WO2020047750A1 - 心律失常的检测方法、装置、电子设备及计算机存储介质 - Google Patents

心律失常的检测方法、装置、电子设备及计算机存储介质 Download PDF

Info

Publication number
WO2020047750A1
WO2020047750A1 PCT/CN2018/104002 CN2018104002W WO2020047750A1 WO 2020047750 A1 WO2020047750 A1 WO 2020047750A1 CN 2018104002 W CN2018104002 W CN 2018104002W WO 2020047750 A1 WO2020047750 A1 WO 2020047750A1
Authority
WO
WIPO (PCT)
Prior art keywords
signal
neural network
ecg signal
arrhythmia
processing
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.)
Ceased
Application number
PCT/CN2018/104002
Other languages
English (en)
French (fr)
Chinese (zh)
Inventor
姚启航
李烨
樊小毛
蔡云鹏
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenzhen Institute of Advanced Technology of CAS
Original Assignee
Shenzhen Institute of Advanced Technology of CAS
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 Shenzhen Institute of Advanced Technology of CAS filed Critical Shenzhen Institute of Advanced Technology of CAS
Priority to PCT/CN2018/104002 priority Critical patent/WO2020047750A1/zh
Priority to CN201880001770.4A priority patent/CN111163690B/zh
Priority to EP18932602.8A priority patent/EP3847958A4/en
Priority to JP2020568775A priority patent/JP7304901B2/ja
Priority to KR1020207035714A priority patent/KR102463764B1/ko
Publication of WO2020047750A1 publication Critical patent/WO2020047750A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

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
    • A61B5/72—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
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
    • A61B5/316—Modalities, i.e. specific diagnostic methods
    • A61B5/318—Heart-related electrical modalities, e.g. electrocardiography [ECG]
    • A61B5/346—Analysis of electrocardiograms
    • A61B5/349—Detecting specific parameters of the electrocardiograph cycle
    • A61B5/361—Detecting fibrillation
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
    • A61B5/316—Modalities, i.e. specific diagnostic methods
    • A61B5/318—Heart-related electrical modalities, e.g. electrocardiography [ECG]
    • A61B5/346—Analysis of electrocardiograms
    • A61B5/349—Detecting specific parameters of the electrocardiograph cycle
    • A61B5/363—Detecting tachycardia or bradycardia
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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 OR CALCULATING; 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
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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
    • G06N3/082—Learning methods modifying the architecture, e.g. adding, deleting or silencing nodes or connections
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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
    • G06N3/09—Supervised learning
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching

Definitions

  • the present application relates to the technical field of electrocardiogram detection, and in particular, to a method, a device, an electronic device, and a computer storage medium for detecting an arrhythmia.
  • Arrhythmia refers to abnormal heart rhythm, frequency or agitation sequence caused by cardiac pacing and conduction dysfunction. It has multiple manifestations such as tachycardia, bradycardia, arrhythmia, and cardiac arrest or fibrillation. It is of great significance for the prevention of cardiovascular diseases.
  • the traditional detection method of arrhythmia is manual detection, that is, the electrocardiogram is analyzed and judged by a doctor with professional knowledge to determine the type of arrhythmia.
  • the application of convolutional neural network models to automatically detect arrhythmia types has gradually appeared.
  • existing convolutional neural network models can only detect arrhythmias on ECG signals of a fixed length, and are difficult to apply to ECG signals of different lengths, that is, they are limited by the different signal lengths commonly present in ECG data. , Poor universality.
  • the purpose of this application is to provide an arrhythmia detection method, device, electronic device and computer storage medium, which can perform arrhythmia detection on ECG signals of different lengths, effectively improving the universality of arrhythmia detection. .
  • a method for detecting an arrhythmia is provided.
  • the method is executed by a processor of an electronic device.
  • the method includes: obtaining an electrocardiogram signal to be detected; and inputting the electrocardiogram signal to a pre-established heart rhythm An arrhythmia detection model; wherein the arrhythmia detection model includes a convolutional neural network and a recurrent neural network connected in sequence; detecting the ECG signal through the arrhythmia detection model to obtain a corresponding one of the ECG signals Test results; the test results include arrhythmia types.
  • the step of detecting the ECG signal by using the arrhythmia detection model to obtain a detection result corresponding to the ECG signal includes: using the convolutional neural network Performing feature extraction on the input ECG signal to obtain feature information of the ECG signal, and inputting the feature information to the recurrent neural network; performing the feature information input through the recurrent neural network Classification processing to determine the type of arrhythmia of the ECG signal.
  • the step of performing feature extraction on the input ECG signal through the convolutional neural network to obtain characteristic information of the ECG signal includes: through the convolution The neural network performs superposition convolution processing and non-linear transformation processing on the ECG signal to obtain a characteristic time series; wherein the characteristic time series includes a characteristic segment extracted from a plurality of ECG signals arranged in time sequence.
  • the step of classifying the input characteristic information through the recurrent neural network to determine an arrhythmia type of the ECG signal includes: using the recurrent neural network. Iteratively process each of the characteristic segments in the characteristic time series to determine the final state of the characteristic time series, and perform classification processing on the final state of the characteristic time series to determine the Arrhythmia types.
  • the training process of the arrhythmia detection model includes: obtaining a training signal; inputting the training signal to the arrhythmia detection model to be trained, and passing a loss function of the arrhythmia detection model Calculate a loss function value; based on the loss function value, train parameters of the arrhythmia detection model through a back propagation algorithm until the loss function value converges to a preset value.
  • the step of obtaining a training signal includes: obtaining an original ECG signal; preprocessing the original ECG signal to generate a training signal; wherein the preprocessing includes noise processing One or more of a horizontal stretching process, a horizontal compression process, and a partial masking process.
  • the pre-processing includes noise processing; the step of pre-processing the original ECG signal to generate a training signal includes: adding and pre-processing to the original ECG signal; Set the random noise corresponding to the amplitude of the signal-to-noise ratio to obtain the training signal.
  • the pre-processing includes a transverse stretching process and / or a transverse compression process; and the step of pre-processing the original ECG signal to generate a training signal includes: A horizontal stretching process and / or a horizontal compression process are performed on the original ECG signal in proportion to obtain a training signal.
  • the pre-processing includes partial covering processing; the step of pre-processing the original ECG signal to generate a training signal includes: randomly intercepting the original ECG signal A signal having a duration that is shorter than a preset duration; the intercepted signal in the original ECG signal is set to zero to obtain a training signal.
  • one or more stable training methods among a regularization method, a random removal method of a network connection structure, and a label interference method are used.
  • the convolutional neural network includes multiple convolutional layers and multiple pooling layers; wherein each of the convolutional layers includes a convolutional unit and a non-linear transformation unit connected in sequence. And batch normalization unit; the recurrent neural network includes a plurality of long-term and short-term memory units; each of the long-term and short-term memory units includes a forget gate structure, an update gate structure, and an output gate structure.
  • the present application also provides an arrhythmia detection device, the device is disposed on the processor side of the electronic device, and the device includes: a signal acquisition module configured to acquire an ECG signal to be detected; a model input module , Configured to input the ECG signal into a pre-established arrhythmia detection model; wherein the arrhythmia detection model includes a convolutional neural network and a recurrent neural network connected in sequence; a model detection module configured to pass the The arrhythmia detection model detects the ECG signal to obtain a detection result corresponding to the ECG signal; the detection result includes an arrhythmia type.
  • the model detection module is configured as a feature extraction unit configured to perform feature extraction on the input ECG signal through the convolutional neural network to obtain the ECG signal. Feature information, and input the feature information to the recurrent neural network; a type determination unit configured to classify the input feature information through the recurrent neural network to determine an arrhythmia type of the ECG signal .
  • the feature extraction unit is configured to perform superposition convolution processing and non-linear transformation processing on the ECG signal through the convolutional neural network to obtain a characteristic time series;
  • the characteristic time series includes a plurality of characteristic segments extracted from a plurality of ECG signals arranged in time sequence.
  • the type determination unit is configured to perform iterative processing on each of the characteristic time series in the characteristic time series through the recurrent neural network to determine the characteristic time series. State, and classify the final state of the characteristic actual sequence to determine the type of arrhythmia of the ECG signal.
  • the device further includes: a training signal acquisition module configured to acquire a training signal; a loss value calculation module configured to input the training signal to an arrhythmia detection model to be trained, and A loss function value of the preset arrhythmia detection model is calculated; a training module is configured to train parameters of the arrhythmia detection model based on the loss function value through a back propagation algorithm until the loss The function value converges to a preset value.
  • the training signal acquisition module is configured to: obtain an original ECG signal; pre-process the original ECG signal to generate a training signal; wherein the pre-processing includes noise processing, One or more of a transverse stretching process, a transverse compression process, and a partial masking process.
  • the convolutional neural network includes multiple convolutional layers and multiple pooling layers; wherein each of the convolutional layers includes a convolutional unit and a non-linear transformation unit connected in sequence. And batch normalization unit; the recurrent neural network includes a plurality of long-term and short-term memory units; each of the long-term and short-term memory units includes a forget gate structure, an update gate structure, and an output gate structure.
  • an electronic device includes a memory and a processor.
  • the memory is configured to store a program that supports the processor to execute the method according to any one of the first aspect.
  • the processor is configured to: For executing a program stored in the memory.
  • a computer storage medium for storing computer software instructions used by the device according to any one of the second aspects.
  • the above-mentioned arrhythmia detection method, device, electronic device and computer storage medium input the ECG signal to be detected into a pre-established arrhythmia detection model, and obtain a detection result including the arrhythmia type through the arrhythmia detection model.
  • the arrhythmia detection model proposed in this application is fused with a convolutional neural network.
  • the recurrent neural network based on the characteristics of the recurrent neural network, it can detect ECG signals of different lengths, effectively improving the universality of arrhythmia detection.
  • FIG. 1 shows a flowchart of a method for detecting an arrhythmia provided by the present application
  • FIG. 2 shows a schematic diagram of an arrhythmia detection model provided by the present application
  • FIG. 3 shows a schematic diagram of another arrhythmia detection model provided by the present application.
  • FIG. 5 shows a flowchart of a training method for an arrhythmia detection model provided by the present application
  • FIG. 6 shows a structural block diagram of an arrhythmia detection device provided by the present application
  • FIG. 7 is a schematic structural diagram of an electronic device provided by the present application.
  • the existing automatic detection method of arrhythmia uses a convolutional neural network model.
  • the ECG signal that this convolutional neural network model can detect is a fixed length, that is, the length of the ECG signal of the input model needs to be limited because the reason is that Traditional convolutional neural network models usually rely on the output of the fully connected layer of the convolutional neural network, and the number of neurons in the fully connected layer needs to correspond to the input signal. Therefore, the convolutional neural network model is based on the limitations of the fully connected layer.
  • the length of the input signal is required, which requires a fixed segmentation of the ECG signal to obtain a fixed-length model input, which seriously affects the accurate detection of arrhythmia.
  • the convolutional neural network model when the input signal of the convolutional neural network model is limited to a waveform length, although the convolutional neural network model has a recognition performance for arrhythmia with obvious waveform anomalies, it is powerless to the arrhythmia mainly exposed to rhythm characteristics. At the same time, this method needs to mark a single waveform. The cost of the mark is too high and it does not have good applicability. If the input signal of the convolutional neural network model is amplified to the record length level, it will be limited by the problem of different signal lengths commonly existing in the ECG data set. The scale of the convolutional neural network model needs to be increased corresponding to the length of the input signal, and it also causes problems such as large computational overhead and overfitting.
  • the present application provides an arrhythmia detection method, device, electronic device, and computer storage medium, which can perform arrhythmia detection on ECG signals of different lengths (that is, variable-length ECG signals), which can effectively improve At least one of the above problems is described in detail below.
  • each step in the method is performed by a processor of an electronic device.
  • the electronic device may be a computer, a mobile phone, or a wearable heart monitoring device.
  • Methods include:
  • an ECG signal to be detected is acquired.
  • the ECG signal may be an ECG signal of a user to be detected acquired by a processor of the electronic device in real time, or may be an ECG signal stored in the specified position, which is called by the processor from the specified position.
  • the ECG signal is input to a pre-established arrhythmia detection model.
  • the arrhythmia detection model includes a convolutional neural network and a recurrent neural network connected in sequence. Referring to the schematic diagram of an arrhythmia detection model shown in FIG. 2, the structure of an arrhythmia detection model is briefly illustrated.
  • the feature of the input ECG signal can be extracted through a Convolutional Neural Network (CNN) to obtain the feature information of the ECG signal, and the feature information is input to a recurrent neural network (Recurrent Neural Networks (RNN); then classify the input feature information through a recurrent neural network to determine the type of arrhythmia of the ECG signal.
  • CNN Convolutional Neural Network
  • RNN recurrent Neural Networks
  • the above arrhythmia detection model proposed in this embodiment can fuse CNN and RNN, use CNN to enhance the original signal features through convolution operation, and reduce noise; use RNN to introduce a directional loop, which can be mainly used to process sequence data, and Able to deal with the problem of back-to-back correlation between inputs, so as to perform better classification processing.
  • the ECG signal is detected by using an arrhythmia detection model to obtain a detection result corresponding to the ECG signal; the detection result includes an arrhythmia type.
  • Arrhythmia types include various types such as sinus tachycardia, sinus bradycardia, premature beats, paroxysmal supraventricular tachycardia, paroxysmal ventricular tachycardia, atrial flutter and atrial fibrillation.
  • the signal characteristics of the ECG signal corresponding to the arrhythmia type are different, and the arrhythmia detection model can determine the corresponding arrhythmia type based on the characteristics of the ECG signal.
  • the arrhythmia detection method provided in the present application inputs an ECG signal to be detected into a pre-established arrhythmia detection model, and obtains a detection result including an arrhythmia type through the arrhythmia detection model.
  • the arrhythmia detection model proposed in this application is fused with a convolutional neural network.
  • the recurrent neural network based on the characteristics of the recurrent neural network, it can detect ECG signals of different lengths, effectively improving the universality of arrhythmia detection.
  • the above arrhythmia detection model can also be called a time step convolutional neural network model.
  • the arrhythmia detection model includes a convolutional sub-network (ie, the aforementioned convolutional neural network) and a recurrent sub-network (ie, the aforementioned recurrent sub-network).
  • the convolutional subnetwork in Figure 3 can expose the features of the signal by superposing the convolution and non-linear transformation on the basis of the input signal, and generate a time series composed of multiple feature segments, each feature Segments correspond to features extracted from a segment of ECG signals and are input into a cyclic subnetwork.
  • the cyclic subnetwork obtains the feature segment corresponding to each time point as an input by iterating over time, and updates the internally saved state of the cyclic subnetwork.
  • the end unit of the cyclic sub-network can output the classification result according to the final state including the entire signal information.
  • the convolutional neural network includes multiple convolutional layers and multiple pooling layers; the convolutional layer can The convolution operation is performed on the input signal through multiple convolution kernels, thereby generating a signal with more prominent features.
  • the pooling layer can use the maximum pooling operation, that is, taking the maximum output of the input adjacent elements to reduce the size of the data stream.
  • the convolutional neural network uses a total of 10 convolutional layers and 5 pooling layers; among the numbers after the convolutional layer, the former represents the size of the convolution kernel used by the convolutional layer, and the latter Represents the number of convolution kernels used.
  • the convolution layer 3-64 3 represents the size of the convolution kernel used by the convolution layer is 3; 64 represents the number of convolution kernels used by the convolution layer is 64.
  • This method can maintain the signal length without being changed by the convolution layer, which can further ensure the accuracy and reliability of model detection.
  • the number of convolution kernels increases leading to the increasing number of mapping features at each signal point
  • the number of signal points can be reduced through the pooling layer to maintain a stable data flow in the model and avoid information at the layer level.
  • the phenomenon of loss may occur due to the different expression ability between layers.
  • the structure of the convolutional sub-network shown in FIG. 3 is only an implementation manner. In actual applications, the structure of the convolutional sub-network can be flexibly changed according to needs, and should not be considered as a limitation.
  • each convolution layer includes a convolution unit, a non-linear transformation unit, and a batch normalization unit connected in sequence.
  • the convolution unit is used to perform a convolution operation.
  • the nonlinear transformation unit mainly introduces a linear rectification function, which is also called a modified linear unit, and is an activation function.
  • the linear rectification function as the activation function of the neuron, defines the non-linear output of the neuron after the linear transformation. That is, the linear rectification function is mainly used to add a non-linear transformation to the convolutional layer, so that the convolutional neural network has the ability to express complex features by extracting features.
  • the input information received can be normalized to keep the input of each convolutional layer similarly distributed, so that the convolution kernel parameters can be updated more stably.
  • the recurrent neural network includes a plurality of long-term and short-term memory units; in one embodiment, the recurrent neural network may include three long-term and short-term memories Network unit.
  • the short-term memory network unit can remember historical information and use it to process time series data.
  • each long-short-term memory unit includes a forget gate structure, an update gate structure, and an output gate structure.
  • the long-term and short-term memory network unit internally selects to save or update its internal state variables through three gate structures—forget gate, update gate, and output gate. Through its internal state variables, the long-term and short-term memory network unit can save the valid information in the sequence input, and continuously use the information of the earlier position in the sequence in each state update, so as to achieve the purpose of comprehensively using the entire ECG signal information .
  • the three long-term and short-term memory network units have state variables with lengths of 128, 32, and 9, respectively.
  • the long-term and short-term memory network units located at the back end continuously refine and summarize the information contained in the front-end state variables.
  • the long-term short-term memory unit outputs the state variable at the last moment as the output of the entire network structure.
  • the forget gate determines how much of the unit state (that is, the above state variables) at the previous moment is retained until the current moment; the input gate determines how much of the network input is saved to the unit state at the current moment; the long-term memory network unit uses the output gate to Control how much the state of the unit outputs to the current output value of the memory network unit.
  • the method includes:
  • Step S402 Acquire the ECG signal to be detected.
  • step S404 the ECG signal is input to a pre-established arrhythmia detection model.
  • the arrhythmia detection model includes a convolutional neural network and a recurrent neural network connected in sequence.
  • Step S406 the convolutional neural network in the arrhythmia detection model is used to superimpose convolution processing and non-linear transformation processing on the ECG signals to obtain a characteristic time series; wherein the characteristic time series includes a plurality of ECGs arranged in chronological order. The corresponding extracted feature segments in the signal.
  • Step S408 Iteratively process each feature segment in the characteristic time series through a recurrent neural network to determine the final state of the characteristic time series, and classify the final state of the characteristic time series to determine the type of arrhythmia of the ECG signal.
  • the purpose of recurrent neural networks is to process sequence data.
  • the layers are fully connected, and the nodes of each layer are unconnected. But this ordinary neural network is incapable of many problems. Because a lot of information needs to be related.
  • the current output of a sequence is also related to the previous output; the specific manifestation is that the recurrent neural network memorizes the previous information and applies it to the calculation of the current output, that is, the recurrent neural network There are connections between adjacent nodes in the layer, based on this, the recurrent neural network can theoretically process sequence data of any length.
  • the recurrent neural network in the arrhythmia detection model provided in this embodiment stores the information of the past ECG signals that have been analyzed. When calculating the results, only the newly acquired signals need to be processed, which greatly reduces the calculation amount.
  • the arrhythmia detection method provided in the present application inputs an ECG signal to be detected into a pre-established arrhythmia detection model, and obtains a detection result including an arrhythmia type through the arrhythmia detection model.
  • the arrhythmia detection model incorporates a convolutional neural network and a recurrent neural network. Based on the characteristics of the recurrent neural network, it can detect ECG signals of different lengths, effectively improving the universality of arrhythmia detection; and it can also detect historical ECG signals.
  • Based on the memory information based on the memory information, it can make the neural network training requires less calculations each time, and the required calculations and memory are significantly reduced, making it possible to deploy the method on mobile platforms. .
  • this application provides a training process of an arrhythmia detection model.
  • the method includes the following steps:
  • Step S502 Obtain a training signal.
  • the original ECG signal may be obtained first; then the original ECG signal is pre-processed to generate a training signal; wherein the pre-processing includes noise processing, lateral stretching processing, lateral compression processing, and partial masking One or more of the processes.
  • step S504 the training signal is input to the arrhythmia detection model to be trained, and the loss function value is calculated through the loss function of the arrhythmia detection model.
  • Step S506 Based on the loss function value, the parameters of the arrhythmia detection model are trained by a back propagation algorithm until the loss function value converges to a preset value.
  • this embodiment further provides an example of generating the following training signals.
  • the foregoing steps of preprocessing the original ECG signals and generating training signals can be performed by referring to one or more of the following methods:
  • a random signal corresponding to the amplitude of a preset signal-to-noise ratio can be added to the original ECG signal to obtain a training signal.
  • the reason for the noise processing of the training signal is that when the ECG signal of the person to be detected is actually collected, due to the limitation of the acquisition, the collected ECG signal necessarily contains noise (the white noise is taken as an example for description below).
  • this embodiment will detect the original ECG signal (that is, the original Pure ECG signal with noise) for noise processing, such as adding random white noise corresponding to the signal-to-noise ratio amplitude to the original ECG signal after estimating the signal-to-noise ratio of the signal, thereby generating a segment of the original ECG signal that interferes with different white noise
  • the new signal that is, the training signal
  • the original electrocardiographic signal may be subjected to lateral stretching processing and / or lateral compression processing according to a preset ratio to obtain a training signal.
  • the reason for performing lateral stretching processing and / or lateral compression processing on the training signal is that the human heart rate fluctuates within a certain range, which is determined by the individual's current physical state. In addition, heart rate varies among individuals. Therefore, by laterally stretching / compressing the signal with a certain range of factors (that is, a preset ratio), that is, by stretching / compressing the original ECG signal, generating ECG signals at various heart rates (that is, training signals) ), Training the arrhythmia detection model by using the ECG signals at various heart rates to improve the detection effect of the arrhythmia detection model on the ECG signals at various heart rates, can make the arrhythmia detection model more focused on the waveform characteristics and Global characteristics such as heart rate variability that are not affected by heart rate.
  • a certain range of factors that is, a preset ratio
  • a signal with a duration shorter than a preset duration can be randomly intercepted from the original ECG signal; then the intercepted signal from the original ECG signal is set to zero to obtain a training signal.
  • the reason for partially covering the training signals is that when the ECG signals of the person to be detected are actually collected, the signal strength may fluctuate or even the leads may fall off.
  • the ideal model must be able to avoid the influence of partial signal loss as much as possible, so as to autonomously extract features from valid signal segments.
  • the intercepted signal can be randomly set to zero in the original ECG signal, forming the effect of partial signal loss, such as randomly selecting a signal segment of no more than 1.5 seconds in each ECG signal and setting the new signal segment The number segment is set to zero, thus generating various signals of lead loss.
  • the detection effect of the arrhythmia detection model on the partially missing ECG signal can be improved, and the arrhythmia detection model can be avoided to focus only on the significant part of the ECG signal
  • the signal segment can help the arrhythmia detection model to extract the signal features more completely and improve the detection effect.
  • the above training signal generation method can enhance the training data of the arrhythmia detection model.
  • a new ECG signal that roughly belongs to the unified data distribution is generated.
  • This transformation is mainly some characteristics of randomness in the actual ECG signal collected by simulation.
  • the enhanced data ie, training signals
  • the arrhythmia detection model provided in this embodiment can support the end-to-end training method, that is, the convolutional neural network and the recurrent neural network can be well connected, so that the two networks can be organically integrated and promote each other.
  • one or more of a regularization method, a random removal method of a network connection structure, and a label interference method may be used.
  • Stable training method To facilitate understanding, the above training methods are described in detail below:
  • Neural network models usually have many parameters and are prone to overfitting. Based on this, when training the model in this embodiment, a regularization manner may be adopted to prevent overfitting.
  • the regularization method may adopt an L1 regularization method or an L2 regularization method.
  • the constraints on the model's internal parameters can effectively control the situation where some parameters are updated to dominate the model's direction. In practical applications, a certain percentage of the sum of the squares of all parameters can be added to the loss function during model training in order to establish a soft constraint of the internal parameters, thereby reducing the possibility of excessive model parameters and enabling the model to train stably.
  • the random removal method of the network connection structure may be to randomly remove a fixed-size part of the connection in the neural network structure. For example, in the connection of the previous convolution layer to the next convolution layer in the neural network, part of the convolution layer is filtered. The device does not work. In this way, the original network structure is destroyed, and different smaller neural network structures can be mainly trained during each training. This random connection combination can make the model avoid the strong dependence on some connections. In order to use the sorted model parameters more effectively and get a more stable and reliable model.
  • the label interference method may be that when the model is trained, an appropriate amount of random labels are introduced randomly. For example, the correct label corresponding to the training signal A is B, but during training, the training signal A is attached with a label C, and interference through this wrong label In this way, each time the model is trained, it can be based on a smaller data set, so that the trained model is approximated by averaging the models generated from multiple different sampled data sets. This averaging helps to obtain a more stable model.
  • the training method of the arrhythmia detection model provided in this embodiment helps to make the model perform stable training and obtain an arrhythmia detection model capable of obtaining accurate and reliable detection results in practical applications.
  • the real-time processing implementation of this model can also flexibly make trade-offs between accuracy and real-time performance.
  • the data distribution of the ECG signal can be kept consistent with the training signal.
  • the convolution layer of the model can be made. Fill in zeros at both ends of the data sequence corresponding to the training signal. Most of the data except the data at both ends can be pure data, and no zeroing operation is required.
  • the model can be allowed to fill in zeros at both ends of the data sequence corresponding to the training signal in all convolution layer calculations, so that the convolution layer no longer affects the data length required for word prediction updates.
  • the ECG signal is actually a combination of multiple digits, and a series of ECG signals is actually a sequence of numbers.
  • the purpose of the above zero padding is to help the convolution operation and prevent the signal length from being affected by the convolution operation.
  • the above-mentioned arrhythmia detection method uses an arrhythmia detection model (also known as a time-step convolutional neural network), which integrates a convolutional neural network and a recurrent neural network, and is applicable. Based on ECG signals of different lengths (that is, compatible with different variable-length ECG signals), features in the ECG signals are automatically extracted and integrated.
  • the arrhythmia detection model can be used for real-time processing. Because the recurrent neural network can be used to store the information of past ECG signals, each time you calculate the results, you only need to process the newly obtained signals, compared to other The neural network model needs to extract and classify the entire signal from scratch.
  • the arrhythmia detection model provided in this embodiment can greatly reduce the calculation amount, can support incremental real-time processing, and can be used in real-time processing environments. Compared with traditional convolutional neural networks, for a network model, the amount of memory required by the model provided in this embodiment is significantly reduced, making it possible to deploy the method on a mobile platform.
  • the above arrhythmia detection model can fully and comprehensively extract local features (such as a single waveform in the ECG signal) and global features (such as heart rate and heart rate variability) from the variable-length ECG signal due to its superior detection performance It has the advantages of small memory overhead and low calculation overhead in real-time processing. It has good applicability in the application areas of automatic analysis of ECG signals such as hospital monitoring and patient self-monitoring.
  • the present application provides the following experimental data.
  • the present application has achieved an average accuracy rate of up to 77.3%.
  • the time-stepped convolutional neural network that is, the arrhythmia detection model proposed in this application
  • Table 1 Arrhythmia classification performance of traditional convolutional neural network and time step convolutional neural network
  • the parameters of the time-stepped convolutional neural network model proposed in the present application can be obtained according to calculations.
  • the amount of parameters of the time-step convolutional network model is about half that of the VGG16 model.
  • the specific estimation process is shown in Table 2 below.
  • the time step convolution network model proposed in this application greatly reduces the amount of parameters required through the convolution layer and the parameter reuse in the short-term and long-term memory network.
  • the method for detecting arrhythmia has the following obvious advantages by using a time step convolutional neural network:
  • this embodiment further provides an arrhythmia detection device, which can be disposed on the processor side of an electronic device. See the structure of an arrhythmia detection device shown in FIG. 6 Block diagram of the device including:
  • the signal acquisition module 602 is configured to acquire an ECG signal to be detected.
  • the model input module 604 is configured to input the ECG signal to a pre-established arrhythmia detection model; wherein the arrhythmia detection model includes a convolutional neural network and a recurrent neural network connected in sequence.
  • the convolutional neural network may include multiple convolutional layers and multiple pooling layers; wherein each convolutional layer includes a convolutional unit, a non-linear transformation unit, and a batch normalization unit connected in sequence; a loop
  • the neural network may include multiple long-term and short-term memory units; each long-term and short-term memory unit includes a forget gate structure, an update gate structure, and an output gate structure.
  • the model detection module 606 is configured to detect the ECG signal by using an arrhythmia detection model to obtain a detection result corresponding to the ECG signal; the detection result includes an arrhythmia type.
  • the arrhythmia detection device inputs an ECG signal to be detected into a pre-established arrhythmia detection model, and obtains a detection result including an arrhythmia type through the arrhythmia detection model.
  • the arrhythmia detection model proposed in this application is fused with a convolutional neural network.
  • the recurrent neural network based on the characteristics of the recurrent neural network, it can detect ECG signals of different lengths, effectively improving the universality of arrhythmia detection.
  • the above model detection module includes a feature extraction unit and a type determination unit: wherein,
  • a feature extraction unit configured to perform feature extraction on an input ECG signal through a convolutional neural network to obtain feature information of the ECG signal and input the feature information to a recurrent neural network;
  • the type determination unit is configured to perform classification processing on the input characteristic information through a recurrent neural network to determine an arrhythmia type of the ECG signal.
  • the above feature extraction unit may be further configured to perform a convolution processing and a non-linear transformation process on the ECG signal through a convolutional neural network to obtain a characteristic time series; wherein, the characteristic time series includes a plurality of segments arranged in chronological order from multiple segments. Feature segments extracted from the ECG signal.
  • the above-mentioned type determination unit may be further configured to iteratively process each feature segment in the feature time series through a recurrent neural network to determine the state of the feature time series, and perform classification processing on the final state of the feature actual sequence to determine the ECG The type of arrhythmia of the signal.
  • this embodiment also provides a training method for an arrhythmia detection model. Based on this, the above device further includes a training signal acquisition module, a loss value calculation module, and a training module connected in sequence;
  • a training signal acquisition module configured to acquire a training signal
  • a loss value calculation module configured to input a training signal to an arrhythmia detection model to be trained, and calculate a loss function value through a loss function of a preset arrhythmia detection model
  • the training module is configured to train the parameters of the arrhythmia detection model through a back propagation algorithm based on the loss function value until the loss function value converges to a preset value.
  • the above training signal acquisition module may be further configured to: obtain the original ECG signal; pre-process the original ECG signal to generate a training signal; wherein the pre-processing includes noise processing, lateral stretching processing, lateral compression processing, and partial covering processing One or more.
  • random noise corresponding to the amplitude of a preset signal-to-noise ratio may be added to the original ECG signal to obtain a training signal.
  • the original electrocardiographic signal may be subjected to lateral stretching processing and / or lateral compression processing according to a preset ratio to obtain a training signal.
  • a signal with a duration shorter than a preset duration can be randomly intercepted from the original ECG signal; the intercepted signal from the original ECG signal is set to zero to obtain training signal.
  • the electronic device includes a memory and a processor.
  • the memory is configured to store a program that supports the processor to execute any of the foregoing arrhythmia detection methods.
  • the processor is configured to execute the storage in the memory. program of.
  • the electronic device may be a computer, a mobile phone, or a wearable heart monitoring device.
  • this embodiment also provides a computer storage medium configured to store computer software instructions used by any of the foregoing arrhythmia detection devices.
  • the electronic device includes: a processor 70, a memory 71, a bus 72, and a communication interface 73.
  • the processor 70 and the communication interface 73 and the memory 71 are connected through a bus 72; the processor 70 is configured to execute an executable module stored in the memory 71, such as a computer program.
  • the memory 71 may include a high-speed random access memory (RAM, Random Access Memory), and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.
  • RAM Random Access Memory
  • non-volatile memory such as at least one disk memory.
  • the communication connection between the network element of the system and at least one other network element is realized through at least one communication interface 73 (which can be wired or wireless), and the Internet, a wide area network, a local network, a metropolitan area network, and the like can be used.
  • the bus 72 may be an ISA bus, a PCI bus, an EISA bus, or the like.
  • the bus can be divided into an address bus, a data bus, a control bus, and the like. For ease of representation, only a two-way arrow is used in FIG. 7, but it does not mean that there is only one bus or one type of bus.
  • the memory 71 is configured to store a program, and the processor 70 executes the program after receiving an execution instruction.
  • the method performed by the apparatus defined by the flow process disclosed in any one of the embodiments of the present application may be applied to the processor 70 Medium, or implemented by the processor 70.
  • the processor 70 may be an integrated circuit chip and has a signal processing capability. In the implementation process, each step of the foregoing method may be completed by an integrated logic circuit of hardware in the processor 70 or an instruction in a form of software.
  • the above-mentioned processor 70 may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc .; it may also be a Digital Signal Processor (DSP) ), Application Specific Integrated Circuit (ASIC), Field-Programmable Gate Array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.
  • DSP Digital Signal Processor
  • ASIC Application Specific Integrated Circuit
  • FPGA Field-Programmable Gate Array
  • a general-purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
  • the steps combined with the method disclosed in this application can be directly embodied as being executed by a hardware decoding processor, or executed and completed by using a combination of hardware and software modules in the decoding processor.
  • the software module may be located in a mature storage medium such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, and the like.
  • the storage medium is located in the memory 71, and the processor 70 reads the information in the memory 71 and completes the steps of the foregoing method in combination with its hardware.
  • the arrhythmia detection method, device, electronic device and computer program product provided by the present application include a computer-readable storage medium storing program code, and the program code includes instructions that can be used to execute the foregoing method embodiments For specific implementation of the method described in the method embodiment, details are not described herein.
  • the terms “installation”, “connected”, and “connected” should be understood in a broad sense, for example, they may be fixed connections, detachable connections, or Integrated connection; it can be mechanical or electrical connection; it can be directly connected, or it can be indirectly connected through an intermediate medium, or it can be the internal connection of two elements.
  • installation should be understood in a broad sense, for example, they may be fixed connections, detachable connections, or Integrated connection; it can be mechanical or electrical connection; it can be directly connected, or it can be indirectly connected through an intermediate medium, or it can be the internal connection of two elements.
  • the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium.
  • the technical solution of this application is essentially a part that contributes to the existing technology or a part of the technical solution can be embodied in the form of a software product.
  • the computer software product is stored in a storage medium, including Several instructions are used to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in the embodiments of the present application.
  • the foregoing storage media include: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disks or optical disks and other media that can store program codes .
  • an arrhythmia detection model including a convolutional neural network and a recurrent neural network connected in sequence can detect arrhythmia of ECG signals of different lengths, effectively improving the universality of arrhythmia detection .

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Biomedical Technology (AREA)
  • Artificial Intelligence (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Mathematical Physics (AREA)
  • Cardiology (AREA)
  • Public Health (AREA)
  • Medical Informatics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computing Systems (AREA)
  • Computational Linguistics (AREA)
  • Software Systems (AREA)
  • Pathology (AREA)
  • Surgery (AREA)
  • Animal Behavior & Ethology (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Veterinary Medicine (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Signal Processing (AREA)
  • Physiology (AREA)
  • Psychiatry (AREA)
  • Fuzzy Systems (AREA)
  • Databases & Information Systems (AREA)
  • Epidemiology (AREA)
  • Primary Health Care (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
PCT/CN2018/104002 2018-09-04 2018-09-04 心律失常的检测方法、装置、电子设备及计算机存储介质 Ceased WO2020047750A1 (zh)

Priority Applications (5)

Application Number Priority Date Filing Date Title
PCT/CN2018/104002 WO2020047750A1 (zh) 2018-09-04 2018-09-04 心律失常的检测方法、装置、电子设备及计算机存储介质
CN201880001770.4A CN111163690B (zh) 2018-09-04 2018-09-04 心律失常的检测方法、装置、电子设备及计算机存储介质
EP18932602.8A EP3847958A4 (en) 2018-09-04 2018-09-04 ARRHYTHMIA DETECTION METHOD AND DEVICE, ELECTRONIC DEVICE AND COMPUTER STORAGE MEDIUM
JP2020568775A JP7304901B2 (ja) 2018-09-04 2018-09-04 不整脈検出方法、装置、電子装置およびコンピュータ記憶媒体
KR1020207035714A KR102463764B1 (ko) 2018-09-04 2018-09-04 부정맥 검출 방법, 장치, 전자장치 및 컴퓨터 기억 매체

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2018/104002 WO2020047750A1 (zh) 2018-09-04 2018-09-04 心律失常的检测方法、装置、电子设备及计算机存储介质

Publications (1)

Publication Number Publication Date
WO2020047750A1 true WO2020047750A1 (zh) 2020-03-12

Family

ID=69721489

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/104002 Ceased WO2020047750A1 (zh) 2018-09-04 2018-09-04 心律失常的检测方法、装置、电子设备及计算机存储介质

Country Status (5)

Country Link
EP (1) EP3847958A4 (https=)
JP (1) JP7304901B2 (https=)
KR (1) KR102463764B1 (https=)
CN (1) CN111163690B (https=)
WO (1) WO2020047750A1 (https=)

Cited By (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111588349A (zh) * 2020-05-28 2020-08-28 京东方科技集团股份有限公司 一种健康分析装置及电子设备
CN112001482A (zh) * 2020-08-14 2020-11-27 佳都新太科技股份有限公司 振动预测及模型训练方法、装置、计算机设备和存储介质
CN112022142A (zh) * 2020-08-07 2020-12-04 上海联影智能医疗科技有限公司 心电信号类型识别方法、装置及介质
CN112070067A (zh) * 2020-10-12 2020-12-11 乐普（北京）医疗器械股份有限公司 一种光体积描计信号的散点图分类方法和装置
CN112464721A (zh) * 2020-10-28 2021-03-09 中国石油天然气集团有限公司 微地震事件自动识别方法及装置
CN112818773A (zh) * 2021-01-19 2021-05-18 青岛歌尔智能传感器有限公司 心率检测方法、设备及存储介质
CN113768514A (zh) * 2021-08-09 2021-12-10 西安理工大学 基于卷积神经网络与门控循环单元的心律失常分类方法
CN114359625A (zh) * 2021-12-13 2022-04-15 重庆邮电大学 一种基于二维图像的深度学习心率失常分类方法
US20220125376A1 (en) * 2019-02-13 2022-04-28 Kyoto University Sleep apnea syndrome determination apparatus, sleep apnea syndrome determination method, and sleep apnea syndrome determination program
CN114587375A (zh) * 2022-03-28 2022-06-07 联通(广东)产业互联网有限公司 心电图的预处理方法、关键波段提取方法、设备和介质
WO2022244291A1 (ja) * 2021-05-21 2022-11-24 株式会社カルディオインテリジェンス プログラム、出力装置及びデータ処理方法
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
CN115624322A (zh) * 2022-11-17 2023-01-20 北京科技大学 一种基于高效时空建模的非接触生理信号检测方法及系统
CN116942175A (zh) * 2022-04-12 2023-10-27 广州视源电子科技股份有限公司 用于心电信号的特征波检测方法、装置、设备及存储介质
CN117257324A (zh) * 2023-11-22 2023-12-22 齐鲁工业大学(山东省科学院) 基于卷积神经网络和ecg信号的房颤检测方法
CN118845033A (zh) * 2024-07-12 2024-10-29 重庆理工大学 基于改进的lstm增强心电图（ecg）分类方法

Families Citing this family (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112450942B (zh) * 2020-11-26 2023-01-24 中国人民解放军南部战区总医院 心电信号的监测方法、系统、装置及介质
CN114692667B (zh) * 2020-12-30 2025-06-10 华为技术有限公司 一种模型训练方法及相关装置
CN112597986B (zh) * 2021-03-05 2021-06-08 腾讯科技（深圳）有限公司 生理电信号分类处理方法、装置、计算机设备和存储介质
CN115316996B (zh) * 2021-05-10 2024-10-18 广州视源电子科技股份有限公司 心律异常识别模型训练方法、装置、设备及存储介质
KR102573059B1 (ko) * 2021-05-13 2023-08-31 경북대학교 산학협력단 부정맥 판단 방법 및 장치, 그리고 이를 구현하기 위한 프로그램이 기록된 기록매체
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
CN114343665B (zh) * 2021-12-31 2022-11-25 贵州省人民医院 一种基于图卷积空时特征融合选择的心律失常识别方法
EP4542573A4 (en) * 2022-07-22 2025-05-28 Medical AI Co., Ltd. METHOD, PROGRAM AND APPARATUS FOR PREDICTING HEALTH STATE USING ELECTROCARDIOGRAM
KR102549010B1 (ko) * 2022-08-30 2023-06-28 주식회사 휴이노 복합 인공 신경망을 이용하여 부정맥을 추정하기 위한 방법, 시스템 및 비일시성의 컴퓨터 판독 가능한 기록 매체
CN115708684A (zh) * 2022-10-24 2023-02-24 卫软(江苏)科技有限公司 一种基于心电信息激活的心电监测方法及装置

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106725426A (zh) * 2016-12-14 2017-05-31 深圳先进技术研究院 一种心电信号分类的方法及系统
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
CN107961007A (zh) * 2018-01-05 2018-04-27 重庆邮电大学 一种结合卷积神经网络和长短时记忆网络的脑电识别方法
CN108030488A (zh) * 2017-11-30 2018-05-15 北京医拍智能科技有限公司 基于卷积神经网络的心律失常的检测系统
CN108095716A (zh) * 2017-11-21 2018-06-01 郑州鼎创智能科技有限公司 一种基于置信规则库和深度神经网络的心电信号检测方法

Family Cites Families (21)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH08221378A (ja) * 1995-02-10 1996-08-30 Ricoh Co Ltd 学習機械
US20160189730A1 (en) * 2014-12-30 2016-06-30 Iflytek Co., Ltd. Speech separation method and system
JP6678930B2 (ja) * 2015-08-31 2020-04-15 インターナショナル・ビジネス・マシーンズ・コーポレーションＩｎｔｅｒｎａｔｉｏｎａｌ Ｂｕｓｉｎｅｓｓ Ｍａｃｈｉｎｅｓ Ｃｏｒｐｏｒａｔｉｏｎ 分類モデルを学習する方法、コンピュータ・システムおよびコンピュータ・プログラム
WO2017072250A1 (en) * 2015-10-27 2017-05-04 CardioLogs Technologies An automatic method to delineate or categorize an electrocardiogram
JP6897673B2 (ja) * 2016-04-06 2021-07-07 ソニーグループ株式会社 情報処理装置、情報処理方法および情報提供方法
JP2018005773A (ja) * 2016-07-07 2018-01-11 株式会社リコー 異常判定装置及び異常判定方法
JP6945987B2 (ja) * 2016-10-28 2021-10-06 キヤノン株式会社 演算回路、その制御方法及びプログラム
EP3558101B1 (en) * 2016-12-21 2022-06-08 Emory University Methods and systems for determining abnormal cardiac activity
JP6813033B2 (ja) * 2017-01-19 2021-01-13 株式会社島津製作所 分析データ解析方法および分析データ解析装置
CN107341452B (zh) * 2017-06-20 2020-07-14 东北电力大学 基于四元数时空卷积神经网络的人体行为识别方法
CN107562784A (zh) * 2017-07-25 2018-01-09 同济大学 基于ResLCNN模型的短文本分类方法
CN107516075B (zh) * 2017-08-03 2020-10-09 安徽华米智能科技有限公司 心电信号的检测方法、装置及电子设备
CN107943525A (zh) * 2017-11-17 2018-04-20 魏茨怡 一种基于循环神经网络的手机app交互方式
CN107958044A (zh) * 2017-11-24 2018-04-24 清华大学 基于深度时空记忆网络的高维序列数据预测方法和系统
GB201720059D0 (en) * 2017-12-01 2018-01-17 Ucb Biopharma Sprl Three-dimensional medical image analysis method and system for identification of vertebral fractures
CN108039203A (zh) * 2017-12-04 2018-05-15 北京医拍智能科技有限公司 基于深度神经网络的心律失常的检测系统
CN107870306A (zh) * 2017-12-11 2018-04-03 重庆邮电大学 一种基于深度神经网络下的锂电池荷电状态预测算法
CN108073704B (zh) * 2017-12-18 2020-07-14 清华大学 一种liwc词表扩展方法
CN108108768B (zh) * 2017-12-29 2020-09-25 清华大学 基于卷积神经网络的光伏玻璃缺陷分类方法及装置
CN108418792B (zh) * 2018-01-29 2020-12-22 华北电力大学 基于深度循环神经网络的网络逃避行为检测方法
CN108255656B (zh) * 2018-02-28 2020-12-22 湖州师范学院 一种应用于间歇过程的故障检测方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106725426A (zh) * 2016-12-14 2017-05-31 深圳先进技术研究院 一种心电信号分类的方法及系统
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
CN108095716A (zh) * 2017-11-21 2018-06-01 郑州鼎创智能科技有限公司 一种基于置信规则库和深度神经网络的心电信号检测方法
CN108030488A (zh) * 2017-11-30 2018-05-15 北京医拍智能科技有限公司 基于卷积神经网络的心律失常的检测系统
CN107961007A (zh) * 2018-01-05 2018-04-27 重庆邮电大学 一种结合卷积神经网络和长短时记忆网络的脑电识别方法

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP3847958A4 *

Cited By (26)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20220125376A1 (en) * 2019-02-13 2022-04-28 Kyoto University Sleep apnea syndrome determination apparatus, sleep apnea syndrome determination method, and sleep apnea syndrome determination program
US12310743B2 (en) * 2019-02-13 2025-05-27 Kyoto University Sleep apnea syndrome determination apparatus, sleep apnea syndrome determination method, and sleep apnea syndrome determination program
CN111588349A (zh) * 2020-05-28 2020-08-28 京东方科技集团股份有限公司 一种健康分析装置及电子设备
CN111588349B (zh) * 2020-05-28 2023-12-01 京东方科技集团股份有限公司 一种健康分析装置及电子设备
CN112022142A (zh) * 2020-08-07 2020-12-04 上海联影智能医疗科技有限公司 心电信号类型识别方法、装置及介质
CN112022142B (zh) * 2020-08-07 2023-10-17 上海联影智能医疗科技有限公司 心电信号类型识别方法、装置及介质
CN112001482B (zh) * 2020-08-14 2024-05-24 佳都科技集团股份有限公司 振动预测及模型训练方法、装置、计算机设备和存储介质
CN112001482A (zh) * 2020-08-14 2020-11-27 佳都新太科技股份有限公司 振动预测及模型训练方法、装置、计算机设备和存储介质
CN112070067A (zh) * 2020-10-12 2020-12-11 乐普（北京）医疗器械股份有限公司 一种光体积描计信号的散点图分类方法和装置
CN112070067B (zh) * 2020-10-12 2023-11-21 乐普（北京）医疗器械股份有限公司 一种光体积描计信号的散点图分类方法和装置
CN112464721A (zh) * 2020-10-28 2021-03-09 中国石油天然气集团有限公司 微地震事件自动识别方法及装置
CN112818773A (zh) * 2021-01-19 2021-05-18 青岛歌尔智能传感器有限公司 心率检测方法、设备及存储介质
WO2022244291A1 (ja) * 2021-05-21 2022-11-24 株式会社カルディオインテリジェンス プログラム、出力装置及びデータ処理方法
JP7254399B1 (ja) * 2021-05-21 2023-04-10 株式会社カルディオインテリジェンス プログラム、出力装置及びデータ処理方法
CN113768514A (zh) * 2021-08-09 2021-12-10 西安理工大学 基于卷积神经网络与门控循环单元的心律失常分类方法
CN113768514B (zh) * 2021-08-09 2024-03-22 西安理工大学 基于卷积神经网络与门控循环单元的心律失常分类方法
CN114359625A (zh) * 2021-12-13 2022-04-15 重庆邮电大学 一种基于二维图像的深度学习心率失常分类方法
CN114587375A (zh) * 2022-03-28 2022-06-07 联通(广东)产业互联网有限公司 心电图的预处理方法、关键波段提取方法、设备和介质
CN116942175A (zh) * 2022-04-12 2023-10-27 广州视源电子科技股份有限公司 用于心电信号的特征波检测方法、装置、设备及存储介质
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
CN115624322B (zh) * 2022-11-17 2023-04-25 北京科技大学 一种基于高效时空建模的非接触生理信号检测方法及系统
CN115624322A (zh) * 2022-11-17 2023-01-20 北京科技大学 一种基于高效时空建模的非接触生理信号检测方法及系统
CN117257324A (zh) * 2023-11-22 2023-12-22 齐鲁工业大学(山东省科学院) 基于卷积神经网络和ecg信号的房颤检测方法
CN117257324B (zh) * 2023-11-22 2024-01-30 齐鲁工业大学(山东省科学院) 基于卷积神经网络和ecg信号的房颤检测方法
CN118845033A (zh) * 2024-07-12 2024-10-29 重庆理工大学 基于改进的lstm增强心电图（ecg）分类方法

Also Published As

Publication number Publication date
JP7304901B2 (ja) 2023-07-07
KR20210037614A (ko) 2021-04-06
CN111163690A (zh) 2020-05-15
CN111163690B (zh) 2023-05-23
EP3847958A1 (en) 2021-07-14
EP3847958A4 (en) 2021-09-08
KR102463764B1 (ko) 2022-11-03
JP2021526063A (ja) 2021-09-30

Similar Documents

Publication Publication Date Title
WO2020047750A1 (zh) 心律失常的检测方法、装置、电子设备及计算机存储介质
Sivapalan et al. ANNet: A lightweight neural network for ECG anomaly detection in IoT edge sensors
US11727279B2 (en) Method and apparatus for performing anomaly detection using neural network
CN111772619B (zh) 一种基于深度学习的心搏识别方法、终端设备及存储介质
CN107516075B (zh) 心电信号的检测方法、装置及电子设备
WO2021017313A1 (zh) 房颤的检测方法、装置、计算机设备及存储介质
WO2022052358A1 (zh) 心电异常检测网络训练方法、心电异常预警方法及装置
CN108113666A (zh) 房颤信号的识别方法、装置和设备
CN106815570B (zh) 一种基于动态模式识别的心电信号st-t段识别方法
CN108968941B (zh) 一种心律失常检测方法、装置及终端
JP6583892B2 (ja) Ｅｃｇ信号の並行解析装置、方法及び移動端末
CN111956207B (zh) 一种心电记录标注方法、装置、设备及存储介质
CN109009084A (zh) 多导联心电信号的qrs波群校验方法、装置及设备、介质
CN110638430A (zh) 多任务级联神经网络ecg信号心律失常疾病分类模型和方法
CN109044348A (zh) 房颤检测装置及存储介质
CN114224355A (zh) 心电信号分类训练方法、分类方法、装置及存储介质
CN114781443B (zh) 一种心电信号分类方法及系统
CN109077720B (zh) 信号处理方法、装置、设备和存储介质
CN110367968B (zh) 一种右束支阻滞检测方法、装置、设备及存储介质
CN115299900A (zh) 基于集成学习的胸位血压估计方法和装置
CN116172572A (zh) 一种基于shap值加权和分段hdbscan的心拍聚类方法及系统
CN114519370A (zh) 一种基于深度学习早停机制的左心室肥大检测方法及系统
CN109044338B (zh) 房颤检测装置及存储介质
CN114451898A (zh) 短时训练卷积神经网络的心电信号分类方法
CN107397547A (zh) 一种ecg典型心拍模板生成方法及系统

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: 18932602

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2020568775

Country of ref document: JP

Kind code of ref document: A

NENP Non-entry into the national phase

Ref country code: DE

ENP Entry into the national phase

Ref document number: 2018932602

Country of ref document: EP

Effective date: 20210406