time or space scale, which is beneficial to improve the classification accuracy of the model.
  • Step 150 Input the four-dimensional tensor data into the trained LepuEcgCatNet heart rate classification model, and obtain the heartbeat classification information corresponding to the heartbeat analysis data;
  • the process of inputting the four-dimensional tensor data into the trained LepuEcgCatNet heart rate classification model for performing the inference operation, and obtaining the heartbeat classification information corresponding to the heartbeat analysis data includes the following steps:
  • the model structure data and parameter data of the trained LepuEcgCatNet heart rate classification model are read according to the Google Protocol Buffers data protocol.
  • the model structure data and parameter data are protected by a symmetric encryption algorithm, so it must be performed before use. Decryption operation
  • the lead heart rate data received and processed in the operating environment is cut, combined and converted to generate four-dimensional tensor data satisfying the input data format requirements of the LepuEcgCatNet heart rate classification model;
  • the four-dimensional tensor data is input into the read LepuEcgCatNet heart rate classification model, and the feature is extracted by layer-by-layer convolution, and then calculated by the full connection layer, softmax regression classification, etc., and finally the heartbeat corresponding to the heartbeat analysis data is obtained.
  • Classified information is input into the read LepuEcgCatNet heart rate classification model, and the feature is extracted by layer-by-layer convolution, and then calculated by the full connection layer, softmax regression classification, etc.
  • the LepuEcgCatNet heartbeat classification model can be divided into two types: single-lead heart beat classification model and multi-lead heart beat classification model;
  • the heartbeat classification of single-lead heart beat data using the LepuEcgCatNet single-lead heart beat classification model;
  • the heartbeat classification of multi-lead heart beat data is generally based on the LepuEcgCatNet multi-lead heart beat classification model, which is called the multi-lead synchronous correlation analysis method; but in some special cases, single guide can also be used.
  • the Lianxin beat classification model separately classifies some or all of the lead heartbeat data of the multi-lead beat data, and then performs classification and voting decision calculation according to the classification result of each lead and the lead weight value reference coefficient to obtain the final classification result.
  • This method is called a multi-lead voting decision method; specifically, the lead weight value reference coefficient is based on the ECG big data for Bayesian statistical analysis to obtain the voting weight coefficients of each lead for different heart beat classifications.
  • Figure 3 illustrates the process of the single-lead heart beat data classification method:
  • the single-lead beat data is cut with the first data amount to generate single-lead heartbeat analysis data, and input into the trained LepuEcgCatNet single-lead heart beat classification model.
  • the feature extraction and analysis of the amplitude and time characterization data are performed to obtain the classification information of the single-lead heartbeat data.
  • FIG. 4 illustrates a process for performing a multi-lead heart beat data classification method using a lead voting decision method
  • the lead heartbeat data is cut by the second data amount, thereby generating heartbeat analysis data of each lead;
  • the feature extraction and analysis of the amplitude and time characterization data of each lead heartbeat analysis data are obtained, and the classification information of each lead is obtained.
  • the classification voting decision is calculated according to the classification information of each lead and the reference weight reference value of the lead, and the classification information is obtained once.
  • FIG. 5 illustrates a process of using a lead synchronization association classification to perform a multi-lead heart beat data classification method
  • the lead heartbeat data is cut by the third data amount, thereby generating heartbeat analysis data of each lead;
  • the characteristic analysis and synchronization of the amplitude and time characterization data of each lead heartbeat analysis data are performed, and the primary classification information of the heartbeat analysis data is obtained.
  • the LepuEcgCatNet heartbeat classification model structure is shown in Figure 2. It is an end-to-end multi-label deep convolutional neural network classification model inspired by mature and open models based on artificial intelligence deep learning convolutional neural network AlexNet, VGG16, and Inception. It is a deep learning model specifically for ECG analysis requirements and data characteristics.
  • the model's network is a 7-layer convolutional network, with each convolution followed by an activation function.
  • the first layer is a convolutional layer of two different scales, followed by six convolutional layers.
  • the convolution kernels of the seven-layer convolution are 96, 256, 256, 384, 384, 384, 256, respectively.
  • the training process of the LepuEcgCatNet heart rate classification model includes the following steps:
  • training data In the first step, we selected training data.
  • the annotations are mainly for common arrhythmia, conduction block and ST segment and T wave changes, which can meet the model training of different application scenarios.
  • the marked information is saved in a preset standard data format.
  • a small sliding is performed on the classification with less sample size to amplify the data. Specifically, it is based on each heart beat and according to a certain step. (For example, 10-50 data points) move 2 times, which can increase the data by 2 times, and improve the recognition accuracy of the classified samples with less data. After the actual results are verified, the generalization ability has also been improved.
  • the training sample is converted into a preset standard data format for storage
  • the third step is to cut, combine and convert the training samples.
  • the converted data meets the requirements of the input data format of the LepuEcgCatNet heart rate classification model, which can be used for the training of the model;
  • the test data is trained on different iterations using independent test data of a certain amount of data.
  • the most accurate model is used as the LepuEcgCatNet heartbeat classification model; and the model and parameter data are stored according to the Google Protocol Buffers data protocol, and the symmetric encryption algorithm is sampled for protection.
  • the length of the interception of the training data may be 1 second to 10 seconds.
  • the sampling rate is 200 Hz, with a sampling length of 2.5 s
  • the obtained data volume is a segment D [500] of 500 ECG voltage values (millivolts)
  • data is performed according to the batch data and the number of leads N.
  • the final four-dimensional tensor data can be obtained: four-dimensional tensor data Inputdata (bulk data, N, 500, 1) or width-mixed four-dimensional tensor data Inputdata (bulk data, 1,500xN, 1) or channel Fusion mode four-dimensional tensor data Inputdata (bulk data, 1,500, N). All the input data is randomly dispersed to start training, which ensures the convergence of the training process. At the same time, it controls the collection of too many samples from the same patient's ECG data, improving the generalization ability of the model, and the accuracy of the real scene.
  • the artificial intelligence-based ECG heartbeat automatic recognition classification method of the invention improves the defect of the classification error which is easily obtained by the traditional method only by relying on the single lead independent analysis, and greatly improves the accuracy of the heartbeat classification. .
  • the steps of a method or algorithm described in connection with the embodiments disclosed herein can be implemented in hardware, a software module executed by a processor, or a combination of both.
  • the software module can be placed in random access memory (RAM), memory, read only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or technical field. Any other form of storage medium known.

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Physics & Mathematics (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Theoretical Computer Science (AREA)
  • Molecular Biology (AREA)
  • Artificial Intelligence (AREA)
  • Public Health (AREA)
  • Medical Informatics (AREA)
  • Cardiology (AREA)
  • Biophysics (AREA)
  • Evolutionary Computation (AREA)
  • Data Mining & Analysis (AREA)
  • Pathology (AREA)
  • General Physics & Mathematics (AREA)
  • Veterinary Medicine (AREA)
  • Animal Behavior & Ethology (AREA)
  • Surgery (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Mathematical Physics (AREA)
  • General Engineering & Computer Science (AREA)
  • Primary Health Care (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Computational Linguistics (AREA)
  • Epidemiology (AREA)
  • Physiology (AREA)
  • Signal Processing (AREA)
  • Biodiversity & Conservation Biology (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Multimedia (AREA)
  • Fuzzy Systems (AREA)
  • Psychiatry (AREA)
  • General Business, Economics & Management (AREA)
  • Business, Economics & Management (AREA)

Abstract

一种基于人工智能的心电图心搏自动识别分类方法，包括：处理接收到的原始心电图数字信号，得到心搏时间序列数据和导联心搏数据（110）；根据心搏时间序列数据，对导联心搏数据进行切割生成导联的心搏分析数据（120）；对导联的心搏分析数据进行数据组合，得到一维心搏分析数组（130）；根据一维心搏分析数组，进行数据维度扩增转换，得到四维张量数据（140）；输入四维张量数据到训练得到的LepuEcgCatNet心搏分类模型，得到心搏分类信息（150）。本方法改进了传统方法仅仅依靠单个导联独立分析进行结果汇总统计比较容易得出分类错误的缺陷，极大地提高了心电图心搏分类的准确率。

Description

基于人工智能的心电图心搏自动识别分类方法
本申请要求于2017年11月27日提交中国专利局、申请号为201711203546.7、发明名称为“基于人工智能的心电图心搏自动识别分类方法”的中国专利申请的优先权。
技术领域
本发明涉及人工智能辅助数据分析处理的技术领域，尤其涉及一种基于人工智能的心电图心搏自动识别分类方法。
背景技术
心血管疾病是威胁人类健康的主要疾病之一，利用有效的手段对心血管疾病进行检测是目前全世界关注的重要课题。心电图(ECG)现代医学中诊断心血管疾病的主要方法，利用ECG诊断各种心血管疾病，本质上就是提取ECG的特征数据对ECG进行分类的过程。专家医生在心电图的阅读分析过程中，都是需要同时比较各个导联(单导数据除外)的信号在时间顺序上的前面变化，导联之间的相关性(空间关系)和变异，然后才能够做出一个比较准确的判断。而这种依赖于医生经验的方式，准确率无法得到保障。
随着科技的进步，利用计算机对ECG进行自动准确的分析已经得到了快速的发展。但是，虽然市场上大多数的心电图分析软件都可以对数据进行自动分析，但由于心电图信号本身的复杂与变异性，目前自动分析软件的准确率远远不够，无法达到临床分析使用的要求。
发明内容
本发明的目的是提供一种基于人工智能的心电图心搏自动识别分类方法，改进了传统方法仅仅依靠单个导联独立分析进行结果汇总统计比较容易得出的分类错误的缺陷，极大地提高了心搏分类的准确率。
为实现上述目的，本发明提供了一种基于人工智能的心电图心搏自动识别分类方法，包括：
处理接收到的原始心电图数字信号，生成心搏时间序列数据和导联心搏数据；
根据心搏时间序列数据，以设定数据量对导联心搏数据进行切割生成导联的心搏分析数据；
将导联的心搏分析数据进行数据组合，得到一维心搏分析数组；
根据一维心搏分析数组，进行数据维度扩增转换，得到四维张量数据；
将四维张量数据输入到训练好的LepuEcgCatNet心搏分类模型，得到所述心搏分析数据所对应的心搏分类信息。
优选的，所述根据心搏时间序列数据，以设定数据量对导联心搏数据进行切割生成导联的心搏分析数据具体包括：
依据所述心搏时间序列数据，确定所述导联心搏数据的中心采样点；
以所述导联中心采样点为中心，按照时间表征数据和预设数据采样频率，对所述导联心搏数据以设定数据量向两侧进行数据取样，得到所述导联的心搏分析数据。
优选的，所述导联心搏分析数据为单导联心搏分析数据，所述将导联的心搏分析数据进行数据组合，得到一维心搏分析数组具体包括：
对所述单导联心搏分析数据，按照心搏时间序列数据，组合为一维心搏分析数组。
优选的，所述导联心搏分析数据为多导联心搏分析数据，所述将导联的心搏分析数据进行数据组合，得到一维心搏分析数组具体包括：
对所述多导联心搏分析数据，按照所述导联参数和心搏时间序列数 据，组合为一维心搏分析数组。
优选的，所述一维心搏分析数组，进行数据维度扩增转换，得到四维张量数据具体包括：
将所述一维心搏分析数组以特定转换方式转换成所述训练得到的心搏分类模型所要求的一个四维张量数据输入格式；所述四维张量数据具有四个因子，分别为高度数据、宽度数据、通道数据和批量数据；
其中，所述特定转换方式具体为：
按照输入样本的长度确定高度数据和宽度数据，根据所述高度数据或宽度数据或者通道数据生成融合数据；其中所述通道数据为导联的数量；
根据批量数据和融合数据生成一个四维张量数据；所述批量数据为输入样本的数量。
优选的，将所述四维张量数据输入到训练得到的LepuEcgCatNet心搏分类模型，得到所述心搏分析数据所对应的心搏分类信息具体包括：
根据LepuEcgCatNet心搏分类模型，对输入的四维张量数据进行逐层卷积提取特征，通过推理运算得到所述心搏分析数据所对应的心搏分类信息。
进一步优选的，所述方法还包括：建立并训练所述LepuEcgCatNet心搏分类模型。
进一步优选的，建立并训练所述LepuEcgCatNet心搏分类模型具体包括：
基于深度卷积神经网络AlexNet，Vgg16，ResNet，和Inception这些成熟公开的深度学习模型，构建端对端多标签的针对心电图心搏分类识别的LepuEcgCatNet多层卷积神经网络的深度学习模型；
训练LepuEcgCatNet心搏分类模型得到模型结构数据和参数数据，存储和加密模型结构数据和参数数据。
进一步优选的，所述训练具体包括：
选取训练样本；
将训练样本转换为预设标准数据格式进行存储；
对所述训练样本进行数据切割、组合和转换，对训练样本输入LepuEcgCatNet心搏分类模型进行训练，对训练好的模型结构数据和参数数据按照Google Protocol Buffers数据协议进行存储，并且使用对称加密算法对模型结构数据和参数数据进行加密保护。
本发明实施例提供的基于人工智能自学习的心电图心搏自动识别分类方法通过处理接收到达原始心电图数字信号，得到心搏时间序列数据和导联心搏数据；根据心搏时间序列数据，对导联心搏数据进行切割生成导联的心搏分析数据；对导联的心搏分析数据进行数据组合，得到一维心搏分析数组；根据一维心搏分析数组，进行数据维度扩增转换，得到四维张量数据；输入四维张量数据到训练得到的LepuEcgCatNet心搏分类模型，得到心搏分类信息。本发明的基于人工智能的心电图心搏自动识别分类方法，改进了传统方法仅仅依靠单个导联独立分析进行结果汇总统计比较容易得出的分类错误的缺陷，极大地提高了心搏分类的准确率。
附图说明
图1为本发明实施例提供的基于人工智能自学习的心电图心搏自动识别分类方法流程图；
图2为本发明实施例提供的LepuEcgCatNet心搏分类模型的示意图；
图3为本发明实施例提供的单导联心搏数据分类方法过程示意图；
图4为本发明实施例提供的多导联投票决策分类方法过程示意图。
图5为本发明实施例提供的导联同步关联分类方法过程示意图。
具体实施方式
下面通过附图和实施例，对本发明的技术方案做进一步的详细描述。
为本发明实施例提供的基于人工智能自学习的心电图心搏自动识别分类方法流程图见图1，主要包括如下步骤：
步骤110，处理接收到的原始心电图数字信号，生成心搏时间序列数据和导联心搏数据；
具体的，心电监测设备将心电图电模拟信号转换为数字信号输出，也可以是通过数据库或者其他文件方式获得的心电图数据，通过数据存储传输装置进行原始数据即原始心电图数字信号的存储，并可以通过WIFI、蓝牙，USB，3G/4G/5G移动通信网络，物联网等方式传输至分析系统硬件模块，并作为输入信号输入到分析系统执行模块。
作为记录心电图图形的原始心电图数字信号，由于不同设备厂家的心电图设备在采集模拟电路，滤波器和采样率诸方面的不同，产生的心电图时间序列数据在导联标记，数据编码格式，增益，精度，每秒的数据长度，基线位置等都有很大的差异，必须进行数据预处理，把所有输入的心电图时间序列数据根据本发明分析流程的要求，进行统一处理。具体的数据处理接收流程包括以下这些步骤：
1.原始数据按照预设数据采样频率重采样，获得在新采样率条件下每个数据点的时间表征数据；其中，时间表征数据用于表示每个数据点在心电图数据信号时间轴上的时间信息；
2.对重采样的数据进行数字信号滤波去除高频，低频的噪音干扰和基线漂移；
3.将滤波后的数据按照预设标准数据格式进行数据格式转换；
4.对转换后数据进行心搏检测处理，识别出每个导联的多个心搏数据，形成导联心搏数据；
5.对导联心搏数据进行干扰识别；
6.将导联心搏数据根据干扰识别结果和时间规则进行合并，生成统一的心搏时间序列数据；
步骤120，根据心搏时间序列数据，以设定数据量对导联心搏数据进行切割生成导联的心搏分析数据；
在切割导联心搏数据前，首先需要确定导联心搏数据的中心采样点。的中心采样点，可以选取导联心搏数据的P波位置、QRS波群的R波位置以及T波等位置。心搏数据的中心采样点的选取需要和心搏分类模型的训练样本的中心采样点的选取要一致。
然后以导联心搏数据的中心采样点为中心，按照时间表征数据和预设数据采样频率，对导联心搏数据以设定数据量向两侧进行数据取样，得到导联的心搏分析数据。设定数据量和预设数据采样频率要与训练得到的LepuEcgCatNet心搏分类模型的输入数据的相关参数相匹配。
步骤130，将导联的心搏分析数据进行数据组合，得到一维心搏分析数组；
具体的，导联的心搏分析数据可以包括单导联心搏分析数据和多导联心搏分析数据，二者的具体处理方式分别如下所述。
对单导联心搏分析数据，按照心搏时间序列数据，组合为一维心搏分析数组。此处一维心搏分析数组长度为心搏时间序列的长度x设定数据量，一维心搏分析数组内容为心搏分析数据按时间序列的排列。
对多导联心搏分析数据进行数据组合得到一维心搏分析数组包含以下步骤：
a.根据心搏时间序列数据获取每个时间表征数据；
b.根据上述时间表征数据获取所有导联的心搏分析数据，并按分类模型定义的导联顺序参数进行排序；
c.根据上述时间表征数据获取各导联对应位置的干扰标志，并按分类模型定义的导联顺序参数进行排序；
d.检查对应导联位置的干扰标志，如果是干扰，则以设定数据量大小的标准干扰数据替换。上述的标准干扰数据可采用全0值、全1值、或 者其他预设值；
e.将上述排列好的各导联心搏分析数据，作为一个样本数据，顺序加入到一个一维心搏分析数组；
f.循环a到e操作，直到把所有心搏时间序列数据对应的导联心搏分析数据都转换为样本数据加入到上述的一维心搏分析数组，完成一维心搏分析数组的样本数据数据组合。
步骤140，根据一维心搏分析数组，进行数据维度扩增转换，得到四维张量数据；
LepuEcgCatNet心搏分类模型对输入数据格式的要求是一个四维张量数据(b,h,w,d)，其中b是批量数据，h是高度数据，w是宽度数据，d是通道数据；具体的，批量数据就是输入样本的样本数量，高度数据x宽度数据＝每个样本的长度，即设定数据量，通道数据＝导联数量。
首先，根据切割导联心搏数据时所用的设定数据量和计算机资源，来确定合适的批量数据值；具体的，计算机资源是指计算机可用于深度学习的内存大小，包括图形显卡的内存；同时，lepuEcgCatNet心搏分类模型还定义了一个导联顺序参数，用于对多导联样本数据输入时进行导联排序的规范处理；
按照批量数据值和导联数量依次从一维心搏分析数组中取出个数为批量数据值，长度为设定数据量x导联数量的一维心搏分析数组数据，接着将该数据转换成一个二维张量数据，第一个维度的值是批量数据值，第二个维度的值是设定数据量x导联数量。也就是说上一步骤组合成的一维心搏分析数组可以转换成多个上述大小的二维张量数据，二维张量数据的每一行就是一个样本数据；
二维张量数据的每一行还需要转换成有三个维度的张量数据，三个维度分别是高度、宽度和通道；也就是说要将二维张量数据转换成一个四维张量数据；
具体的，可以通过高度融合、宽度融合或者通道融合的方式来进行转换：
通过高度融合可以将二维数据张量的每一行转换成三维数据张量，具体为：将设定数据量x导联数量的每一行数据，转换成高度值为导联数量，宽度值为设定数据量，通道值为一的三维张量数据；也就是说二维张量数据转换成了为(批量数据值，导联数量，设定数据量，1)这样一个符合LepuEcgCatNet心搏分类模型输入格式要求的四维张量数据；
通过宽度融合将二维数据张量的每一行转换成三维数据张量，具体为：将设定数据量x导联数量的每一行数据，转换成高度值为1，宽度值为设定数据量x导联数量，通道值为1的三维张量数据；也就是说二维张量数据转换成了(批量数据值，1，设定数据量x导联数量，1)的符合LepuEcgCatNet心搏分类模型输入格式要求的四维张量数据；
通过通道融合将二维数据张量的每一行转换成三维数据张量，具体为：将设定数据量x导联数量的每一行数据，转换成高度值为1，宽度值为设定数据量，通道值为导联数量的三维张量数据；也就是说二维张量数据转换成了第一维度为(批量数据值，1，设定数据量，导联数量)的符合LepuEcgCatNet心搏分类模型输入格式要求的四维张量数据；
不同的融合方式可以使原来心搏分析数据在时间或空间尺度上得到信息特征融合，用利于提高模型分类精度。
步骤150，将四维张量数据输入到训练得到的LepuEcgCatNet心搏分类模型，得到心搏分析数据所对应的心搏分类信息；
具体的，将四维张量数据输入到训练得到的LepuEcgCatNet心搏分类模型进行推理运算，得到心搏分析数据所对应的心搏分类信息的过程包括以下步骤：
第一步，按照Google Protocol Buffers数据协议读取训练好的LepuEcgCatNet心搏分类模型的模型结构数据和参数数据，模型结构数据 和参数数据是经过对称加密算法进行保护的，所以使用前，还必须进行解密操作；
第二步，将运行环境中接收处理后的导联心搏数据进行切割、组合和转换，生成满足LepuEcgCatNet心搏分类模型输入数据格式要求的四维张量数据；
第三步，将四维张量数据输入到读取的LepuEcgCatNet心搏分类模型，经过逐层卷积提取特征，依次经过全连接层、softmax回归分类等计算，最后得到对应心搏分析数据的心搏分类信息。
LepuEcgCatNet心搏分类模型具体的可以分成单导联心搏分类模型和多导联心搏分类模型两种；
单导联心搏数据的心搏分类，使用LepuEcgCatNet单导联心搏分类模型；
多导联心搏数据的心搏分类，一般情况是使用LepuEcgCatNet多导联心搏分类模型，称这种方法为多导联同步关联分析方法；但在某些特殊情况下，也可以使用单导联心搏分类模型分别对多导联心搏数据的部分或者全部导联心搏数据进行独立分类，然后根据各导联的分类结果和导联权重值参考系数进行分类投票决策计算得到最终分类结果，这种方法叫多导联投票决策方法；具体的，导联权重值参考系数是基于心电图大数据进行贝叶斯统计分析得到各导联对不同心搏分类的投票权重系数。
具体的，图3说明了单导联心搏数据分类方法过程：
根据心搏时间序列数据，以第一数据量将单导联心搏数据进行切割生成单导联的心搏分析数据，并输入到训练得到的对应该导联的LepuEcgCatNet单导联心搏分类模型进行幅值和时间表征数据的特征提取和分析，得到单导联的心搏数据的分类信息。
具体的，图4说明了使用导联投票决策方法进行多导联心搏数据分类方法过程；
第一步、根据心搏时间序列数据，以第二数据量对各导联心搏数据进行切割，从而生成各导联的心搏分析数据；
第二步、根据训练得到的各导联对应的LepuEcgCatNet单导联心搏分类模型对各导联的心搏分析数据进行幅值和时间表征数据的特征提取和分析，得到各导联的分类信息；
第三步、根据各导联的分类信息和导联权重值参考系数进行分类投票决策计算，得到一次分类信息。
具体的，图5说明了使用导联同步关联分类进行多导联心搏数据分类方法过程；
第一步、根据心搏时间序列数据，以第三数据量对各导联心搏数据进行切割，从而生成各导联的心搏分析数据；
第二步、根据训练得到的LepuEcgCatNet多导联心搏分类模型对各导联的心搏分析数据进行同步幅值和时间表征数据的特征提取和分析，得到心搏分析数据的一次分类信息。
LepuEcgCatNet心搏分类模型结构如图2所示，是基于人工智能深度学习的卷积神经网络AlexNet，VGG16，Inception这些成熟公开的模型启发下构建的端对端多标签深度卷积神经网络分类模型，是一个专门针对心电图分析要求和数据特征的深度学习模型。具体的讲，该模型的网络是一个7层的卷积网络，每个卷积之后紧跟一个激活函数。第一层是两个不同尺度的卷积层，之后是六个卷积层。七层卷积的卷积核分别是96,256,256,384,384,384,256。除第一层卷积核有两个尺度分别是5和11外，其他层卷积核尺度为5。第三、五、六、七层卷积层后是池化层。最后跟着两个全连接层。具体的，LepuEcgCatNet心搏分类模型的训练过程包括以下步骤：
第一步，选取训练数据，我们采用了训练集包含30万病人的1700万数据样本进行训练。这些样本是根据心电图分析诊断的要求对数据进行准确 的标注产生的，标注主要是针对常见心律失常，传导阻滞以及ST段和T波改变，可满足不同应用场景的模型训练。具体以预设标准数据格式保存标注的信息。在训练数据的预处理上，为增加模型的泛化能力，对于样本量较少的分类做了小幅的滑动来扩增数据，具体的说，就是以每个心搏为基础，按照一定步长(比如10-50个数据点)移动2次，这样就可以增加2倍的数据，提高了对这些数据量比较少的分类样本的识别准确率。经过实际结果验证，泛化能力也得到了改善。
第二步，将训练样本转换为预设标准数据格式进行存储；
第三步，对训练样本进行数据切割、组合和转换，经过转换后的数据则符合LepuEcgCatNet心搏分类模型对输入数据格式的要求，可用于该模型的训练；
第四步，在一个实际训练过程使用了两台GPU服务器进行几十次轮循训练，训练收敛后，使用一定数据量的独立的测试数据对不同迭代次数训练而成的模型进行测试，选取测试精度最高的模型做为LepuEcgCatNet心搏分类模型；并且按照Google Protocol Buffers数据协议存储模型和参数数据，并且采样对称加密算法进行保护。
其中，训练数据的截取的长度，可以是1秒到10秒。比如采样率是200Hz，以2.5s为采样长度，取得的设定数据量大小是500个心电图电压值(毫伏)的一个片段D[500]，依据批量数据和导联个数N,进行数据转换，可得最终四维张量数据：高度融合方式的四维张量数据Inputdata(批量数据,N,500,1)或宽度融合方式的四维张量数据Inputdata(批量数据,1,500xN,1)或通道融合方式的四维张量数据Inputdata(批量数据,1,500,N)。输入数据全部经过随机打散才开始训练，保证了训练过程收敛；同时，控制从同一个病人的心电图数据中收集太多的样本，提高模型的泛化能力，既真实场景下的准确率。
本发明的基于人工智能的心电图心搏自动识别分类方法，改进了传统方 法仅仅依靠单个导联独立分析进行结果汇总统计比较容易得出的分类错误的缺陷，极大地提高了心搏分类的准确率。
专业人员应该还可以进一步意识到，结合本文中所公开的实施例描述的各示例的单元及算法步骤，能够以电子硬件、计算机软件或者二者的结合来实现，为了清楚地说明硬件和软件的可互换性，在上述说明中已经按照功能一般性地描述了各示例的组成及步骤。这些功能究竟以硬件还是软件方式来执行，取决于技术方案的特定应用和设计约束条件。专业技术人员可以对每个特定的应用来使用不同方法来实现所描述的功能，但是这种实现不应认为超出本发明的范围。
结合本文中所公开的实施例描述的方法或算法的步骤可以用硬件、处理器执行的软件模块，或者二者的结合来实施。软件模块可以置于随机存储器(RAM)、内存、只读存储器(ROM)、电可编程ROM、电可擦除可编程ROM、寄存器、硬盘、可移动磁盘、CD-ROM、或技术领域内所公知的任意其它形式的存储介质中。
以上所述的具体实施方式，对本发明的目的、技术方案和有益效果进行了进一步详细说明，所应理解的是，以上所述仅为本发明的具体实施方式而已，并不用于限定本发明的保护范围，凡在本发明的精神和原则之内，所做的任何修改、等同替换、改进等，均应包含在本发明的保护范围之内。

Claims (9)

  1. 一种基于人工智能的心电图心搏自动识别分类方法，其特征在于，所述方法包括：
    处理接收到的原始心电图数字信号，生成心搏时间序列数据和导联心搏数据；
    根据心搏时间序列数据，以设定数据量对导联心搏数据进行切割生成导联的心搏分析数据；
    将导联的心搏分析数据进行数据组合，得到一维心搏分析数组；
    根据一维心搏分析数组，进行数据维度扩增转换，得到四维张量数据；
    将四维张量数据输入到训练好的LepuEcgCatNet心搏分类模型，得到所述心搏分析数据所对应的心搏分类信息。
  2. 根据权利要求1所述的心电图心搏自动识别分类方法，其特征在于，所述根据心搏时间序列数据，以设定数据量对导联心搏数据进行切割生成导联的心搏分析数据具体包括：
    依据所述心搏时间序列数据，确定所述导联心搏数据的中心采样点；
    以所述导联中心采样点为中心，按照时间表征数据和预设数据采样频率，对所述导联心搏数据以设定数据量向两侧进行数据取样，得到所述导联的心搏分析数据。
  3. 根据权利要求1所述的心电图心搏自动识别分类方法，其特征在于，所述导联心搏分析数据为单导联心搏分析数据，所述将导联的心搏分析数据进行数据组合，得到一维心搏分析数组具体包括：
    对所述单导联心搏分析数据，按照心搏时间序列数据，组合为一维心搏分析数组。
  4. 根据权利要求1所述的心电图心搏自动识别分类方法，其特征在于，所述导联心搏分析数据为多导联心搏分析数据，所述将导联的心搏分析数据进行数据组合，得到一维心搏分析数组具体包括：
    对所述多导联心搏分析数据，按照所述导联参数和心搏时间序列数据，组合为一维心搏分析数组。
  5. 根据权利要求1所述的心电图心搏自动识别分类方法，其特征在于，所述一维心搏分析数组，进行数据维度扩增转换，得到四维张量数据具体包括：
    将所述一维心搏分析数组以特定转换方式转换成所述训练得到的心搏分类模型所要求的一个四维张量数据输入格式；所述四维张量数据具有四个因子，分别为高度数据、宽度数据、通道数据和批量数据；
    其中，所述特定转换方式具体为：
    按照输入样本的长度确定高度数据和宽度数据，根据所述高度数据或宽度数据或者通道数据生成融合数据；其中所述通道数据为导联的数量；
    根据批量数据和融合数据生成一个四维张量数据；所述批量数据为输入样本的数量。
  6. 根据权利要求1所述的心电图心搏自动识别分类方法，其特征在于，将所述四维张量数据输入到训练得到的LepuEcgCatNet心搏分类模型，得到所述心搏分析数据所对应的心搏分类信息具体包括：
    根据LepuEcgCatNet心搏分类模型，对输入的四维张量数据进行逐层卷积提取特征，通过推理运算得到所述心搏分析数据所对应的心搏分类信息。
  7. 根据权利要求6所述的心电图心搏自动识别分类方法，其特征在于，所述方法还包括：建立并训练所述LepuEcgCatNet心搏分类模型。
  8. 根据权利要求7所述的心电图心搏自动识别分类方法，其特征在于，建立并训练所述LepuEcgCatNet心搏分类模型具体包括：
    基于深度卷积神经网络AlexNet，Vgg16，ResNet，和Inception这些成熟公开的深度学习模型，构建端对端多标签的针对心电图心搏分类识别的LepuEcgCatNet多层卷积神经网络的深度学习模型；
    训练LepuEcgCatNet心搏分类模型得到模型结构数据和参数数据，存储和加密模型结构数据和参数数据。
  9. 根据权利要求7所述的方法，其特征在于，所述训练具体包括：
    选取训练样本；
    将训练样本转换为预设标准数据格式进行存储；
    对所述训练样本进行数据切割、组合和转换，对训练样本输入LepuEcgCatNet心搏分类模型进行训练，对训练好的模型结构数据和参数数据按照Google Protocol Buffers数据协议进行存储，并且使用对称加密算法对模型结构数据和参数数据进行加密保护。
PCT/CN2018/072350 2017-11-27 2018-01-12 基于人工智能的心电图心搏自动识别分类方法 WO2019100562A1 (zh)

Priority Applications (3)

Application Number Priority Date Filing Date Title
EP18880577.4A EP3692901A4 (en) 2017-11-27 2018-01-12 METHOD FOR AUTOMATIC DETECTION AND CLASSIFICATION OF ELECTROCARDIOGRAM HEARTBATE BASED ON ARTIFICIAL INTELLIGENCE
US16/755,105 US11564612B2 (en) 2017-11-27 2018-01-12 Automatic recognition and classification method for electrocardiogram heartbeat based on artificial intelligence
JP2020518512A JP7018133B2 (ja) 2017-11-27 2018-01-12 人工知能に基づく心電図心拍自動識別分類方法

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201711203546.7A CN107981858B (zh) 2017-11-27 2017-11-27 基于人工智能的心电图心搏自动识别分类方法
CN201711203546.7 2017-11-27

Publications (1)

Publication Number Publication Date
WO2019100562A1 true WO2019100562A1 (zh) 2019-05-31

Family

ID=62033176

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/072350 WO2019100562A1 (zh) 2017-11-27 2018-01-12 基于人工智能的心电图心搏自动识别分类方法

Country Status (5)

Country Link
US (1) US11564612B2 (zh)
EP (1) EP3692901A4 (zh)
JP (1) JP7018133B2 (zh)
CN (1) CN107981858B (zh)
WO (1) WO2019100562A1 (zh)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112401903A (zh) * 2020-11-03 2021-02-26 沈阳东软智能医疗科技研究院有限公司 心电数据的识别方法、装置、存储介质和电子设备
CN112587148A (zh) * 2020-12-01 2021-04-02 上海数创医疗科技有限公司 一种包含模糊化相似性测量方法的模板生成方法和装置
CN112990025A (zh) * 2021-03-19 2021-06-18 北京京东拓先科技有限公司 用于处理数据的方法、装置、设备以及存储介质
CN113384277A (zh) * 2020-02-26 2021-09-14 京东方科技集团股份有限公司 心电数据的分类方法及分类系统
CN115429284A (zh) * 2022-09-16 2022-12-06 山东科技大学 心电信号分类方法、系统、计算机设备以及可读存储介质

Families Citing this family (31)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108720831B (zh) * 2018-05-12 2021-01-15 鲁东大学 一种基于导联深度神经网络的自动心律失常分析方法
CN109033990B (zh) * 2018-07-02 2021-08-17 河南大学 基于类内类间距离的cnn模型心拍分类方法
CN110897629A (zh) * 2018-09-14 2020-03-24 杭州脉流科技有限公司 基于深度学习算法的心电特征提取方法、装置、系统、设备和分类方法
CN109544512B (zh) * 2018-10-26 2020-09-18 浙江大学 一种基于多模态的胚胎妊娠结果预测装置
CN109602414B (zh) * 2018-11-12 2022-01-28 安徽心之声医疗科技有限公司 一种多视角转换的心电信号数据增强方法
CN109222963A (zh) * 2018-11-21 2019-01-18 燕山大学 一种基于卷积神经网络的心电异常识别分类方法
CN109567789B (zh) * 2018-12-03 2021-08-31 东软集团股份有限公司 心电图数据的分割处理方法、装置及可读存储介质
CN109770891B (zh) * 2019-01-31 2022-04-29 上海交通大学 心电信号预处理方法及预处理装置
CN109674465A (zh) * 2019-03-05 2019-04-26 邢台医学高等专科学校 一种心电图自动识别系统
CN110327034B (zh) * 2019-04-15 2022-07-15 浙江工业大学 一种基于深度特征融合网络的心动过速心电图的筛查方法
CN110236521A (zh) * 2019-05-20 2019-09-17 上海数创医疗科技有限公司 用于心电图类型识别的双卷积神经网络
WO2021019984A1 (ja) * 2019-07-29 2021-02-04 株式会社カルディオインテリジェンス 心電図表示装置、心電図表示方法及びプログラム
CN110693483A (zh) * 2019-09-02 2020-01-17 乐普智芯(天津)医疗器械有限公司 一种动态心电图自动分析的方法
CN111161874A (zh) * 2019-12-23 2020-05-15 乐普（北京）医疗器械股份有限公司 一种智能心电分析装置
CN111310572B (zh) * 2020-01-17 2023-05-05 上海乐普云智科技股份有限公司 利用心搏时间序列生成心搏标签序列的处理方法和装置
CN111126350B (zh) * 2020-01-17 2023-05-05 上海乐普云智科技股份有限公司 一种心搏分类结果的生成方法和装置
CN111275093B (zh) * 2020-01-17 2024-01-26 上海乐普云智科技股份有限公司 一种用于多标签标注心电信号的心搏分类方法和装置
CN111797816B (zh) * 2020-07-21 2024-05-14 迪姆软件(北京)有限公司 心电图的分析方法及分析装置
CN111901000A (zh) * 2020-08-03 2020-11-06 电子科技大学 基于Inception结构的信道译码方法
CN112244855B (zh) * 2020-11-13 2022-08-26 上海乐普云智科技股份有限公司 一种面向心电数据的数据处理系统
CN112587150B (zh) * 2020-12-01 2022-02-18 上海数创医疗科技有限公司 利用模糊化相似性模板的心电信号类型识别方法和装置
CN112861607B (zh) * 2020-12-29 2022-10-04 湖北航天飞行器研究所 一种远距离激光活体识别方法
CN112883803B (zh) * 2021-01-20 2023-09-01 武汉中旗生物医疗电子有限公司 一种基于深度学习的心电信号分类方法、装置及存储介质
CN112869753B (zh) * 2021-02-10 2023-03-07 上海交通大学医学院附属瑞金医院 一种针对心电图qrst波形的分析方法、设备、介质和心电图仪
CN113052229B (zh) * 2021-03-22 2023-08-29 武汉中旗生物医疗电子有限公司 一种基于心电数据的心脏病症分类方法及装置
CN113344040A (zh) * 2021-05-20 2021-09-03 深圳索信达数据技术有限公司 图像分类方法、装置、计算机设备和存储介质
CN113349753A (zh) * 2021-07-19 2021-09-07 成都芯跳医疗科技有限责任公司 一种基于便携式动态心电监护仪的心律失常检测方法
JP7126230B1 (ja) 2022-02-04 2022-08-26 公立大学法人会津大学 入浴者見守りシステム
CN114722950B (zh) * 2022-04-14 2023-11-07 武汉大学 一种多模态的多变量时间序列自动分类方法及装置
CN116188867B (zh) * 2023-03-07 2023-10-31 山东省人工智能研究院 一种基于注意力增强网络的多标签心电图像分类方法
CN116612899B (zh) * 2023-07-19 2023-10-10 潍坊医学院附属医院 基于互联网的心血管外科数据处理方法及服务平台

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102697492A (zh) * 2012-05-25 2012-10-03 长春华讯信息科技有限公司 一种实时心电信息分析系统及其分析方法
US9408543B1 (en) * 2012-08-17 2016-08-09 Analytics For Life Non-invasive method and system for characterizing cardiovascular systems for all-cause mortality and sudden cardiac death risk
CN106725420A (zh) * 2015-11-18 2017-05-31 中国科学院苏州纳米技术与纳米仿生研究所 室性早搏识别方法及室性早搏识别系统
WO2017091736A1 (en) * 2015-11-23 2017-06-01 Mayo Foundation For Medical Education And Research Processing physiological electrical data for analyte assessments

Family Cites Families (20)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7343197B2 (en) * 2000-05-30 2008-03-11 Vladimir Shusterman Multi-scale analysis and representation of physiological and health data
US7123954B2 (en) * 2002-09-19 2006-10-17 Sanjiv Mathur Narayan Method for classifying and localizing heart arrhythmias
EP1638463A4 (en) 2003-07-01 2007-11-21 Cardiomag Imaging Inc USE OF MACHINE LEARNING FOR CLASSIFICATION OF MAGNETOCARDIOGRAMS
US7996075B2 (en) 2004-10-20 2011-08-09 Cardionet, Inc. Monitoring physiological activity using partial state space reconstruction
US20070260151A1 (en) * 2006-05-03 2007-11-08 Clifford Gari D Method and device for filtering, segmenting, compressing and classifying oscillatory signals
KR101910576B1 (ko) 2011-11-08 2018-12-31 삼성전자주식회사 인공신경망을 이용하여 신속하게 입력 패턴을 분류하는 방법 및 장치
WO2015051280A1 (en) * 2013-10-03 2015-04-09 University Of Washington Through Its Center For Commercialization Apparatuses and methods for classification of electrocardiogram signals during cardiopulmonary resuscitation
CN104970789B (zh) * 2014-04-04 2017-12-19 中国科学院苏州纳米技术与纳米仿生研究所 心电图分类方法及系统
JP6535729B2 (ja) 2014-08-01 2019-06-26 ピアッジオ・アンド・シー．・エス．ピー．エー．Ｐｉａｇｇｉｏ ＆ Ｃ． Ｓ．Ｐ．Ａ． チェーンテンショニングロッド
CN104161510B (zh) 2014-08-27 2016-03-23 电子科技大学 一种多级导联心电信号qrs波形识别方法
CN104367317B (zh) * 2014-10-15 2017-04-12 北京理工大学 多尺度特征融合的心电图心电信号分类方法
KR20170075012A (ko) 2014-10-31 2017-06-30 아이리듬 테크놀로지스, 아이엔씨 무선 생리학적 모니터링 기기 및 시스템
CN105092266B (zh) 2015-09-17 2017-10-17 山东建筑大学 轮胎水滑性能动态测试实验台及实验方法
CN105902266A (zh) 2016-04-22 2016-08-31 江苏物联网研究发展中心 基于自组织神经网络的心电信号分类方法
CN106108889B (zh) * 2016-07-20 2019-11-08 杨一平 基于深度学习算法的心电图分类方法
CN106214145B (zh) * 2016-07-20 2019-12-10 杨一平 一种基于深度学习算法的心电图分类方法
CN106901723A (zh) * 2017-04-20 2017-06-30 济南浪潮高新科技投资发展有限公司 一种心电图异常自动诊断方法
CN107137072B (zh) * 2017-04-28 2019-09-24 北京科技大学 一种基于1d卷积神经网络的室性异位搏动检测方法
CN107203692B (zh) * 2017-05-09 2020-05-05 哈尔滨工业大学（威海） 基于深度卷积神经网络的心电数据数字信号处理方法
CN107358196B (zh) * 2017-07-12 2020-11-10 北京卫嘉高科信息技术有限公司 一种心搏类型的分类方法、装置及心电仪

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102697492A (zh) * 2012-05-25 2012-10-03 长春华讯信息科技有限公司 一种实时心电信息分析系统及其分析方法
US9408543B1 (en) * 2012-08-17 2016-08-09 Analytics For Life Non-invasive method and system for characterizing cardiovascular systems for all-cause mortality and sudden cardiac death risk
CN106725420A (zh) * 2015-11-18 2017-05-31 中国科学院苏州纳米技术与纳米仿生研究所 室性早搏识别方法及室性早搏识别系统
WO2017091736A1 (en) * 2015-11-23 2017-06-01 Mayo Foundation For Medical Education And Research Processing physiological electrical data for analyte assessments

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP3692901A4 *

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113384277A (zh) * 2020-02-26 2021-09-14 京东方科技集团股份有限公司 心电数据的分类方法及分类系统
CN113384277B (zh) * 2020-02-26 2022-09-20 京东方科技集团股份有限公司 心电数据的分类方法及分类系统
CN112401903A (zh) * 2020-11-03 2021-02-26 沈阳东软智能医疗科技研究院有限公司 心电数据的识别方法、装置、存储介质和电子设备
CN112401903B (zh) * 2020-11-03 2023-12-22 沈阳东软智能医疗科技研究院有限公司 心电数据的识别方法、装置、存储介质和电子设备
CN112587148A (zh) * 2020-12-01 2021-04-02 上海数创医疗科技有限公司 一种包含模糊化相似性测量方法的模板生成方法和装置
CN112587148B (zh) * 2020-12-01 2023-02-17 上海数创医疗科技有限公司 一种包含模糊化相似性测量方法的模板生成方法和装置
CN112990025A (zh) * 2021-03-19 2021-06-18 北京京东拓先科技有限公司 用于处理数据的方法、装置、设备以及存储介质
CN115429284A (zh) * 2022-09-16 2022-12-06 山东科技大学 心电信号分类方法、系统、计算机设备以及可读存储介质
CN115429284B (zh) * 2022-09-16 2024-05-03 山东科技大学 心电信号分类方法、系统、计算机设备以及可读存储介质

Also Published As

Publication number Publication date
EP3692901A1 (en) 2020-08-12
EP3692901A4 (en) 2021-07-07
JP2020535882A (ja) 2020-12-10
US20200237246A1 (en) 2020-07-30
CN107981858A (zh) 2018-05-04
CN107981858B (zh) 2020-12-01
JP7018133B2 (ja) 2022-02-09
US11564612B2 (en) 2023-01-31

Similar Documents

Publication Publication Date Title
WO2019100562A1 (zh) 基于人工智能的心电图心搏自动识别分类方法
US11529103B2 (en) Artificial intelligence self-learning-based automatic electrocardiography analysis method and apparatus
US20200335217A1 (en) Methods and systems using mathematical analysis and machine learning to diagnose disease
JP6986724B2 (ja) 人工知能に基づく心電図干渉識別方法
WO2019100565A1 (zh) 基于人工智能自学习的动态心电图分析方法和装置
US11783924B2 (en) ECG information processing method and ECG workstation
CN108511055B (zh) 基于分类器融合及诊断规则的室性早搏识别系统及方法
CN108090509B (zh) 一种数据长度自适应的心电图分类方法
WO2021037102A1 (zh) 基于图片及心搏信息的心电分析方法、装置、设备及介质
US20220093215A1 (en) Discovering genomes to use in machine learning techniques
CN109009102B (zh) 一种基于脑电图深度学习的辅助诊断方法及系统
CN111126350B (zh) 一种心搏分类结果的生成方法和装置
CN112329609A (zh) 基于2d心拍的特征融合迁移学习心律失常分类系统
CN115281688A (zh) 一种基于多模态深度学习的心脏肥大多标签检测系统
CN115062655A (zh) 一种多导联心电信号分析方法及多导联心肌梗死分析系统
CN113229798B (zh) 模型迁移训练方法、装置、计算机设备及可读存储介质
Saadi et al. Low-cost ecg monitoring system with classification using deep learning
CN112842355A (zh) 基于深度学习目标检测的心电信号心搏检测识别方法
KR102618522B1 (ko) 심전도 신호를 이용한 파킨슨병 분석 장치 및 방법
Tihak et al. Classification of Atrial Fibrillation ECG Signals Using 2D CNN
Xia Reliable and decentralised deep learning for physiological data
Kundra An Energy Aware VLSI Framework for ECG Features Extraction for Wearable Devices in Healthcare
CN115919324A (zh) 心电图的分类方法、训练方法、装置、设备和存储介质
CN116172569A (zh) 基于随机卷积的心电分类方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18880577

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2020518512

Country of ref document: JP

Kind code of ref document: A

ENP Entry into the national phase

Ref document number: 2018880577

Country of ref document: EP

Effective date: 20200504

NENP Non-entry into the national phase

Ref country code: DE