WO2021179712A1 - 心电信号分类模型的训练方法、心电信号分类方法及装置 - Google Patents

心电信号分类模型的训练方法、心电信号分类方法及装置 Download PDF

Info

Publication number
WO2021179712A1
WO2021179712A1 PCT/CN2020/135083 CN2020135083W WO2021179712A1 WO 2021179712 A1 WO2021179712 A1 WO 2021179712A1 CN 2020135083 W CN2020135083 W CN 2020135083W WO 2021179712 A1 WO2021179712 A1 WO 2021179712A1
Authority
WO
WIPO (PCT)
Prior art keywords
ecg signal
abnormal
classification model
label
trained
Prior art date
Application number
PCT/CN2020/135083
Other languages
English (en)
French (fr)
Inventor
张楠
王健宗
瞿晓阳
Original Assignee
平安科技（深圳）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 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Publication of WO2021179712A1 publication Critical patent/WO2021179712A1/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
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235—Details of waveform analysis
    • 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

Definitions

  • This application relates to the field of artificial intelligence, and specifically, to a training method of an ECG signal classification model, an ECG signal classification method, device, computer readable storage medium, and electronic equipment.
  • the ECG signal is one of the earliest biosignals researched and applied to medical clinics. It is easier to detect than other bioelectric signals and has a more intuitive regularity. Therefore, ECG signal analysis has become a diagnostic tool for doctors to diagnose heart diseases. Important means.
  • abnormal events in the ECG signal are very important for the analysis of the ECG signal.
  • the abnormal events in the ECG signal include left atrial hypertrophy, right ventricular hypertrophy, biventricular hypertrophy, anterior myocardial infarction, and posterior wall. Myocardial infarction, etc.
  • the prior art lacks relevant solutions to classify abnormal events in the ECG signal.
  • the embodiments of the present application provide an ECG signal classification model training method, ECG signal classification method, device, medium, and electronic equipment, and furthermore, at least to a certain extent, the ECG signal classification model obtained by training can be used to realize abnormalities. Effective classification of ECG signals.
  • a training method for an ECG signal classification model including: constructing a training sample set, each training sample in the training sample set contains an ECG signal sample and a plurality of abnormal label tags,
  • the multiple abnormal labeling labels are labels for labeling multiple abnormal categories in the ECG signal sample; the feature vector of the ECG signal sample and the heart signal are obtained through the neural network of the ECG signal classification model to be trained.
  • An abnormal label vector of an electrical signal sample generating a label correlation matrix according to the abnormal label vector, and generating an abnormal label probability distribution vector according to the feature vector and the label correlation matrix; according to the abnormal label probability distribution vector, and
  • the multiple abnormal label tags are trained on the ECG signal classification model to be trained to obtain the ECG signal classification model after training.
  • an ECG signal classification method is provided.
  • the ECG signal classification model performs classification prediction on the signal features to obtain the abnormal category corresponding to the ECG signal to be classified; wherein, the ECG signal classification model is trained by the ECG signal classification model provided in the above embodiment Method is obtained through training.
  • an ECG signal classification training device including: a construction unit configured to construct a training sample set, each training sample in the training sample set contains an ECG signal sample and a plurality of An abnormality labeling label, where the multiple abnormality labeling labels are labels for labeling a plurality of abnormal categories in the ECG signal sample; the obtaining unit is configured to obtain the ECG signal through a neural network of the ECG signal classification model to be trained The feature vector of the signal sample and the abnormal label vector of the ECG signal sample; a generating unit configured to generate a label correlation matrix according to the abnormal label vector, and generate an abnormality according to the feature vector and the label correlation matrix Label probability distribution vector; a training unit configured to train the ECG signal classification model to be trained according to the abnormal label probability distribution vector and the multiple abnormal label labels to obtain a trained ECG signal classification model.
  • an ECG signal classification device including: an extracting unit configured to perform feature extraction on an ECG signal to be classified using an ECG signal classification model to obtain the ECG signal to be classified Signal feature; a classification unit configured to classify and predict the signal feature through the ECG signal classification model to obtain the abnormal category corresponding to the ECG signal to be classified; wherein the ECG signal classification model is implemented as described above
  • the training method of the ECG signal classification model provided in the example is obtained by training.
  • a training sample includes an ECG signal sample and a plurality of abnormal labeling labels, and the plurality of abnormal labeling labels are labels for labeling a plurality of abnormal categories in the ECG signal sample; through the nerve of the ECG signal classification model to be trained
  • the network obtains the characteristic vector of the ECG signal sample and the abnormal label vector of the ECG signal sample; generates a label correlation matrix according to the abnormal label vector, and generates a label correlation matrix according to the characteristic vector and the label correlation matrix Abnormal label probability distribution vector; according to the abnormal label probability distribution vector and the multiple abnormal label labels, the ECG signal classification model to be trained is trained to obtain a trained ECG signal classification model.
  • a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the following method is implemented: the ECG signal to be classified through the ECG signal classification model Perform feature extraction to obtain the signal feature of the ECG signal to be classified; classify and predict the signal feature through the ECG signal classification model to obtain the abnormal category corresponding to the ECG signal to be classified; wherein, the The ECG signal classification model is obtained by training the above-mentioned ECG signal classification model training method.
  • an electronic device which includes: a processor and a memory; wherein the memory is used to store executable instructions of the processor; the processor is configured to execute the Executable instructions are executed to execute the following method: constructing a training sample set, each training sample in the training sample set contains an ECG signal sample and a plurality of abnormal labeling labels, and the multiple abnormal labeling labels are Labels labeled with multiple abnormal categories; obtain the feature vector of the ECG signal sample and the abnormal label vector of the ECG signal sample through the neural network of the ECG signal classification model to be trained; generate the label according to the abnormal label vector A correlation matrix, and generate an abnormal label probability distribution vector according to the feature vector and the label correlation matrix; according to the abnormal label probability distribution vector and the multiple abnormal labeling labels, the ECG signal to be trained The classification model is trained, and the trained ECG signal classification model is obtained.
  • an electronic device which includes: a processor and a memory; wherein the memory is configured to store executable instructions of the processor; and the processor is configured to execute The executable instruction executes the following method: extracting features of the ECG signal to be classified through the ECG signal classification model to obtain the signal features of the ECG signal to be classified; performing the signal features on the ECG signal classification model The classification prediction obtains the abnormal category corresponding to the ECG signal to be classified; wherein the ECG signal classification model is obtained by training the ECG signal classification model described above.
  • the technical solution of the present application makes full use of the abnormal label vectors in the ECG signal samples for model training, which improves the prediction accuracy of the ECG signal classification model obtained by training, so that the ECG signal classification model can be used to analyze the ECG signal The abnormal events are effectively classified.
  • FIG. 1 shows a schematic diagram of an exemplary system architecture of a model training system to which the technical solutions of the embodiments of the present application can be applied;
  • Fig. 2 shows a flowchart of a training method of an ECG signal classification model according to an embodiment of the present application
  • Fig. 3 shows a flowchart of a training method of an ECG signal classification model according to an embodiment of the present application
  • Fig. 4 shows a flowchart of a training method of an ECG signal classification model according to an embodiment of the present application
  • Fig. 5 shows a flowchart of a training method of an ECG signal classification model according to an embodiment of the present application
  • Fig. 6 shows a flowchart of a training method of an ECG signal classification model according to an embodiment of the present application
  • Fig. 7 shows a flowchart of an ECG signal classification method according to an embodiment of the present application.
  • Fig. 8 shows a block diagram of a training device for an ECG signal classification model according to an embodiment of the present application
  • Fig. 9 shows a block diagram of an ECG signal classification device according to an embodiment of the present application.
  • FIG. 10 shows a schematic structural diagram of a computer system suitable for implementing an electronic device according to an embodiment of the present application.
  • first, second, third, etc. may be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another. For example, without departing from the scope of the present application, the first element may be referred to as the second element. Similarly, the second element may be referred to as the first element. As used herein, the term “and/or” encompasses any and all combinations of one or more of the associated listed items.
  • the technical solution of this application can be applied to the fields of artificial intelligence, smart city, digital medical care, blockchain and/or big data technology to realize smart medical care.
  • the data involved in this application such as samples, tags, and/or abnormal categories, can be stored in a database, or can be stored in a blockchain, such as distributed storage through a blockchain, which is not limited in this application.
  • artificial intelligence technology has been researched and applied in many fields, such as common smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, unmanned driving, autonomous driving, drones , Robotics, intelligent medical care, intelligent customer service, etc., I believe that with the development of technology, artificial intelligence technology will be applied in more fields and play more and more important values.
  • AI Artificial Intelligence
  • digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge, and use knowledge to obtain the best results.
  • artificial intelligence is a comprehensive technology of computer science, which attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can react in a similar way to human intelligence.
  • Artificial intelligence is to study the design principles and implementation methods of various intelligent machines, so that the machines have the functions of perception, reasoning and decision-making.
  • Artificial intelligence technology is a comprehensive discipline, covering a wide range of fields, including both hardware-level technology and software-level technology.
  • Basic artificial intelligence technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, and mechatronics.
  • Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology, and machine learning/deep learning.
  • Machine Learning is a multi-field interdisciplinary subject, involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other subjects. Specializing in the study of how computers simulate or realize human learning behaviors in order to acquire new knowledge or skills, and reorganize the existing knowledge structure to continuously improve its own performance.
  • Machine learning is the core of artificial intelligence, the fundamental way to make computers intelligent, and its applications cover all fields of artificial intelligence.
  • Machine learning and deep learning usually include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and style teaching learning.
  • an embodiment of the present application proposes a method for training an ECG signal classification model, which is applied to the model training system shown in FIG. 1, please refer to FIG. 1, which is A schematic diagram of the architecture of the model training system in the embodiment of the present application.
  • the model training system includes a server and a terminal device.
  • the model training device can be deployed on a server, or can be deployed on a terminal device with higher computing power. The following will take the model training device deployed on a server as an example for introduction.
  • each training sample in the training sample set contains an ECG signal sample and multiple anomaly labeling tags.
  • the multiple anomaly labeling labels are Labels labeled with multiple abnormal categories, and then obtain the feature vector of the ECG signal sample and the abnormal label vector of the ECG signal sample through the neural network of the ECG signal classification model to be trained, and then generate the label correlation matrix according to the abnormal label vector , And generate an abnormal label probability distribution vector according to the feature vector and the label correlation matrix.
  • the ECG signal classification model to be trained is trained to obtain the trained ECG signal Classification model.
  • the server and the terminal device can communicate through a wireless network, a wired network, or a removable storage medium.
  • the above-mentioned wireless network uses standard communication technologies and/or protocols.
  • the wireless network is usually the Internet, but it can also be any network, including but not limited to Bluetooth, Local Area Network (LAN), Metropolitan Area Network (MAN), Wide Area Network (WAN), mobile, Private network or any combination of virtual private networks).
  • LAN Local Area Network
  • MAN Metropolitan Area Network
  • WAN Wide Area Network
  • mobile Private network or any combination of virtual private networks.
  • customized or dedicated data communication technologies can be used to replace or supplement the aforementioned data communication technologies.
  • the removable storage medium may be a Universal Serial Bus (USB) flash disk, a mobile hard disk, or other removable storage media.
  • USB Universal Serial Bus
  • the server in FIG. 1 may be one server or a server cluster or cloud computing center composed of multiple servers, and the details are not limited here.
  • the terminal device can be a tablet computer, a notebook computer, a palmtop computer, a mobile phone, a personal computer (PC), and a voice interaction device shown in Figure 1, or it can be a monitoring device, a face recognition device, etc., which are not done here. limited.
  • voice interaction devices include, but are not limited to, smart speakers and smart home appliances.
  • FIG. 1 Although only three terminal devices and one server are shown in FIG. 1, it should be understood that the example in FIG. 1 is only used to understand this solution, and the number of specific terminal devices and servers should be flexibly determined based on actual conditions.
  • Fig. 2 shows a flowchart of a training method of an ECG signal classification model according to an embodiment of the present application. Referring to Figure 2, the method includes:
  • Step S210 Construct a training sample set.
  • Each training sample in the training sample set contains an ECG signal sample and a plurality of abnormality labeling labels, and the plurality of abnormality labeling labels are for the plurality of abnormality categories in the ECG signal sample. Marked label;
  • Step S220 Obtain the feature vector of the ECG signal sample and the abnormal label vector of the ECG signal sample through the neural network of the ECG signal classification model to be trained;
  • Step S230 Generate a label correlation matrix according to the abnormal label vector, and generate an abnormal label probability distribution vector according to the feature vector and the label correlation matrix;
  • Step S240 Training the ECG signal classification model to be trained according to the probability distribution vector of the abnormal label and the multiple abnormal label labels to obtain a trained ECG signal classification model.
  • step S210 a training sample set is constructed, and each training sample in the training sample set includes an ECG signal sample and a plurality of abnormal labeling labels, and the plurality of abnormal labeling labels is a reference to a plurality of the ECG signal samples.
  • the label of the anomaly category is a training sample set.
  • the electrocardiogram signal is a signal of changes in electrical activity generated by the heart during each cardiac cycle recorded from the body surface, and is usually recorded in a graphical way of electrocardiograph (Electro Cardio Gram, ECG).
  • ECG Electro Cardio Gram
  • Each training sample in the training sample set includes an ECG signal sample and multiple abnormal category labeling tags. Multiple abnormal category labels are labeled as ECG The label of the abnormal category of multiple abnormal events in the signal sample.
  • ECG signal samples in each training sample can be collected in the following way: ECG signals are collected through ECG leads and sensors, and the analog signals of human physiological parameters are converted into digital signals by analog-to-digital converters. , Stored by the memory.
  • the ECG signal can be collected through 8 leads, and 2500 points are intercepted forward and 2500 points backward according to the position of the apex of the R wave at the same time, and 5000 points of data are intercepted in each lead, and then The 5000 points intercepted by each lead are spliced in the second dimension by the apex of the R wave at the same time.
  • the ECG signal of each lead is expanded from 1*5000 dimensions to 8*5000 dimensions.
  • a heart beat of the ECG signal of the lead is sampled to form a sample of the above-mentioned 8*5000 dimension. Then perform the same operation on the fixed points of the R wave tops of all ECG signal data to obtain a data set U containing (8*5000)*M-dimensional data.
  • Each sample is (8*5000) dimensional, because each sample has It is intercepted according to the position of the apex of the R wave, so M is the number of the apex of the R wave used for interception, that is, the number of ECG signal samples.
  • Each ECG signal sample is an 8*5000 8-lead heart Electrical signal data.
  • abnormal category labels can be manually labeled for the ECG signal samples.
  • the abnormal category labels include but are not limited to left atrial hypertrophy, right atrial hypertrophy, double atrial hypertrophy, left ventricular hypertrophy, and right ventricle Hypertrophy, biventricular hypertrophy, anterior myocardial infarction, posterior myocardial infarction, anterior septal myocardial infarction, lateral myocardial infarction, posterior septal myocardial infarction, sinus arrhythmia, atrial premature beats, ventricular premature beats, supraventricular tachycardia, ventricular Tachycardia, atrial flutter, atrial fibrillation, ventricular flutter, ventricular fibrillation, left bundle branch block, right bundle branch block, atrial escape, ventricular escape, tachycardia, bradycardia, atrioventricular block, etc. Wait.
  • step S220 the feature vector of the ECG signal sample and the abnormal label vector of the ECG signal sample are obtained through the neural network of the ECG signal classification model to be trained.
  • the neural network of the ECG signal classification model to be trained in this embodiment may include an input layer, a feature extraction layer, a crowdsourcing layer, and an output layer.
  • the feature extraction layer of the ECG signal classification model can be composed of any network with the feature extraction function of the ECG signal, such as the convolutional neural network (Convolutional Neural Network) of the convolutional layer, the pooling layer, and the fully connected layer. Networks, CNN).
  • Convolutional Neural Network Convolutional Neural Network
  • the ECG signal sample in order to obtain the feature vector of the ECG signal sample, can be input to the feature extraction layer, through the neural network in the feature extraction layer, such as the convolutional layer, the pooling layer, and the fully connected layer.
  • the convolutional neural network is formed, and the feature vector is obtained.
  • the dimension of the feature vector output by the feature extraction layer is the dimension of the feature vector after dimension reduction set in advance.
  • supervised learning methods For the training of the feature extraction layer, either supervised learning methods or unsupervised learning methods can be used.
  • Unsupervised learning means that in practical applications, there are a large number of unlabeled or a small number of labeled samples, from which the learning between samples Relevant to each other, unsupervised learning methods can achieve the training of the feature extraction layer by establishing optimization goals.
  • the supervised learning method is used for training, the samples and the label information in the sample set can be used as the input and expected output respectively, and the initial feature extraction layer can be trained by the machine learning method.
  • the sample data in the sample data set of the training feature extraction layer includes data from the same source as the ECG signal sample.
  • the training of the feature extraction layer can end when the preset optimization goal is reached or the preset number of iterations is completed.
  • the abnormal label vector of the ECG signal sample can also be obtained through the neural network of the ECG signal classification model to be trained.
  • a preset trained deep learning algorithm model can be used to process the ECG signal samples, and after the trained deep learning algorithm model, the abnormal label vectors of the ECG signal samples can be obtained. Record the abnormal category that appears in the ECG signal as 1, and if it does not appear, it is recorded as 0. For example, if there are 8 types of abnormal events in a certain ECG signal sample, the abnormal label vector can be expressed as [1, 0, 1, 0, 1, 0, 0, 1], the abnormal label vector indicates that this ECG signal sample contains abnormalities of category 1, category 3, category 5, and category 8.
  • the trained deep learning algorithm model in this embodiment at least includes: Convolutional Neural Network and Multilayer Perceptron (MLP); correspondingly, the process of using the trained deep learning algorithm model to process the training sample set can be It includes inputting the ECG signal sample into the convolutional neural network, extracting and obtaining the characteristic information of the ECG signal sample; inputting the characteristic information of the ECG signal sample into the multilayer perceptron, so that the multilayer perceptron is based on the preset abnormality category
  • the label performs vector dimension mapping on the feature information to obtain the abnormal category label vector of the ECG signal sample.
  • both the convolutional neural network and the multi-layer perceptron belong to relatively mature network architectures, and this embodiment does not limit their architecture composition.
  • this application uses a convolutional neural network and a multilayer perceptron to sequentially perform the foregoing steps to achieve the corresponding functions.
  • step S230 a label correlation matrix is generated according to the abnormal label vector, and an abnormal label probability distribution vector is generated according to the feature vector and the label correlation matrix.
  • the correlation between each abnormal label can be calculated.
  • the correlation between the two abnormal labels is calculated by the following model:
  • P x, y is the correlation between the abnormal label x and the abnormal label y.
  • the feature vector and the label correlation matrix can be multiplied to obtain the abnormal label probability distribution vector.
  • the abnormal probability distribution vector represents the probability of abnormal occurrence in the ECG signal sample.
  • the probability distribution vector of the abnormal label of the obtained ECG signal sample is (0.1, 0.2, 0.5, 0.1, 0.1)
  • the probability distribution vector of the abnormal label can correspond to the abnormal label A, the abnormal label B, the abnormal label C, and the abnormal label respectively.
  • Label D and abnormal label E can correspond to the abnormal label A, the abnormal label B, the abnormal label C, and the abnormal label respectively.
  • generating a label correlation matrix based on an abnormal label vector can also be implemented in the following manner, which specifically includes step S310-step S320, which are described in detail as follows:
  • Step S310 Obtain a label word vector matrix, a trainable parameter matrix, and a conditional probability matrix of the ECG signal sample according to the abnormal label vector.
  • the word2vec technology can be used to train the abnormal label vectors of the ECG signal samples in an unsupervised manner to obtain the label word vector matrix. For example, suppose that each label vector in the anomaly category label vector is regarded as a word, and there are 30 words in the anomaly category label vector. After each word is encoded, a 300-dimensional vector is obtained. Therefore, the label word vector matrix is expressed as 30*300 dimensional matrix.
  • a convolutional neural network CNN, Convolutional Neural Networks
  • CNN Convolutional Neural Networks
  • conditional probability matrix of the ECG signal sample does not need to be trained, and directly counts the conditional probability of the labels appearing in pairs in the abnormal label vector to obtain the conditional probability matrix. For example, if the label vector is [1, 1, 0], you can The obtained conditional probability matrix m is:
  • Step S320 Obtain the label correlation matrix according to the label word vector matrix of the ECG signal sample, the trainable parameter matrix, and the conditional probability matrix.
  • the label correlation matrix of the ECG signal sample can be obtained.
  • step S240 the ECG signal classification model to be trained is trained according to the abnormal label probability distribution vector and the multiple abnormal label labels to obtain a trained ECG signal classification model.
  • the server may train the ECG signal classification model to be trained according to the obtained probability distribution vector of abnormal labels and multiple abnormal label labels, so as to obtain the trained ECG signal classification model.
  • the model parameters may be updated based on the loss function to obtain the trained ECG signal classification model, as shown in FIG. 4, step S240 It can specifically include:
  • Step S410 Determine the value of the loss function according to the multiple abnormal label labels and the probability distribution vector of the abnormal label
  • Step S420 Based on the value of the loss function, update the model parameters of the ECG signal classification model to be trained to obtain a trained ECG signal classification model.
  • the server inputs each training sample into the ECG signal classification model to be trained, and each training sample contains ECG signal samples, so that the prediction results of the ECG signal samples can be obtained; based on each prediction result and multiple Anomaly label, determine the value of the loss function, based on the value of the loss function, update the model parameters of the ECG signal classification model to be trained, and obtain the ECG signal classification model after training.
  • step S410 may specifically include:
  • Step S510 Use abnormal labels corresponding to probability values greater than a first preset threshold in the abnormal probability distribution vector as multiple abnormal prediction labels of the ECG signal sample;
  • Step S520 Obtain the difference between the multiple abnormal prediction tags and the multiple abnormal annotation tags
  • Step S530 Determine the average value of the obtained sum of differences, and use the determined average value as the value of the loss function.
  • the model classification system may use abnormal labels corresponding to probability values greater than the first preset threshold in the abnormal probability distribution vector as multiple abnormal prediction labels of the ECG signal sample.
  • the probability distribution vector of the abnormal label is (0.1, 0.2, 0.3, 0.5, 0.1, 0.4, 0.1), and the probability distribution vector of the abnormal label can correspond to the abnormal label A, the abnormal label B, the abnormal label C, the abnormal label D,
  • the probability distribution vector of the abnormal label can correspond to the abnormal label A, the abnormal label B, the abnormal label C, the abnormal label D,
  • the multiple abnormal prediction labels of the ECG signal sample can be obtained as the abnormal label C, the abnormal label D, and the abnormal label F.
  • the model classification system can obtain the difference between multiple abnormal prediction labels and multiple abnormal annotation labels, determine the difference sum of each difference obtained, and use the determined difference sum as the value of the loss function.
  • the loss function may include a 0-1 loss (Zero-one Loss) function, a perceptron loss (Perceptron Loss) function, a hinge loss (Hinge Loss) function, a cross entropy loss function, a square error loss (Square Loss) function, Any one of Absolute Loss function, Exponential Loss function and regular function.
  • step S420 may specifically include:
  • Step S610 When the value of the loss function exceeds a second preset threshold, determine a corresponding error signal based on the loss function of the ECG signal classification model to be trained;
  • Step S620 Backpropagate the error signal in the ECG signal classification model to be trained, and update the model parameters of the ECG signal classification model to be trained during the propagation process to obtain the trained ECG signal Classification model.
  • the server determines that the value of the loss function exceeds the second preset threshold, the corresponding error signal is determined based on the loss function of the ECG signal classification model to be trained, the error signal is backpropagated in the ECG signal classification model to be trained, and the During the propagation process, the model parameters of the ECG signal classification model to be trained are updated to obtain the ECG signal classification model after training.
  • the training samples are input into the input layer of the neural network model, after passing through the hidden layer, finally reach the output layer and output the result.
  • This is the forward propagation process of the neural network model. Since the output result of the neural network model has an error with the actual result, the error between the output result and the actual result is calculated, and the error is propagated back from the output layer to the hidden layer until it propagates to the input layer.
  • the back propagation process In the process, adjust the value of the model parameters according to the error, and continuously iterate the above process until convergence.
  • FIG. 7 shows a flowchart of an ECG signal classification method according to an embodiment of the present application.
  • the ECG signal classification method includes:
  • Step S710 Perform feature extraction on the ECG signal to be classified using the ECG signal classification model to obtain the signal feature of the ECG signal to be classified;
  • Step S720 Perform classification prediction on the signal feature through the ECG signal classification model to obtain the abnormal category corresponding to the ECG signal to be classified.
  • the terminal may set an ECG signal classification client, and based on the ECG signal classification, the client sends a classification request with the ECG signal to be classified to the server.
  • the server parses the classification request to obtain the ECG signal to be classified, inputs the ECG signal classification model, and outputs the abnormal category corresponding to the ECG signal to be classified.
  • the server may perform feature extraction on the ECG signal to be classified through the ECG signal classification model to obtain the signal characteristics of the ECG signal to be classified; then, use the ECG signal classification model to classify and predict the signal features to obtain the ECG signal to be classified The abnormal category corresponding to the signal.
  • FIG. 8 shows a block diagram of a training device for an ECG signal classification model according to an embodiment of the present application.
  • the training device 800 for an ECG signal classification model according to an embodiment of the present application includes: The construction unit 802, the acquisition unit 804, the generation unit 806, and the training unit 808.
  • the construction unit 802 is configured to construct a training sample set, and each training sample in the training sample set includes an ECG signal sample and a plurality of abnormal labeling labels, and the multiple abnormal labeling labels are Labels labeled with multiple abnormal categories; an acquiring unit 804 configured to acquire the feature vector of the ECG signal sample and the abnormal label vector of the ECG signal sample through the neural network of the ECG signal classification model to be trained; generating unit 806, configured to generate a label correlation matrix according to the abnormal label vector, and generate an abnormal label probability distribution vector according to the feature vector and the label correlation matrix; the training unit 808 is configured to generate an abnormal label probability distribution according to the abnormal label probability distribution The vector and the multiple abnormal label tags are trained on the ECG signal classification model to be trained to obtain the ECG signal classification model after training.
  • the generating unit 806 is configured to obtain the label word vector matrix, the trainable parameter matrix, and the conditional probability matrix of the ECG signal sample according to the abnormal label vector;
  • the label word vector matrix of the electrical signal sample, the trainable parameter matrix, and the conditional probability matrix are used to obtain the label correlation matrix.
  • the training unit 808 includes: a determining subunit, configured to determine the value of the loss function according to the plurality of abnormal labeling labels and the probability distribution vector of the abnormal label; updating the subunit, configuring To update the model parameters of the ECG signal classification model to be trained based on the value of the loss function to obtain the ECG signal classification model after training.
  • the determining subunit is configured to use an abnormal label corresponding to a probability value greater than a first preset threshold in the abnormal probability distribution vector as the multiple abnormal prediction labels of the ECG signal sample Obtain the difference between the plurality of abnormal prediction tags and the plurality of abnormal annotation tags; determine the average value of the sum of the obtained differences, and use the determined average value as the value of the loss function.
  • the update subunit is configured to determine a corresponding error signal based on the loss function of the ECG signal classification model to be trained when the value of the loss function exceeds a second preset threshold;
  • the error signal is back propagated in the ECG signal classification model to be trained, and the model parameters of the ECG signal classification model to be trained are updated during the propagation process to obtain the ECG signal classification model after training.
  • Fig. 9 shows a block diagram of an ECG signal classification device according to an embodiment of the present application.
  • the ECG signal classification device 900 includes: an extraction unit 902 and a classification unit 904.
  • the extraction unit 902 is configured to perform feature extraction on the ECG signal to be classified using the ECG signal classification model to obtain the signal characteristics of the ECG signal to be classified;
  • the classification unit 904 is configured to perform feature extraction on the ECG signal to be classified using the ECG signal classification model. Signal features are classified and predicted to obtain the abnormal category corresponding to the ECG signal to be classified; wherein, the ECG signal classification model is obtained by training the ECG signal classification model training method provided in the foregoing embodiment.
  • FIG. 10 shows a schematic structural diagram of a computer system suitable for implementing an electronic device according to an embodiment of the present application.
  • the computer system 1000 includes a central processing unit (Central Processing Unit, CPU) 1001, which can be loaded into a random storage unit according to a program stored in a read-only memory (Read-Only Memory, ROM) 1002 or from a storage part 1008. Access to the program in the memory (Random Access Memory, RAM) 1003 to execute various appropriate actions and processing, for example, execute the method described in the foregoing embodiment. In RAM 1003, various programs and data required for system operation are also stored.
  • the CPU 1001, the ROM 1002, and the RAM 1003 are connected to each other through a bus 1004.
  • An input/output (Input/Output, I/O) interface 1005 is also connected to the bus 1004.
  • the following components are connected to the I/O interface 1005: input part 1006 including keyboard, mouse, etc.; output part 1007 including cathode ray tube (Cathode Ray Tube, CRT), liquid crystal display (LCD), etc., and speakers, etc. ; A storage part 1008 including a hard disk, etc.; and a communication part 1009 including a network interface card such as a LAN (Local Area Network) card and a modem.
  • the communication section 1009 performs communication processing via a network such as the Internet.
  • the driver 1010 is also connected to the I/O interface 1005 as needed.
  • a removable medium 1011 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 1010 as needed, so that the computer program read therefrom is installed into the storage portion 1008 as needed.
  • the process described above with reference to the flowchart can be implemented as a computer software program.
  • the embodiments of the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a computer program for executing the method shown in the flowchart.
  • the computer program may be downloaded and installed from the network through the communication part 1009, and/or installed from the removable medium 1011.
  • CPU central processing unit
  • the computer-readable medium shown in the embodiments of the present application may be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two.
  • the computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or a combination of any of the above.
  • Computer-readable storage media may include, but are not limited to: electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable Erasable Programmable Read Only Memory (EPROM), flash memory, optical fiber, portable compact disk read-only memory (Compact Disc Read-Only Memory, CD-ROM), optical storage device, magnetic storage device, or any suitable of the above The combination.
  • the computer-readable storage medium may be any tangible medium that contains or stores a program, and the program may be used by or in combination with an instruction execution system, apparatus, or device.
  • a computer-readable signal medium may include a data signal propagated in a baseband or as a part of a carrier wave, and a computer-readable computer program is carried therein.
  • This propagated data signal can take many forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing.
  • the computer-readable signal medium may also be any computer-readable medium other than the computer-readable storage medium.
  • the computer-readable medium may send, propagate, or transmit the program for use by or in combination with the instruction execution system, apparatus, or device .
  • the computer program contained on the computer-readable medium can be transmitted by any suitable medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.
  • each block in the flowchart or block diagram may represent a module, program segment, or part of the code, and the above-mentioned module, program segment, or part of the code includes one or more executables for realizing the specified logic function. instruction.
  • the functions marked in the block may also occur in a different order from the order marked in the drawings. For example, two blocks shown one after another can actually be executed substantially in parallel, and they can sometimes be executed in the reverse order, depending on the functions involved.
  • each block in the block diagram or flowchart, and the combination of blocks in the block diagram or flowchart can be implemented by a dedicated hardware-based system that performs the specified function or operation, or can be implemented by It is realized by a combination of dedicated hardware and computer instructions.
  • the units described in the embodiments of the present application may be implemented in software or hardware, and the described units may also be provided in a processor. Among them, the names of these units do not constitute a limitation on the unit itself under certain circumstances.
  • the present application also provides a computer-readable (storage) medium.
  • the computer-readable medium may be included in the electronic device described in the above-mentioned embodiments; it may also exist alone without being assembled into The electronic device.
  • the above-mentioned computer-readable medium carries one or more programs, and when the above-mentioned one or more programs are executed by an electronic device, the electronic device realizes the method described in the above-mentioned embodiment.
  • the storage media involved in this application may be non-volatile or volatile.
  • modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory.
  • the features and functions of two or more modules or units described above may be embodied in one module or unit.
  • the features and functions of a module or unit described above can be further divided into multiple modules or units to be embodied.
  • the example embodiments described here can be implemented by software, or can be implemented by combining software with necessary hardware. Therefore, the technical solution according to the embodiments of the present application can be embodied in the form of a software product, which can be stored in a storage medium (can be a CD-ROM, U disk, mobile hard disk, etc.) or computer readable on the network Instructions, the storage medium may be volatile or non-volatile.
  • the computer-readable instructions include several instructions to enable a computing device (which may be a personal computer, a server, a touch terminal, or a network device, etc.) to execute the method according to the embodiments of the present application.

Landscapes

  • Health & Medical Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Physics & Mathematics (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Molecular Biology (AREA)
  • Psychiatry (AREA)
  • Physiology (AREA)
  • Biophysics (AREA)
  • Pathology (AREA)
  • Biomedical Technology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Signal Processing (AREA)
  • Surgery (AREA)
  • Animal Behavior & Ethology (AREA)
  • General Health & Medical Sciences (AREA)
  • Public Health (AREA)
  • Veterinary Medicine (AREA)
  • Evolutionary Computation (AREA)
  • Fuzzy Systems (AREA)
  • Mathematical Physics (AREA)
  • Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

一种心电信号分类模型的训练方法、心电信号分类方法、装置、计算机可读存储介质及电子设备。该心电信号分类模型的训练方法包括：构建训练样本集，每个训练样本包含有心电信号样本以及多个异常标注标签(S210)；通过待训练心电信号分类模型的神经网络获取心电信号样本的特征向量以及心电信号样本的异常标签向量(S220)；根据异常标签向量生成标签相关性矩阵，并根据特征向量和标签相关性矩阵，生成异常标签概率分布向量(S230)；根据异常标签概率分布向量以及多个异常标注标签，对待训练心电信号分类模型进行训练，得到训练后的心电信号分类模型(S240)。本申请能够通过模型对心电信号中的异常事件进行有效分类。

Description

心电信号分类模型的训练方法、心电信号分类方法及装置
本申请要求于2020年10月22日提交中国专利局、申请号为202011142763.1，发明名称为“心电信号分类模型的训练方法、心电信号分类方法及装置”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及人工智能领域，具体而言，涉及一种心电信号分类模型的训练方法、心电信号分类方法、装置、计算机可读存储介质及电子设备。
背景技术
心电信号是人类最早研究并应用于医学临床的生物信号之一，它比其它生物电信号更易于检测，并且具有较直观的规律性，因而也就使得心电信号分析成为医生诊断心脏疾病的重要手段。
传统分析方法常通过提取心电信号特征对信号进行分类。发明人意识到，近年来，随着深度神经网络技术的兴起，应用深度学习方法对心电信号分类的研究也日渐增多。然而，无论是传统分析方法还是机器学习分类的方法都只是对心电信号的分类，即识别心电信号是正常信号还是异常信号。
发明人发现，事实上，心电信号中的异常事件对于心电信号分析至关重要，心电信号中的异常事件包括左心房肥大、右心室肥大、双心室肥大、前壁心肌梗死、后壁心肌梗死等等，然而现有技术中缺乏对心电信号中的异常事件分类的相关方案。
发明内容
本申请的实施例提供了一种心电信号分类模型的训练方法、心电信号分类方法、装置、介质及电子设备，进而至少在一定程度上能够通过训练得到的心电信号分类模型实现对异常心电信号的有效分类。
本申请的其他特性和优点将通过下面的详细描述变得显然，或部分地通过本申请的实践而习得。
根据本申请实施例的一个方面，提供了一种心电信号分类模型的训练方法，包括：构建训练样本集，所述训练样本集中每个训练样本包含有心电信号样本以及多个异常标注标签，所述多个异常标注标签为对所述心电信号样本中的多个异常类别标注的标签；通过待训练心电信号分类模型的神经网络获取所述心电信号样本的特征向量以及所述心电信号样本的异常标签向量；根据所述异常标签向量生成标签相关性矩阵，并根据所述特征向量和所述标签相关性矩阵，生成异常标签概率分布向量；根据所述异常标签概率分布向量以及所述多个异常标注标签，对所述待训练心电信号分类模型进行训练，得到训练后的心电信号分类模型。
根据本申请实施例的一个方面，提供了一种心电信号分类方法，包括：通过心电信号分类模型对待分类心电信号进行特征提取，得到所述待分类心电信号的信号特征；通过所述心电信号分类模型对所述信号特征进行分类预测，得到所述待分类心电信号对应的异常类别；其中，所述心电信号分类模型通过上述实施例提供的心电信号分类模型的训练方法进行训练得到。
根据本申请实施例的一个方面，提供了一种心电信号分类的训练装置，包括：构建单元，配置为构建训练样本集，所述训练样本集中每个训练样本包含有心电信号样本以及多个异常标注标签，所述多个异常标注标签为对所述心电信号样本中的多个异常类别标注的标签；获取单元，配置为通过待训练心电信号分类模型的神经网络获取所述心电信号样本的特征向量以及所述心电信号样本的异常标签向量；生成单元，配置为根据所述异常标签向量生成标签相关性矩阵，并根据所述特征向量和所述标签相关性矩阵，生成异常标签概率分布向量；训练单元，配置为根据所述异常标签概率分布向量以及所述 多个异常标注标签，对所述待训练心电信号分类模型进行训练，得到训练后的心电信号分类模型。
根据本申请实施例的一个方面，提供了一种心电信号分类装置，包括：提取单元，配置为通过心电信号分类模型对待分类心电信号进行特征提取，得到所述待分类心电信号的信号特征；分类单元，配置为通过所述心电信号分类模型对所述信号特征进行分类预测，得到所述待分类心电信号对应的异常类别；其中，所述心电信号分类模型通过上述实施例提供的心电信号分类模型的训练方法进行训练得到。
根据本申请实施例的一个方面，提供了一种计算机可读存储介质，其上存储有计算机程序，所述计算机程序被处理器执行时实现以下方法：构建训练样本集，所述训练样本集中每个训练样本包含有心电信号样本以及多个异常标注标签，所述多个异常标注标签为对所述心电信号样本中的多个异常类别标注的标签；通过待训练心电信号分类模型的神经网络获取所述心电信号样本的特征向量以及所述心电信号样本的异常标签向量；根据所述异常标签向量生成标签相关性矩阵，并根据所述特征向量和所述标签相关性矩阵，生成异常标签概率分布向量；根据所述异常标签概率分布向量以及所述多个异常标注标签，对所述待训练心电信号分类模型进行训练，得到训练后的心电信号分类模型。
根据本申请实施例的一个方面，提供了一种计算机可读存储介质，其上存储有计算机程序，所述计算机程序被处理器执行时实现以下方法：通过心电信号分类模型对待分类心电信号进行特征提取，得到所述待分类心电信号的信号特征；通过所述心电信号分类模型对所述信号特征进行分类预测，得到所述待分类心电信号对应的异常类别；其中，所述心电信号分类模型通过上述心电信号分类模型的训练方法进行训练得到。
根据本申请实施例的一个方面，提供了一种电子设备，其中，包括：处理器和存储器；其中，存储器用于存储所述处理器的可执行指令；所述处理器配置为经由执行所述可执行指令来执行以下方法：构建训练样本集，所述训练样本集中每个训练样本包含有心电信号样本以及多个异常标注标签，所述多个异常标注标签为对所述心电信号样本中的多个异常类别标注的标签；通过待训练心电信号分类模型的神经网络获取所述心电信号样本的特征向量以及所述心电信号样本的异常标签向量；根据所述异常标签向量生成标签相关性矩阵，并根据所述特征向量和所述标签相关性矩阵，生成异常标签概率分布向量；根据所述异常标签概率分布向量以及所述多个异常标注标签，对所述待训练心电信号分类模型进行训练，得到训练后的心电信号分类模型。
根据本申请实施例的一个方面，提供了一种电子设备，其中，包括：处理器和存储器；其中，存储器，用于存储所述处理器的可执行指令；所述处理器配置为经由执行所述可执行指令来执行以下方法：通过心电信号分类模型对待分类心电信号进行特征提取，得到所述待分类心电信号的信号特征；通过所述心电信号分类模型对所述信号特征进行分类预测，得到所述待分类心电信号对应的异常类别；其中，所述心电信号分类模型通过上述心电信号分类模型的训练方法进行训练得到。
本申请技术方案充分利用心电信号样本中的异常标签向量进行模型训练，提高了训练得到的心电信号分类模型的预测准确度，使得能够通过训练后的心电信号分类模型对心电信号中的异常事件进行有效分类。
应当理解的是，以上的一般描述和后文的细节描述仅是示例性和解释性的，并不能限制本申请。
附图说明
此处的附图被并入说明书中并构成本说明书的一部分，示出了符合本申请的实施例，并与说明书一起用于解释本申请的原理。显而易见地，下面描述中的附图仅仅是本申请的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以 根据这些附图获得其他的附图。在附图中：
图1示出了可以应用本申请实施例的技术方案的模型训练系统的一个示例性系统架构的示意图；
图2示出了根据本申请的一个实施例的心电信号分类模型的训练方法的流程图；
图3示出了根据本申请的一个实施例的心电信号分类模型的训练方法的流程图；
图4示出了根据本申请的一个实施例的心电信号分类模型的训练方法的流程图；
图5示出了根据本申请的一个实施例的心电信号分类模型的训练方法的流程图；
图6示出了根据本申请的一个实施例的心电信号分类模型的训练方法的流程图；
图7示出了根据本申请的一个实施例的心电信号分类方法的流程图；
图8示出了根据本申请的一个实施例的心电信号分类模型的训练装置的框图；
图9示出了根据本申请的一个实施例的心电信号分类装置的框图；
图10示出了适于用来实现本申请实施例的电子设备的计算机系统的结构示意图。
具体实施方式
现在将参考附图更全面地描述示例实施方式。然而，示例实施方式能够以多种形式实施，且不应被理解为限于在此阐述的范例；相反，提供这些实施方式使得本申请将更加全面和完整，并将示例实施方式的构思全面地传达给本领域的技术人员。
此外，所描述的特征、结构或特性可以以任何合适的方式结合在一个或更多实施例中。在下面的描述中，提供许多具体细节从而给出对本申请的实施例的充分理解。然而，本领域技术人员将意识到，可以实践本申请的技术方案而没有特定细节中的一个或更多，或者可以采用其它的方法、组元、装置、步骤等。在其它情况下，不详细示出或描述公知方法、装置、实现或者操作以避免模糊本申请的各方面。
需要说明的是，本申请的说明书和权利要求书及上述附图中使用的术语仅用于描述实施例，并不旨在限制本申请的范围。应该理解的是，术语“包括”、“包含”、“具有”等在本文中使用时指定存在所陈述的特点、整体、步骤、操作、元件、组件和/或其群组，但并不排除存在或添加其他特点、整体、步骤、操作、元件、组件和/或其群组中的一个或多个。
将进一步理解的是，尽管术语“第一”、“第二”、“第三”等可以在本文中用于描述各种元件，但是这些元件不应受这些术语的限制。这些术语仅用于区分一个元件和另一个元件。例如，在不脱离本申请的范围的情况下，第一元件可以被称为第二元件。类似地，第二元件可以被称为第一元件。如本文所使用的，术语“和/或”包含关联的列出的项目中的一个或多个的任何和所有组合。
附图中所示的方框图仅仅是功能实体，不一定必须与物理上独立的实体相对应。即，可以采用软件形式来实现这些功能实体，或在一个或多个硬件模块或集成电路中实现这些功能实体，或在不同网络和/或处理器装置和/或微控制器装置中实现这些功能实体。
附图中所示的流程图仅是示例性说明，不是必须包括所有的内容和操作/步骤，也不是必须按所描述的顺序执行。例如，有的操作/步骤还可以分解，而有的操作/步骤可以合并或部分合并，因此实际执行的顺序有可能根据实际情况改变。
应当理解，在本申请中，“至少一个(项)”是指一个或者多个，“多个”是指两个或两个以上。
本申请的技术方案可应用于人工智能、智慧城市、数字医疗、区块链和/或大数据技术领域，以实现智能医疗。可选的，本申请涉及的数据如样本、标签和/或异常类别等可存储于数据库中，或者可以存储于区块链中，比如通过区块链分布式存储，本申请不做限定。
随着人工智能技术研究和进步，人工智能技术在多个领域展开研究和应用，例如常 见的智能家居、智能穿戴设备、虚拟助理、智能音箱、智能营销、无人驾驶、自动驾驶、无人机、机器人、智能医疗、智能客服等，相信随着技术的发展，人工智能技术将在更多的领域得到应用，并发挥越来越重要的价值。
本申请实施例提供的方案涉及人工智能的机器学习等技术，具体通过如下实施例进行说明，首先对几个名词进行解释和说明：
人工智能(Artificial Intelligence，AI)是利用数字计算机或者数字计算机控制的机器模拟、延伸和扩展人的智能，感知环境、获取知识并使用知识获得最佳结果的理论、方法、技术及应用系统。换句话说，人工智能是计算机科学的一个综合技术，它企图了解智能的实质，并生产出一种新的能以人类智能相似的方式做出反应的智能机器。人工智能也就是研究各种智能机器的设计原理与实现方法，使机器具有感知、推理与决策的功能。
人工智能技术是一门综合学科，涉及领域广泛，既有硬件层面的技术也有软件层面的技术。人工智能基础技术一般包括如传感器、专用人工智能芯片、云计算、分布式存储、大数据处理技术、操作/交互系统、机电一体化等技术。人工智能软件技术主要包括计算机视觉技术、语音处理技术、自然语言处理技术以及机器学习/深度学习等几大方向。
机器学习(Machine Learning，ML)是一门多领域交叉学科，涉及概率论、统计学、逼近论、凸分析、算法复杂度理论等多门学科。专门研究计算机怎样模拟或实现人类的学习行为，以获取新的知识或技能，重新组织已有的知识结构使之不断改善自身的性能。机器学习是人工智能的核心，是使计算机具有智能的根本途径，其应用遍及人工智能的各个领域。机器学习和深度学习通常包括人工神经网络、置信网络、强化学习、迁移学习、归纳学习、式教学习等技术。
为了对心电信号中的异常事件进行分类，本申请实施例提出了一种心电信号分类模型的训练方法，该方法应用于图1所示的模型训练系统，请参阅图1，图1为本申请实施例中模型训练系统的一个架构示意图，如图所示，模型训练系统中包括服务器和终端设备。而模型训练装置可以部署于服务器，也可以部署于具有较高计算力的终端设备，下面将以模型训练装置部署于服务器为示例进行介绍。
具体的，服务器在对模型进行训练之前，可以先构建训练样本集，训练样本集中每个训练样本包含有心电信号样本以及多个异常标注标签，多个异常标注标签为对心电信号样本中的多个异常类别标注的标签，然后，通过待训练心电信号分类模型的神经网络获取心电信号样本的特征向量以及心电信号样本的异常标签向量，进而，根据异常标签向量生成标签相关性矩阵，并根据特征向量和标签相关性矩阵，生成异常标签概率分布向量，最后，根据异常标签概率分布向量以及多个异常标注标签，对待训练心电信号分类模型进行训练，得到训练后的心电信号分类模型。
服务器和终端设备之间可以通过无线网络、有线网络或可移动存储介质进行通信。其中，上述的无线网络使用标准通信技术和/或协议。无线网络通常为因特网、但也可以是任何网络，包括但不限于蓝牙、局域网(Local Area Network，LAN)、城域网(Metropolitan Area Network，MAN)、广域网(Wide Area Network，WAN)、移动、专用网络或者虚拟专用网络的任何组合)。在一些实施例中，可以使用定制或专用数据通信技术取代或者补充上述数据通信技术。可移动存储介质可以为通用串行总线(Universal Serial Bus，USB)闪存盘、移动硬盘或其他可移动存储介质等。
其中，图1中的服务器可以是一台服务器或多台服务器组成的服务器集群或云计算中心等，具体此处均不限定。终端设备可以为图1中示出的平板电脑、笔记本电脑、掌上电脑、手机、个人电脑(personal computer，PC)及语音交互设备，也可以为监控设备、人脸识别设备等，此处不做限定。其中，语音交互设备包含但不仅限于智能音响以及智 能家电。
虽然图1中仅示出了三个终端设备和一个服务器，但应当理解，图1中的示例仅用于理解本方案，具体终端设备和服务器的数量均应当结合实际情况灵活确定。
以下对本申请实施例的技术方案的实现细节进行详细阐述：
图2示出了根据本申请的一个实施例的心电信号分类模型的训练方法的流程图。参照图2所示，所述方法包括：
步骤S210、构建训练样本集，所述训练样本集中每个训练样本包含有心电信号样本以及多个异常标注标签，所述多个异常标注标签为对所述心电信号样本中的多个异常类别标注的标签；
步骤S220、通过待训练心电信号分类模型的神经网络获取所述心电信号样本的特征向量以及所述心电信号样本的异常标签向量；
步骤S230、根据所述异常标签向量生成标签相关性矩阵，并根据所述特征向量和所述标签相关性矩阵，生成异常标签概率分布向量；
步骤S240、根据所述异常标签概率分布向量以及所述多个异常标注标签，对所述待训练心电信号分类模型进行训练，得到训练后的心电信号分类模型。
下面对这些步骤进行详细描述。
在步骤S210中，构建训练样本集，所述训练样本集中每个训练样本包含有心电信号样本以及多个异常标注标签，所述多个异常标注标签为对所述心电信号样本中的多个异常类别标注的标签。
本实施例中，心电信号为从体表记录的心脏每一心动周期所产生的电活动变化的信号，通常通过心电图(Electro Cardio Gram，ECG)图形化的方式记录。在训练心电信号分类模型之前，需要构建用于训练模型的训练样本集，训练样本集中每个训练样本包括心电信号样本和多个异常类别标注标签，多个异常类别标注标签为对心电信号样本中多个异常事件的异常类别标注的标签。
需要说明的是，对于每个训练样本中的心电信号样本可以通过如下方式采集得到：通过心电导联和传感器采集心电信号，由模数转换器把人体生理参数的模拟信号转化为数字信号，由存储器存储。
更具体而言，可以通过8导联采集心电信号，根据同一时刻R波顶点的位置向前截取2500个点，向后截取2500个点，每个导联截取到5000个点的数据，随后把相同时刻R波顶点对每个导联所截取的5000个点进行第二维度拼接，每导联的心电信号由1*5000维扩增为8*5000维，此时已经将原始每个导联的心电信号的一个心拍经过采样形成上述8*5000维的一个样本。然后对所有心电信号数据的R波顶定点进行同样的操作，得到包含(8*5000)*M维数据的数据集U，每个样本都是(8*5000)维，由于每个样本都是根据R波顶点的位置截取的，所以M为截取所使用的R波顶点的个数，也就是心电信号样本的个数，每个心电信号样本都是8*5000的8导联心电信号数据。
在采集得到心电信号样本后，可以对心电信号样本通过人工标注出多个异常类别标签，异常类别标签包括但不限于左心房肥大、右心房肥大、双心房肥大、左心室肥大、右心室肥大、双心室肥大、前壁心肌梗死、后壁心肌梗死、前间壁心肌梗死、侧壁心肌梗死、后间壁心肌梗死、窦性心律不齐、房性早搏、室性早搏、室上速、室速、房扑、房颤、室扑、室颤、左束支阻滞、右束支阻滞、房性逸博、室性逸博、心动过速、心动过缓、房室传导阻滞等等。
在步骤S220中，通过待训练心电信号分类模型的神经网络获取所述心电信号样本的特征向量以及所述心电信号样本的异常标签向量。
本实施例的待训练心电信号分类模型的神经网络可以包括输入层、特征提取层、众 包层及输出层。在实际应用中，心电信号分类模型的特征提取层可由任何具备对心电信号进行特征提取功能的网络构成，如可由卷积层、池化层及全连接层的卷积神经网络(Convolutional Neural Networks，CNN)构成。
具体到本步骤中，为了获取心电信号样本的特征向量，可以将心电信号样本输入至特征提取层，通过特征提取层中的神经网络，例如由卷积层、池化层以及全连接层构成的卷积神经网络，得到特征向量。需要说明的是，特征提取层输出的特征向量的维度为预先设置的降维后的特征向量的维度。
对于特征提取层的训练可以采用监督学习方法或非监督学习方法，非监督学习方法(Unsupervised Learning)是指在实际应用中，存在大量没有进行标注的或者少量标注的样本，从中学习样本之间的相互联系，非监督学习方法可以通过建立优化目标，来实现对特征提取层的训练。而采用监督学习方法训练时，可以将样本集中的样本和标注信息分别作为输入和期望输出，利用机器学习方法训练初始的特征提取层。
在本申请实施例的一些可选实现方式中，训练特征提取层的样本数据集合中的样本数据包括与心电信号样本来源相同的数据。特征提取层的训练可以在达到预设的优化目标或完成预设次数的迭代时结束。
进一步，除了通过特征提取层获取心电信号样本的特征向量以外，在本实施例中，还可以通过待训练心电信号分类模型的神经网络获取所述心电信号样本的异常标签向量。
具体地，可以利用预设的已训练深度学习算法模型对心电信号样本进行处理，而经过该已训练深度学习算法模型，可以获得心电信号样本的异常标签向量。将在心电信号中出现的异常类别记为1，没有出现则记为0，例如，假如某个心电信号样本有8类异常事件，异常标签向量可以表示为[1，0，1，0，1，0，0，1]，该异常标签向量则表示这个心电信号样本包含类别1、类别3、类别5和类别8的异常。
本实施方式中的已训练深度学习算法模型至少包括：卷积神经网络和多层感知器(MLP，Multilayer Perceptron)；相应的，利用该已训练深度学习算法模型对训练样本集进行处理的过程可包括将所述心电信号样本输入卷积神经网络，提取获得心电信号样本的特征信息；将心电信号样本的特征信息输入多层感知器，以使多层感知器根据预设的异常类别标签对特征信息进行向量维度的映射，获得心电信号样本的异常类别标签向量。
需要说明的是，卷积神经网络和多层感知器均属于较为成熟的网络架构，本实施方式对其架构组成不进行限制。但是，为了实现对心电信号样本进行分析以得到相应的异常类别标签向量，本申请通过利用卷积神经网络和多层感知器依次执行前述步骤，以实现相应功能。
在步骤S230中，根据所述异常标签向量生成标签相关性矩阵，并根据所述特征向量和所述标签相关性矩阵，生成异常标签概率分布向量。
在一些实施例中，在对心电信号样本进行处理得到异常标签向量后，可以计算各个异常标签之间的相关性，在本实施例中，两个异常标签相关性由如下模型计算得出：
Figure PCTCN2020135083-appb-000001
其中，P x,y为异常标签x和异常标签y的相关性。
然后，根据两个异常标签相关性可以生成标签相关性矩阵ρ，表示如下：
Figure PCTCN2020135083-appb-000002
在通过步骤S220获取到心电信号样本的特征向量，并根据异常标签向量生成标签相 关性矩阵后，可以将特征向量以及标签相关性矩阵相乘，获得异常标签概率分布向量。其中，异常概率分布向量表示的是心电信号样本中异常发生的概率。例如，假设获得的心电信号样本的异常标签概率分布向量是(0.1,0.2,0.5,0.1,0.1)，异常标签概率分布向量可以分别对应于异常标签A、异常标签B、异常标签C、异常标签D和异常标签E。
在其他实施例中，如图3所示，根据异常标签向量生成标签相关性矩阵还可以通过如下方式实现，具体包括步骤S310-步骤S320，现详细说明如下：
步骤S310、根据所述异常标签向量，获取所述心电信号样本的标签词向量矩阵、可训练参数矩阵以及条件概率矩阵。
具体的，可以利用word2vec技术以无监督的方式，对心电信号样本的异常标签向量进行训练，得到标签词向量矩阵。例如，假设将异常类别标签向量中每个标签向量看作一个词语，异常类别标签向量中有30个词语，每个词语经过编码后得到一个300维的向量，因此，标签词向量矩阵则表示为30*300维的矩阵。而对于心电信号样本的可训练参数矩阵，则可以采用卷积神经网络(CNN，Convolutional Neural Networks)作为一个可训练的特征检测器对异常标签向量进行特征提取，得到可训练参数矩阵。
而心电信号样本的条件概率矩阵则不用通过训练，直接统计异常标签向量中标签两两出现的条件概率，得到条件概率矩阵，举例说明，比如标签向量为[1，1，0]，则可以得到的条件概率矩阵m为：
Figure PCTCN2020135083-appb-000003
步骤S320、根据所述心电信号样本的标签词向量矩阵、所述可训练参数矩阵以及所述条件概率矩阵，获得所述标签相关性矩阵。
具体而言，将标签词向量矩阵、条件概率矩阵以及可训练参数矩阵相乘，可以得到心电信号样本的标签相关性矩阵。
继续参见图2，在步骤S240中，根据所述异常标签概率分布向量以及所述多个异常标注标签，对所述待训练心电信号分类模型进行训练，得到训练后的心电信号分类模型。
具体实现中，服务器可以根据获得的异常标签概率分布向量以及多个异常标注标签，对待训练心电信号分类模型进行训练，从而得到训练后的心电信号分类模型。
在本申请的一个实施例中，在训练待训练心电信号分类模型的过程中，可以基于损失函数，更新模型参数，从而得到训练后的心电信号分类模型，如图4所示，步骤S240可以具体包括：
步骤S410、根据所述多个异常标注标签以及所述异常标签概率分布向量，确定损失函数的值；
步骤S420、基于所述损失函数的值，更新所述待训练心电信号分类模型的模型参数，得到训练后的心电信号分类模型。
在该实施例中，服务器将每个训练样本输入待训练心电信号分类模型，每个训练样本中包含有心电信号样本，从而可以得到心电信号样本的预测结果；基于各个预测结果以及多个异常标注标签，确定损失函数的值，基于损失函数的值，更新待训练心电信号分类模型的模型参数，得到训练后的心电信号分类模型。
在本申请的一个实施例中，如图5所示，步骤S410可以具体包括：
步骤S510、将所述异常概率分布向量中大于第一预设阈值的概率值对应的异常标签作为所述心电信号样本的多个异常预测标签；
步骤S520、获取所述多个异常预测标签和所述多个异常标注标签之间的差异；
步骤S530、确定所获取的各个差异之和的平均值，将确定的所述平均值作为所述损 失函数的值。
在该实施例中，模型分类系统可以将异常概率分布向量中大于第一预设阈值的概率值对应的异常标签作为心电信号样本的多个异常预测标签。
举例说明，异常标签概率分布向量是(0.1，0.2，0.3，0.5，0.1，0.4，0.1)，异常标签概率分布向量可以分别对应于异常标签A、异常标签B、异常标签C、异常标签D、异常标签E、异常标签F和异常标签G，假设预设阈值为0.2，则可以得到心电信号样本的多个异常预测标签为异常标签C、异常标签D以及异常标签F。
然后，模型分类系统可以获取多个异常预测标签和多个异常标注标签之间的差异，确定获取的各个差异的差异和，将确定的差异和作为损失函数的值。
可选地，损失函数可以包括0-1损失(Zero-one Loss)函数、感知损失(Perceptron Loss)函数、铰链损失(Hinge Loss)函数、交叉熵损失函数、平方误差损失(Square Loss)函数、绝对值损失(Absolute Loss)函数、指数误差(Exponential Loss)函数和正则函数中的任意一种。
在本申请的一个实施例中，如图6所示，步骤S420可以具体包括：
步骤S610、当所述损失函数的值超出第二预设阈值时，基于所述待训练心电信号分类模型的损失函数确定相应的误差信号；
步骤S620、将所述误差信号在所述待训练心电信号分类模型中反向传播，并在传播的过程中更新所述待训练心电信号分类模型的模型参数，得到训练后的心电信号分类模型。
当服务器确定损失函数的值超出第二预设阈值时，基于待训练心电信号分类模型的损失函数确定相应的误差信号，将误差信号在待训练心电信号分类模型中反向传播，并在传播的过程中更新所述待训练心电信号分类模型的模型参数，得到训练后的心电信号分类模型。
这里对反向传播进行说明，将训练样本输入神经网络模型的输入层，经过隐藏层，最后达到输出层并输出结果，这是神经网络模型的前向传播过程。由于神经网络模型的输出结果与实际结果有误差，则计算输出结果与实际结果之间的误差，并将该误差从输出层向隐藏层反向传播，直至传播到输入层，在反向传播过程中，根据误差调整模型参数的值，不断迭代上述过程，直至收敛。
图7示出了根据本申请的一个实施例的心电信号分类方法的流程图，参照图7所示，心电信号分类方法包括：
步骤S710、通过心电信号分类模型对待分类心电信号进行特征提取，得到所述待分类心电信号的信号特征；
步骤S720、通过所述心电信号分类模型对所述信号特征进行分类预测，得到所述待分类心电信号对应的异常类别。
这里，在实际应用中，终端可以设置心电信号分类客户端，基于心电信号分类客户端发送带有待分类心电信号的分类请求给服务器。在实际实施时，服务器解析分类请求得到待分类心电信号，输入心电信号分类模型，输出待分类心电信号对应的异常类别。
具体地，服务器可以通过心电信号分类模型对待分类心电信号进行特征提取，得到待分类心电信号的信号特征；然后，通过心电信号分类模型对信号特征进行分类预测，得到待分类心电信号对应的异常类别。
以下介绍本申请的装置实施例，可以用于执行本申请上述实施例中的心电信号分类模型的训练方法。对于本申请装置实施例中未披露的细节，请参照本申请上述的心电信号分类模型的训练方法的实施例。
图8示出了根据本申请的一个实施例的心电信号分类模型的训练装置的框图，参照 图8所示，根据本申请的一个实施例的心电信号分类模型的训练装置800，包括：构建单元802、获取单元804、生成单元806和训练单元808。
其中，构建单元802，配置为构建训练样本集，所述训练样本集中每个训练样本包含有心电信号样本以及多个异常标注标签，所述多个异常标注标签为对所述心电信号样本中的多个异常类别标注的标签；获取单元804，配置为通过待训练心电信号分类模型的神经网络获取所述心电信号样本的特征向量以及所述心电信号样本的异常标签向量；生成单元806，配置为根据所述异常标签向量生成标签相关性矩阵，并根据所述特征向量和所述标签相关性矩阵，生成异常标签概率分布向量；训练单元808，配置为根据所述异常标签概率分布向量以及所述多个异常标注标签，对所述待训练心电信号分类模型进行训练，得到训练后的心电信号分类模型。
在本申请的一些实施例中，所述生成单元806配置为：根据所述异常标签向量，获取所述心电信号样本的标签词向量矩阵、可训练参数矩阵以及条件概率矩阵；根据所述心电信号样本的标签词向量矩阵、所述可训练参数矩阵以及所述条件概率矩阵，获得所述标签相关性矩阵。
在本申请的一些实施例中，所述训练单元808包括：确定子单元，配置为根据所述多个异常标注标签以及所述异常标签概率分布向量，确定损失函数的值；更新子单元，配置为基于所述损失函数的值，更新所述待训练心电信号分类模型的模型参数，得到训练后的心电信号分类模型。
在本申请的一些实施例中，所述确定子单元配置为将所述异常概率分布向量中大于第一预设阈值的概率值对应的异常标签作为所述心电信号样本的多个异常预测标签；获取所述多个异常预测标签和所述多个异常标注标签之间的差异；确定所获取的各个差异之和的平均值，将确定的所述平均值作为所述损失函数的值。
在本申请的一些实施例中，所述更新子单元配置为当所述损失函数的值超出第二预设阈值时，基于所述待训练心电信号分类模型的损失函数确定相应的误差信号；将所述误差信号在所述待训练心电信号分类模型中反向传播，并在传播的过程中更新所述待训练心电信号分类模型的模型参数，得到训练后的心电信号分类模型。
图9示出了根据本申请的一个实施例的心电信号分类装置的框图。
参见图9所示，根据本申请的一个实施例的心电信号分类装置900，包括：提取单元902以及分类单元904。
提取单元902，配置为通过心电信号分类模型对待分类心电信号进行特征提取，得到所述待分类心电信号的信号特征；分类单元904，配置为通过所述心电信号分类模型对所述信号特征进行分类预测，得到所述待分类心电信号对应的异常类别；其中，所述心电信号分类模型通过上述实施例提供的心电信号分类模型的训练方法进行训练得到。
图10示出了适于用来实现本申请实施例的电子设备的计算机系统的结构示意图。
需要说明的是，图10示出的电子设备的计算机系统1000仅是一个示例，不应对本申请实施例的功能和使用范围带来任何限制。
如图10所示，计算机系统1000包括中央处理单元(Central Processing Unit，CPU)1001，其可以根据存储在只读存储器(Read-Only Memory，ROM)1002中的程序或者从存储部分1008加载到随机访问存储器(Random Access Memory，RAM)1003中的程序而执行各种适当的动作和处理，例如执行上述实施例中所述的方法。在RAM 1003中，还存储有系统操作所需的各种程序和数据。CPU 1001、ROM 1002以及RAM 1003通过总线1004彼此相连。输入/输出(Input/Output，I/O)接口1005也连接至总线1004。
以下部件连接至I/O接口1005：包括键盘、鼠标等的输入部分1006；包括诸如阴极射线管(Cathode Ray Tube，CRT)、液晶显示器(Liquid Crystal Display，LCD)等以及 扬声器等的输出部分1007；包括硬盘等的存储部分1008；以及包括诸如LAN(Local Area Network，局域网)卡、调制解调器等的网络接口卡的通信部分1009。通信部分1009经由诸如因特网的网络执行通信处理。驱动器1010也根据需要连接至I/O接口1005。可拆卸介质1011，诸如磁盘、光盘、磁光盘、半导体存储器等等，根据需要安装在驱动器1010上，以便于从其上读出的计算机程序根据需要被安装入存储部分1008。
特别地，根据本申请的实施例，上文参考流程图描述的过程可以被实现为计算机软件程序。例如，本申请的实施例包括一种计算机程序产品，其包括承载在计算机可读介质上的计算机程序，该计算机程序包含用于执行流程图所示的方法的计算机程序。在这样的实施例中，该计算机程序可以通过通信部分1009从网络上被下载和安装，和/或从可拆卸介质1011被安装。在该计算机程序被中央处理单元(CPU)1001执行时，执行本申请的系统中限定的各种功能。
需要说明的是，本申请实施例所示的计算机可读介质可以是计算机可读信号介质或者计算机可读存储介质或者是上述两者的任意组合。计算机可读存储介质例如可以是——但不限于——电、磁、光、电磁、红外线、或半导体的系统、装置或器件，或者任意以上的组合。计算机可读存储介质的更具体的例子可以包括但不限于：具有一个或多个导线的电连接、便携式计算机磁盘、硬盘、随机访问存储器(RAM)、只读存储器(ROM)、可擦式可编程只读存储器(Erasable Programmable Read Only Memory，EPROM)、闪存、光纤、便携式紧凑磁盘只读存储器(Compact Disc Read-Only Memory，CD-ROM)、光存储器件、磁存储器件、或者上述的任意合适的组合。在本申请中，计算机可读存储介质可以是任何包含或存储程序的有形介质，该程序可以被指令执行系统、装置或者器件使用或者与其结合使用。而在本申请中，计算机可读的信号介质可以包括在基带中或者作为载波一部分传播的数据信号，其中承载了计算机可读的计算机程序。这种传播的数据信号可以采用多种形式，包括但不限于电磁信号、光信号或上述的任意合适的组合。计算机可读的信号介质还可以是计算机可读存储介质以外的任何计算机可读介质，该计算机可读介质可以发送、传播或者传输用于由指令执行系统、装置或者器件使用或者与其结合使用的程序。计算机可读介质上包含的计算机程序可以用任何适当的介质传输，包括但不限于：无线、有线等等，或者上述的任意合适的组合。
附图中的流程图和框图，图示了按照本申请各种实施例的系统、方法和计算机程序产品的可能实现的体系架构、功能和操作。其中，流程图或框图中的每个方框可以代表一个模块、程序段、或代码的一部分，上述模块、程序段、或代码的一部分包含一个或多个用于实现规定的逻辑功能的可执行指令。也应当注意，在有些作为替换的实现中，方框中所标注的功能也可以以不同于附图中所标注的顺序发生。例如，两个接连地表示的方框实际上可以基本并行地执行，它们有时也可以按相反的顺序执行，这依所涉及的功能而定。也要注意的是，框图或流程图中的每个方框、以及框图或流程图中的方框的组合，可以用执行规定的功能或操作的专用的基于硬件的系统来实现，或者可以用专用硬件与计算机指令的组合来实现。
描述于本申请实施例中所涉及到的单元可以通过软件的方式实现，也可以通过硬件的方式来实现，所描述的单元也可以设置在处理器中。其中，这些单元的名称在某种情况下并不构成对该单元本身的限定。
作为另一方面，本申请还提供了一种计算机可读(存储)介质，该计算机可读介质可以是上述实施例中描述的电子设备中所包含的；也可以是单独存在，而未装配入该电子设备中。上述计算机可读介质承载有一个或者多个程序，当上述一个或者多个程序被一个该电子设备执行时，使得该电子设备实现上述实施例中所述的方法。
可选的，本申请涉及的存储介质如计算机可读存储介质可以是非易失性的，也可以 是易失性的。
应当注意，尽管在上文详细描述中提及了用于动作执行的设备的若干模块或者单元，但是这种划分并非强制性的。实际上，根据本申请的实施方式，上文描述的两个或更多模块或者单元的特征和功能可以在一个模块或者单元中具体化。反之，上文描述的一个模块或者单元的特征和功能可以进一步划分为由多个模块或者单元来具体化。
通过以上的实施方式的描述，本领域的技术人员易于理解，这里描述的示例实施方式可以通过软件实现，也可以通过软件结合必要的硬件的方式来实现。因此，根据本申请实施方式的技术方案可以以软件产品的形式体现出来，该软件产品是可以存储在存储介质(可以是CD-ROM，U盘，移动硬盘等)中或网络上的计算机可读指令，所述存储介质可以是易失性的，也可以是非易失性的。所述计算机可读指令包括若干指令以使得一台计算设备(可以是个人计算机、服务器、触控终端、或者网络设备等)执行根据本申请实施方式的方法。
本领域技术人员在考虑说明书及实践这里公开的实施方式后，将容易想到本申请的其它实施方案。本申请旨在涵盖本申请的任何变型、用途或者适应性变化，这些变型、用途或者适应性变化遵循本申请的一般性原理并包括本申请未公开的本技术领域中的公知常识或惯用技术手段。
应当理解的是，本申请并不局限于上面已经描述并在附图中示出的精确结构，并且可以在不脱离其范围进行各种修改和改变。本申请的范围仅由所附的权利要求来限制。

Claims (20)

  1. 一种心电信号分类模型的训练方法，其中，所述方法包括：
    构建训练样本集，所述训练样本集中每个训练样本包含有心电信号样本以及多个异常标注标签，所述多个异常标注标签为对所述心电信号样本中的多个异常类别标注的标签；
    通过待训练心电信号分类模型的神经网络获取所述心电信号样本的特征向量以及所述心电信号样本的异常标签向量；
    根据所述异常标签向量生成标签相关性矩阵，并根据所述特征向量和所述标签相关性矩阵，生成异常标签概率分布向量；
    根据所述异常标签概率分布向量以及所述多个异常标注标签，对所述待训练心电信号分类模型进行训练，得到训练后的心电信号分类模型。
  2. 根据权利要求1所述的方法，其中，根据所述异常标签向量生成标签相关性矩阵，包括：
    根据所述异常标签向量，获取所述心电信号样本的标签词向量矩阵、可训练参数矩阵以及条件概率矩阵；
    根据所述心电信号样本的标签词向量矩阵、所述可训练参数矩阵以及所述条件概率矩阵，获得所述标签相关性矩阵。
  3. 根据权利要求1所述的方法，其中，根据所述异常标签概率分布向量以及所述多个异常标注标签，对所述待训练心电信号分类模型进行训练，得到训练后的心电信号分类模型，包括：
    根据所述多个异常标注标签以及所述异常标签概率分布向量，确定损失函数的值；
    基于所述损失函数的值，更新所述待训练心电信号分类模型的模型参数，得到训练后的心电信号分类模型。
  4. 根据权利要求3所述的方法，其中，根据所述多个异常标注标签以及所述异常标签概率分布向量，确定损失函数的值，包括：
    将所述异常概率分布向量中大于第一预设阈值的概率值对应的异常标签作为所述心电信号样本的多个异常预测标签；
    获取所述多个异常预测标签和所述多个异常标注标签之间的差异；
    确定所获取的各个差异之和的平均值，将确定的所述平均值作为所述损失函数的值。
  5. 根据权利要求3所述的方法，其中，基于所述损失函数的值，更新所述待训练心电信号分类模型的模型参数，得到训练后的心电信号分类模型，包括：
    当所述损失函数的值超出第二预设阈值时，基于所述待训练心电信号分类模型的损失函数确定相应的误差信号；
    将所述误差信号在所述待训练心电信号分类模型中反向传播，并在传播的过程中更新所述待训练心电信号分类模型的模型参数，得到训练后的心电信号分类模型。
  6. 一种心电信号分类方法，其中，所述方法包括：
    通过心电信号分类模型对待分类心电信号进行特征提取，得到所述待分类心电信号的信号特征；
    通过所述心电信号分类模型对所述信号特征进行分类预测，得到所述待分类心电信号对应的异常类别；
    其中，所述心电信号分类模型通过权利要求1至5任一项所述的方法进行训练得到。
  7. 一种心电信号分类模型的训练装置，其中，所述装置包括：
    构建单元，配置为构建训练样本集，所述训练样本集中每个训练样本包含有心电信号样本以及多个异常标注标签，所述多个异常标注标签为对所述心电信号样本中的多个 异常类别标注的标签；
    获取单元，配置为通过待训练心电信号分类模型的神经网络获取所述心电信号样本的特征向量以及所述心电信号样本的异常标签向量；
    生成单元，配置为根据所述异常标签向量生成标签相关性矩阵，并根据所述特征向量和所述标签相关性矩阵，生成异常标签概率分布向量；
    训练单元，配置为根据所述异常标签概率分布向量以及所述多个异常标注标签，对所述待训练心电信号分类模型进行训练，得到训练后的心电信号分类模型。
  8. 一种心电信号分类装置，其中，所述装置包括：
    提取单元，配置为通过心电信号分类模型对待分类心电信号进行特征提取，得到所述待分类心电信号的信号特征；
    分类单元，配置为通过所述心电信号分类模型对所述信号特征进行分类预测，得到所述待分类心电信号对应的异常类别；
    其中，所述心电信号分类模型通过权利要求1至5任一项所述的方法进行训练得到。
  9. 一种计算机可读存储介质，其上存储有计算机程序，其中，所述计算机程序被处理器执行时实现以下方法：
    构建训练样本集，所述训练样本集中每个训练样本包含有心电信号样本以及多个异常标注标签，所述多个异常标注标签为对所述心电信号样本中的多个异常类别标注的标签；
    通过待训练心电信号分类模型的神经网络获取所述心电信号样本的特征向量以及所述心电信号样本的异常标签向量；
    根据所述异常标签向量生成标签相关性矩阵，并根据所述特征向量和所述标签相关性矩阵，生成异常标签概率分布向量；
    根据所述异常标签概率分布向量以及所述多个异常标注标签，对所述待训练心电信号分类模型进行训练，得到训练后的心电信号分类模型。
  10. 根据权利要求9所述的计算机可读存储介质，其中，根据所述异常标签向量生成标签相关性矩阵时，具体实现：
    根据所述异常标签向量，获取所述心电信号样本的标签词向量矩阵、可训练参数矩阵以及条件概率矩阵；
    根据所述心电信号样本的标签词向量矩阵、所述可训练参数矩阵以及所述条件概率矩阵，获得所述标签相关性矩阵。
  11. 根据权利要求9所述的计算机可读存储介质，其中，根据所述异常标签概率分布向量以及所述多个异常标注标签，对所述待训练心电信号分类模型进行训练，得到训练后的心电信号分类模型时，具体实现：
    根据所述多个异常标注标签以及所述异常标签概率分布向量，确定损失函数的值；
    基于所述损失函数的值，更新所述待训练心电信号分类模型的模型参数，得到训练后的心电信号分类模型。
  12. 根据权利要求11所述的计算机可读存储介质，其中，根据所述多个异常标注标签以及所述异常标签概率分布向量，确定损失函数的值时，具体实现：
    将所述异常概率分布向量中大于第一预设阈值的概率值对应的异常标签作为所述心电信号样本的多个异常预测标签；
    获取所述多个异常预测标签和所述多个异常标注标签之间的差异；
    确定所获取的各个差异之和的平均值，将确定的所述平均值作为所述损失函数的值。
  13. 根据权利要求11所述的计算机可读存储介质，其中，基于所述损失函数的值，更新所述待训练心电信号分类模型的模型参数，得到训练后的心电信号分类模型时，具 体实现：
    当所述损失函数的值超出第二预设阈值时，基于所述待训练心电信号分类模型的损失函数确定相应的误差信号；
    将所述误差信号在所述待训练心电信号分类模型中反向传播，并在传播的过程中更新所述待训练心电信号分类模型的模型参数，得到训练后的心电信号分类模型。
  14. 一种计算机可读存储介质，其上存储有计算机程序，其中，所述计算机程序被处理器执行时实现以下方法：
    通过心电信号分类模型对待分类心电信号进行特征提取，得到所述待分类心电信号的信号特征；
    通过所述心电信号分类模型对所述信号特征进行分类预测，得到所述待分类心电信号对应的异常类别；
    其中，所述心电信号分类模型通过权利要求1至5任一项所述的方法进行训练得到。
  15. 一种电子设备，其中，包括：
    处理器；以及
    存储器，用于存储所述处理器的可执行指令；
    其中，所述处理器配置为经由执行所述可执行指令来执行以下方法：
    构建训练样本集，所述训练样本集中每个训练样本包含有心电信号样本以及多个异常标注标签，所述多个异常标注标签为对所述心电信号样本中的多个异常类别标注的标签；
    通过待训练心电信号分类模型的神经网络获取所述心电信号样本的特征向量以及所述心电信号样本的异常标签向量；
    根据所述异常标签向量生成标签相关性矩阵，并根据所述特征向量和所述标签相关性矩阵，生成异常标签概率分布向量；
    根据所述异常标签概率分布向量以及所述多个异常标注标签，对所述待训练心电信号分类模型进行训练，得到训练后的心电信号分类模型。
  16. 根据权利要求15所述的电子设备，其中，根据所述异常标签向量生成标签相关性矩阵时，具体执行：
    根据所述异常标签向量，获取所述心电信号样本的标签词向量矩阵、可训练参数矩阵以及条件概率矩阵；
    根据所述心电信号样本的标签词向量矩阵、所述可训练参数矩阵以及所述条件概率矩阵，获得所述标签相关性矩阵。
  17. 根据权利要求15所述的电子设备，其中，根据所述异常标签概率分布向量以及所述多个异常标注标签，对所述待训练心电信号分类模型进行训练，得到训练后的心电信号分类模型时，具体执行：
    根据所述多个异常标注标签以及所述异常标签概率分布向量，确定损失函数的值；
    基于所述损失函数的值，更新所述待训练心电信号分类模型的模型参数，得到训练后的心电信号分类模型。
  18. 根据权利要求17所述的电子设备，其中，根据所述多个异常标注标签以及所述异常标签概率分布向量，确定损失函数的值时，具体执行：
    将所述异常概率分布向量中大于第一预设阈值的概率值对应的异常标签作为所述心电信号样本的多个异常预测标签；
    获取所述多个异常预测标签和所述多个异常标注标签之间的差异；
    确定所获取的各个差异之和的平均值，将确定的所述平均值作为所述损失函数的值。
  19. 根据权利要求17所述的电子设备，其中，基于所述损失函数的值，更新所述待 训练心电信号分类模型的模型参数，得到训练后的心电信号分类模型时，具体执行：
    当所述损失函数的值超出第二预设阈值时，基于所述待训练心电信号分类模型的损失函数确定相应的误差信号；
    将所述误差信号在所述待训练心电信号分类模型中反向传播，并在传播的过程中更新所述待训练心电信号分类模型的模型参数，得到训练后的心电信号分类模型。
  20. 一种电子设备，其中，包括：
    处理器；以及
    存储器，用于存储所述处理器的可执行指令；
    其中，所述处理器配置为经由执行所述可执行指令来执行以下方法：
    通过心电信号分类模型对待分类心电信号进行特征提取，得到所述待分类心电信号的信号特征；
    通过所述心电信号分类模型对所述信号特征进行分类预测，得到所述待分类心电信号对应的异常类别；
    其中，所述心电信号分类模型通过权利要求1至5任一项所述的方法进行训练得到。
PCT/CN2020/135083 2020-10-22 2020-12-10 心电信号分类模型的训练方法、心电信号分类方法及装置 WO2021179712A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202011142763.1 2020-10-22
CN202011142763.1A CN112353402B (zh) 2020-10-22 2020-10-22 心电信号分类模型的训练方法、心电信号分类方法及装置

Publications (1)

Publication Number Publication Date
WO2021179712A1 true WO2021179712A1 (zh) 2021-09-16

Family

ID=74511736

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/135083 WO2021179712A1 (zh) 2020-10-22 2020-12-10 心电信号分类模型的训练方法、心电信号分类方法及装置

Country Status (2)

Country Link
CN (1) CN112353402B (zh)
WO (1) WO2021179712A1 (zh)

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114129169A (zh) * 2021-11-22 2022-03-04 中节能风力发电股份有限公司 一种生物电信号数据识别方法、系统、介质和设备
CN114139031A (zh) * 2021-10-28 2022-03-04 马上消费金融股份有限公司 数据分类方法、装置、电子设备及存储介质
CN114328936A (zh) * 2022-03-01 2022-04-12 支付宝(杭州)信息技术有限公司 建立分类模型的方法和装置
CN114880646A (zh) * 2022-07-13 2022-08-09 华南师范大学 一种身份识别方法、装置、电子设备及存储介质
CN115913769A (zh) * 2022-12-20 2023-04-04 石家庄曲竹闻网络科技有限公司 基于人工智能的数据安全存储方法及系统
CN116186271A (zh) * 2023-04-19 2023-05-30 北京亚信数据有限公司 一种医疗专业名词分类模型训练方法、分类方法及装置
CN116842330A (zh) * 2023-08-31 2023-10-03 庆云县人民医院 一种可对比历史记录的保健信息处理方法及装置
CN117481606A (zh) * 2023-11-13 2024-02-02 齐鲁工业大学(山东省科学院) 一种基于改进孪生网络的小样本心电信号分类方法

Families Citing this family (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
GB2606700A (en) * 2021-04-21 2022-11-23 Prevayl Innovations Ltd Method of preparing training data for use in training a health event identification machine-learning model
CN113349792B (zh) * 2021-05-31 2022-10-11 平安科技（深圳）有限公司 基于多导联心电信号的分类方法、装置、设备及介质
CN113837216B (zh) * 2021-06-01 2024-05-10 腾讯科技（深圳）有限公司 数据分类方法、训练方法、装置、介质及电子设备
CN113378967B (zh) * 2021-06-28 2022-11-08 哈尔滨工业大学 一种基于卷积神经网络与迁移学习的结构健康监测多元数据异常诊断方法
CN113688244A (zh) * 2021-08-31 2021-11-23 中国平安人寿保险股份有限公司 基于神经网络的文本分类方法、系统、设备及存储介质
CN115577258A (zh) * 2022-09-08 2023-01-06 中国电信股份有限公司 振动信号识别模型训练方法、电机故障检测方法及装置
CN115607166B (zh) * 2022-10-08 2023-08-22 中国医学科学院阜外医院 一种心电信号智能分析方法及系统、智能心电辅助系统
CN117031421A (zh) * 2023-08-14 2023-11-10 广州桐富科技发展有限公司 雷达信号模拟系统的异常分析方法及系统

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200015694A1 (en) * 2015-10-27 2020-01-16 Cardiologs Technologies Sas Automatic method to delineate or categorize an electrocardiogram
CN110807468A (zh) * 2019-09-19 2020-02-18 平安科技（深圳）有限公司 检测异常邮件的方法、装置、设备和存储介质
CN111160139A (zh) * 2019-12-13 2020-05-15 中国科学院深圳先进技术研究院 心电信号的处理方法、装置及终端设备
CN111275093A (zh) * 2020-01-17 2020-06-12 上海优加利健康管理有限公司 一种用于多标签标注心电信号的心搏分类方法和装置
CN111700608A (zh) * 2020-07-24 2020-09-25 武汉中旗生物医疗电子有限公司 一种心电信号多分类方法及装置
KR20210015306A (ko) * 2019-08-01 2021-02-10 성균관대학교산학협력단 클래스 확률 출력망 기반 심장 상태 분류 방법 및 장치

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105320967A (zh) * 2015-11-04 2016-02-10 中科院成都信息技术股份有限公司 基于标签相关性的多标签AdaBoost集成方法
CN109840531B (zh) * 2017-11-24 2023-08-25 华为技术有限公司 训练多标签分类模型的方法和装置
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
CN110807495B (zh) * 2019-11-08 2023-09-12 腾讯科技（深圳）有限公司 多标签分类方法、装置、电子设备及存储介质
CN110840402B (zh) * 2019-11-19 2021-02-26 山东大学 一种基于机器学习的房颤信号识别方法及系统
CN111012332A (zh) * 2020-01-20 2020-04-17 太原理工大学 基于神经网络的12导联心电信号多标签分类方法

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200015694A1 (en) * 2015-10-27 2020-01-16 Cardiologs Technologies Sas Automatic method to delineate or categorize an electrocardiogram
KR20210015306A (ko) * 2019-08-01 2021-02-10 성균관대학교산학협력단 클래스 확률 출력망 기반 심장 상태 분류 방법 및 장치
CN110807468A (zh) * 2019-09-19 2020-02-18 平安科技（深圳）有限公司 检测异常邮件的方法、装置、设备和存储介质
CN111160139A (zh) * 2019-12-13 2020-05-15 中国科学院深圳先进技术研究院 心电信号的处理方法、装置及终端设备
CN111275093A (zh) * 2020-01-17 2020-06-12 上海优加利健康管理有限公司 一种用于多标签标注心电信号的心搏分类方法和装置
CN111700608A (zh) * 2020-07-24 2020-09-25 武汉中旗生物医疗电子有限公司 一种心电信号多分类方法及装置

Cited By (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114139031A (zh) * 2021-10-28 2022-03-04 马上消费金融股份有限公司 数据分类方法、装置、电子设备及存储介质
CN114139031B (zh) * 2021-10-28 2024-03-19 马上消费金融股份有限公司 数据分类方法、装置、电子设备及存储介质
CN114129169A (zh) * 2021-11-22 2022-03-04 中节能风力发电股份有限公司 一种生物电信号数据识别方法、系统、介质和设备
CN114328936A (zh) * 2022-03-01 2022-04-12 支付宝(杭州)信息技术有限公司 建立分类模型的方法和装置
CN114880646A (zh) * 2022-07-13 2022-08-09 华南师范大学 一种身份识别方法、装置、电子设备及存储介质
CN114880646B (zh) * 2022-07-13 2023-08-08 华南师范大学 一种身份识别方法、装置、电子设备及存储介质
CN115913769B (zh) * 2022-12-20 2023-09-08 海口盛通达投资控股有限责任公司 基于人工智能的数据安全存储方法及系统
CN115913769A (zh) * 2022-12-20 2023-04-04 石家庄曲竹闻网络科技有限公司 基于人工智能的数据安全存储方法及系统
CN116186271B (zh) * 2023-04-19 2023-07-25 北京亚信数据有限公司 一种医疗专业名词分类模型训练方法、分类方法及装置
CN116186271A (zh) * 2023-04-19 2023-05-30 北京亚信数据有限公司 一种医疗专业名词分类模型训练方法、分类方法及装置
CN116842330A (zh) * 2023-08-31 2023-10-03 庆云县人民医院 一种可对比历史记录的保健信息处理方法及装置
CN116842330B (zh) * 2023-08-31 2023-11-24 庆云县人民医院 一种可对比历史记录的保健信息处理方法及装置
CN117481606A (zh) * 2023-11-13 2024-02-02 齐鲁工业大学(山东省科学院) 一种基于改进孪生网络的小样本心电信号分类方法
CN117481606B (zh) * 2023-11-13 2024-05-14 齐鲁工业大学(山东省科学院) 一种基于改进孪生网络的小样本心电信号分类方法

Also Published As

Publication number Publication date
CN112353402B (zh) 2022-09-27
CN112353402A (zh) 2021-02-12

Similar Documents

Publication Publication Date Title
WO2021179712A1 (zh) 心电信号分类模型的训练方法、心电信号分类方法及装置
Mincholé et al. Machine learning in the electrocardiogram
Kilic Artificial intelligence and machine learning in cardiovascular health care
Wiens et al. Machine learning for healthcare: on the verge of a major shift in healthcare epidemiology
Sayantan et al. Classification of ECG beats using deep belief network and active learning
Li et al. SLC-GAN: An automated myocardial infarction detection model based on generative adversarial networks and convolutional neural networks with single-lead electrocardiogram synthesis
Wang et al. Inter-patient ECG arrhythmia heartbeat classification based on unsupervised domain adaptation
CN111488793B (zh) 心电信号分类方法、装置、电子设备和存储介质
EP4042445A1 (en) Systems and methods for reduced lead electrocardiogram diagnosis using deep neural networks and rule-based systems
Feng et al. A novel bidirectional LSTM network based on scale factor for atrial fibrillation signals classification
Muthalaly et al. Applications of machine learning in cardiac electrophysiology
Sanchez de la Nava et al. Artificial intelligence for a personalized diagnosis and treatment of atrial fibrillation
Jang et al. Unsupervised feature learning for electrocardiogram data using the convolutional variational autoencoder
Karthiga et al. Hybrid optimized convolutional neural network for efficient classification of ECG signals in healthcare monitoring
Bond et al. Machine learning with electrocardiograms: A call for guidelines and best practices for ‘stress testing’algorithms
Roy et al. ECG-NET: A deep LSTM autoencoder for detecting anomalous ECG
Seetharam et al. Applications of machine learning in cardiology
Ran et al. Label correlation embedding guided network for multi-label ECG arrhythmia diagnosis
Li et al. Inter-patient automated arrhythmia classification: A new approach of weight capsule and sequence to sequence combination
Ying et al. FedECG: A federated semi-supervised learning framework for electrocardiogram abnormalities prediction
Tao et al. A resource-efficient ECG diagnosis model for mobile health devices
Qu et al. Quantum conditional generative adversarial network based on patch method for abnormal electrocardiogram generation
Dimitri et al. A preliminary evaluation of Echo State Networks for Brugada syndrome classification
Cheng et al. MSW-Transformer: Multi-Scale Shifted Windows Transformer Networks for 12-Lead ECG Classification
Wu et al. Autonomous detection of myocarditis based on the fusion of improved quantum genetic algorithm and adaptive differential evolution optimization back propagation neural network

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: 20924055

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20924055

Country of ref document: EP

Kind code of ref document: A1