WO2019200746A1 - Ecg信号的检测方法、装置、计算机设备和存储介质 - Google Patents

Ecg信号的检测方法、装置、计算机设备和存储介质 Download PDF

Info

Publication number
WO2019200746A1
WO2019200746A1 PCT/CN2018/095494 CN2018095494W WO2019200746A1 WO 2019200746 A1 WO2019200746 A1 WO 2019200746A1 CN 2018095494 W CN2018095494 W CN 2018095494W WO 2019200746 A1 WO2019200746 A1 WO 2019200746A1
Authority
WO
WIPO (PCT)
Prior art keywords
ecg signal
signal
training
rhythm type
ecg
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/CN2018/095494
Other languages
English (en)
French (fr)
Inventor
王健宗
吴天博
刘新卉
刘莉红
马进
肖京
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.)
Ping An Technology Shenzhen Co Ltd
Original Assignee
Ping An Technology Shenzhen Co Ltd
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 Ping An Technology Shenzhen Co Ltd filed Critical Ping An Technology Shenzhen Co Ltd
Publication of WO2019200746A1 publication Critical patent/WO2019200746A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Classifications

    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
    • A61B5/316—Modalities, i.e. specific diagnostic methods
    • A61B5/318—Heart-related electrical modalities, e.g. electrocardiography [ECG]
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—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

  • the present application relates to the field of computer technologies, and in particular, to a method, an apparatus, a computer device, and a storage medium for detecting an ECG signal.
  • Arrhythmia refers to a variety of symptoms caused by abnormal electrical conduction system of the heart, including the general expression of irregular, too fast, or too slow heartbeat, and is an important group of diseases in cardiovascular diseases. It can occur alone or in combination with other cardiovascular diseases. The proportion of patients with arrhythmia is located in the first two of various cardiovascular diseases, which seriously endangers the health of Chinese residents. Therefore, fast and accurate arrhythmia detection technology is extremely important.
  • ECG signals ECG signals, Electrocardiogram
  • noise due to the diversity of ECG signals (ECG signals, Electrocardiogram) and the existence of noise, there are large errors in machine detection.
  • ECG signals ECG signals, Electrocardiogram
  • the mainstream arrhythmia detection technology is artificially diagnosed by experts on ECG signals, which has low diagnostic speed and diagnostic efficiency. Low, it costs a lot of manpower and material costs.
  • the main purpose of the present application is to provide a method, a device, a computer device and a storage medium for detecting ECG signals, which overcome the defects of low diagnostic speed and low diagnostic efficiency when manually diagnosed arrhythmia.
  • the present application provides a method for detecting an ECG signal, including the following steps:
  • the application also provides an apparatus for detecting an ECG signal, comprising:
  • a training unit configured to input a first ECG signal of a known signal rhythm type into a convolutional neural network for training, training a corresponding training parameter, and obtaining a convolutional neural network of the training parameter as a detection model of the ECG signal;
  • a detecting unit configured to input a second ECG signal to be detected into the detection model of the ECG signal for calculation, and output a signal rhythm type of the second ECG signal.
  • the application further provides a computer device comprising a memory and a processor, wherein the memory stores computer readable instructions, and the processor implements the steps of the method when the computer readable instructions are executed.
  • the present application also provides a computer non-transitory readable storage medium having stored thereon computer readable instructions that, when executed by a processor, implement the steps of the above methods.
  • the method, device, computer device and storage medium for detecting ECG signals provided in the present application input a first ECG signal of a known signal rhythm type into a convolutional neural network for training, and training corresponding training parameters, which will result
  • the convolutional neural network of the training parameter is used as a detection model of the ECG signal;
  • the second ECG signal to be detected is input into the detection model of the ECG signal for calculation, and the signal rhythm type of the second ECG signal is output; the training is completed.
  • the ECG signal detection model can replace the medical expert's automatic detection of the signal rhythm type in the patient's ECG signal, saving a lot of manpower and material cost; at the same time, the accuracy of detecting the signal rhythm type in the patient's ECG signal can reach or exceed the expert.
  • FIG. 1 is a schematic diagram of steps of a method for detecting an ECG signal in an embodiment of the present application
  • FIG. 2 is a schematic diagram of steps of a method for detecting an ECG signal in another embodiment of the present application
  • FIG. 3 is a structural block diagram of an apparatus for detecting an ECG signal in an embodiment of the present application
  • FIG. 4 is a structural block diagram of an apparatus for detecting an ECG signal in another embodiment of the present application.
  • FIG. 5 is a structural block diagram of an apparatus for detecting an ECG signal in still another embodiment of the present application.
  • FIG. 6 is a schematic block diagram showing the structure of a computer device according to an embodiment of the present application.
  • an embodiment of the present application provides a method for detecting an ECG signal, including the following steps:
  • Step S1 input the first ECG signal of the known signal rhythm type into the convolutional neural network for training, train the corresponding training parameters, and obtain the convolutional neural network of the training parameters as the detection model of the ECG signal;
  • the ECG signal generally includes 14 kinds of signal rhythm types, wherein 12 abnormal rhythms, one sinus rhythm and one noise rhythm, the signal rhythm expresses the state information of the ECG signal, when the ECG signal is detected as above
  • the signal rhythm type of the ECG signal becomes a scientific means of detecting whether the patient has a heart rhythm abnormality.
  • the signal rhythm type of the first ECG signal is known in advance.
  • the above convolutional neural network (Convolutional Neural Networks, CNN) can be a Sequence to sequence CNN model, which is a deep learning network for one-dimensional time series models, which is suitable for ECG time series data (ie ECG signal data); therefore, for ECG signals are more accurate and faster when training.
  • the source of the first ECG signal may be selected from the detected ECG signals in the history detection database, or may be temporarily collected and marked with a signal rhythm type. For example, in a specific embodiment, finding as many target patients as possible to wear a cardiac monitoring device to continuously monitor their heart rhythm for a number of days, collecting a plurality of ECG signals having a frequency of 200 Hz for n seconds (generally less than 60 seconds); After the ECG signal, the expert (doctor) divides the above ECG signal into regions, and the expert judges the signal rhythm type according to the ECG signal characteristics of each region, and uses the labeling tool to mark, that is, the specific signal rhythm type of the ECG signal of each region is marked. .
  • the ECG signal labeled with the signal rhythm type described above is used as the first ECG signal.
  • the first ECG signal is input into the convolutional neural network, and the unknown training result is trained according to the unknown training parameter, and the training result is associated with the training parameter, and different training parameters result in different training results;
  • the training result of the expected output is the signal rhythm type of the first ECG signal.
  • the signal rhythm type of the first ECG signal is known, and only the first ECG signal is input into the convolutional neural network for training, and the output training result is the signal rhythm type known to the first ECG signal.
  • the training parameter when the convolutional neural network detects the signal rhythm type of the ECG signal in the present embodiment can be obtained.
  • the training parameters are input into the convolutional neural network to obtain the detection model of the ECG signal, and the detection model of the trained ECG signal can be used to detect the unknown ECG signal, predict the result, thereby replacing the manual detection, reducing the manpower, Material cost; and the detection speed is obviously improved, and the detection efficiency is improved.
  • Step S2 inputting a second ECG signal to be detected into a detection model of the ECG signal for calculation, and outputting a signal rhythm type of the second ECG signal.
  • the detection model of the ECG signal is the detection model obtained by the training in the above step S1.
  • the medical device can collect the first
  • the second ECG signal is input to the detection model of the ECG signal, and the detection model of the ECG signal outputs the second ECG.
  • the predicted result of the signal that is, the type of signal rhythm of the second ECG signal is predicted, thereby judging whether the patient has an arrhythmia based on the predicted result.
  • the machine automatically detects the type of signal rhythm in the patient's ECG signal, thereby judging whether the patient has arrhythmia according to the signal rhythm type, improving the diagnosis speed of diagnosis of arrhythmia, saving a lot of manpower and material cost; meanwhile, the diagnostic accuracy can reach Even more than experts.
  • the method before the step of inputting a first ECG signal of a known signal rhythm type into a convolutional neural network for training and training corresponding training parameters, the method includes:
  • Step S101 performing normalization processing on the first ECG signal.
  • the normalization process is a commonly used method of convolutional neural networks, and is mainly used to normalize the first ECG signal to a specified range.
  • the first ECG signal is normalized to a mean value of 0 and the variance is 1 as an input to the convolutional neural network; when the convolutional neural network outputs, a prediction result for the first ECG signal can be output every 1 s, all The predicted results combine to correspond to the entire first ECG signal sequence.
  • the above-described normalization process is to subtract the mean of all ECG signal data for each second ECG signal data and divide by the standard deviation of all ECG signal data. It should be understood that, when the signal rhythm type of the second ECG signal is detected in the above step S2, the second ECG signal is subjected to normalization processing in accordance with the same processing procedure.
  • the step of inputting the second ECG signal to be detected into the detection model of the ECG signal, and outputting the signal rhythm type of the second ECG signal includes:
  • the second ECG signal to be detected is input to an input layer in the detection model of the ECG signal, and is outputted through a 32-layer convolution layer and then output through the output layer to obtain a signal rhythm type of the second ECG signal.
  • the second ECG signal is input to the detection model of the ECG signal for calculation to output the signal rhythm type of the second ECG signal.
  • the convolutional neural network in the above step S1 has the same network structure as the detection model of the ECG signal in this step, and the only difference is that the training parameters are unknown. Therefore, the training process in step S1 is consistent with the calculation process in the step S2. The loss function used is consistent, the optimization method of the network model is consistent, and the training process in step S1 is not described again, which can refer to the specific calculation process in step S2.
  • the detection model of the convolutional neural network and the ECG signal in this embodiment is a 34-layer convolutional neural network, and the network structure specifically includes an input layer, an output layer, and 16 residual blocks, wherein each residual block includes 2 Convolutional layer.
  • the size of each convolutional layer filter is: filter length is 16 x 1, filter number is 64k, k starts at 1, and increments by 1 every 4 residual blocks k.
  • filter length is 16 x 1
  • filter number is 64k
  • k starts at 1
  • the ECG signal feature in the network structure it is processed by a common method in the neural network, for example, every other residual block, and downsampled by a coefficient of 2.
  • the method before the step of convolving the 32-layer convolution layer and outputting through the output layer, the method includes:
  • the loss function used in the detection model of the convolutional neural network and the ECG signal is a cross entropy loss function
  • the optimization method of the convolutional neural network is the Adam method
  • the cross entropy loss function is used to measure the volume.
  • the Adam method dynamically adjusts the learning rate for each parameter based on the first-order moment estimate and the second-order moment estimate of the gradient of each parameter according to the loss function. The learning rate will gradually decrease after Loss (loss function) no longer decreases. The reason why loss does not decrease is that the learning rate is too large, so the way to reduce the learning rate is to reduce the loss.
  • Optimized using the Adam method The optimization method can also use SGD (random steepest descent method), Momentum (momentum optimization) and other methods, but after experiment comparison, it is found that using Adam method is the best.
  • the first ECG signal of the known signal rhythm type is input into the convolutional neural network for training, the corresponding training parameters are trained, and the convolutional neural network with the training parameters is used as the detection of the ECG signal.
  • step S1 of the model it includes:
  • S102 Input a third ECG signal of a known signal rhythm type to a detection model of the ECG signal for verification, verify whether a signal rhythm type output by the detection model of the ECG signal and a signal rhythm type of the third ECG signal the same.
  • a training set and a test set are set, and the training set and the test set ratio may be set to 3:1; wherein the training set includes the first ECG signal, and the test set includes the third ECG signal, which are all The source of the third ECG signal is the same as the source of the first ECG signal, and is not described here.
  • the third ECG signal of the known signal rhythm type is input for calculation, and the predicted result (signal) of the output is determined. Whether the rhythm type is the same as the signal rhythm type of the third ECG signal; if the same, the detection model of the above ECG signal has a good training effect.
  • the second ECG signal of the unknown signal rhythm type is detected using the detection model of the ECG signal described above. In the verification process, the process of calculating the third ECG signal input into the detection model of the ECG signal is consistent with the specific implementation of the foregoing step S2, and details are not described herein.
  • the third ECG signal is input into the detection model to calculate a fusion matrix (the confusion matrix is also called an error matrix, which is a standard format for accuracy evaluation), and the detection model is obtained in the arrhythmia disease. Correct prediction of classification results.
  • a fusion matrix the confusion matrix is also called an error matrix, which is a standard format for accuracy evaluation
  • step S2 after detecting the abnormal signal rhythm type in the second ECG signal (ie, determining that the arrhythmia is abnormal), according to the arrhythmia signal rhythm type keyword/feature, go to the historical diagnosis database. Search/match similar/similar arrhythmia cases so that the doctor can refer to the diagnosis, analyze the similar cases, and analyze the arrhythmia disease conveniently.
  • the historical diagnosis database stores a large number of historical diagnostic records of the hospital, including patient information, cases, specific information of arrhythmia, etc.; the information of the patient having arrhythmia detected by the detection model in the present application is also deposited into In the database.
  • the detected patient information similar to/similar to the ECG signal is matched in a database of historical diagnoses prior to detection in the detection model of the second ECG signal of the patient to be detected being fed into the ECG signal,
  • the detected patient information includes the arrhythmia result of the historical diagnosis, and finally the result of the detection of the patient to be detected by the detection model is compared with the arrhythmia result of the detected patient to determine the difference, and if the difference is too large, the detection result is too large.
  • One of them may be misdiagnosed; it may be tested again or diagnosed by a specialist for correction.
  • a first ECG signal of a known signal rhythm type is input into a convolutional neural network for training, and training corresponding training parameters is obtained.
  • the convolutional neural network of the training parameter is used as a detection model of the ECG signal; the second ECG signal to be detected is input into the detection model of the ECG signal for calculation, and the signal rhythm type of the second ECG signal is output; the training is completed.
  • the ECG signal detection model can replace the medical expert, fully automatic detection of the signal rhythm type in the patient's ECG signal, saving a lot of manpower and material cost; at the same time, its accuracy in detecting the signal rhythm type in the patient's ECG signal can reach or exceed the expert. .
  • an apparatus for detecting an ECG signal including:
  • the training unit 10 is configured to input the first ECG signal of the known signal rhythm type into the convolutional neural network for training, train the corresponding training parameters, and obtain the convolutional neural network of the training parameters as the detection model of the ECG signal. ;
  • the ECG signal generally includes 14 kinds of signal rhythm types, 12 abnormal rhythms, one sinus rhythm and one noise rhythm, and the signal rhythm expresses state information of the ECG signal, when the ECG signal is detected as above
  • the signal rhythm type of the ECG signal becomes a scientific means of detecting whether the patient has a heart rhythm abnormality.
  • the signal rhythm type of the first ECG signal is known in advance.
  • the above convolutional neural network (Convolutional Neural Networks, CNN) can be a Sequence to sequence CNN model, which is a deep learning network for one-dimensional time series models, which is suitable for ECG time series data (ie ECG signal data); therefore, for ECG signals are more accurate and faster when training.
  • the source of the first ECG signal may be selected from the detected ECG signals in the history detection database, or may be temporarily collected and marked with a signal rhythm type. For example, in a specific embodiment, finding as many target patients as possible to wear a cardiac monitoring device to continuously monitor their heart rhythm for a number of days, collecting a plurality of ECG signals having a frequency of 200 Hz for n seconds (generally less than 60 seconds); After the ECG signal, the expert (doctor) divides the above ECG signal into regions, and the expert judges the signal rhythm type according to the ECG signal characteristics of each region, and uses the labeling tool to mark, that is, the specific signal rhythm type of the ECG signal of each region is marked. .
  • the ECG signal labeled with the signal rhythm type described above is used as the first ECG signal.
  • the training unit 10 inputs the first ECG signal into the convolutional neural network, and outputs an unknown training result according to the unknown training parameter, the training result is associated with the training parameter, and different training parameters result in different training.
  • the first ECG signal is input into the convolutional neural network, and we hope that the training result of the expected output is the signal rhythm type of the first ECG signal.
  • the signal rhythm type of the first ECG signal is known, and only the first ECG signal is input into the convolutional neural network for training, and the output training result is the signal rhythm type known to the first ECG signal.
  • the training parameter when the convolutional neural network detects the signal rhythm type of the ECG signal in the embodiment can be obtained.
  • the training parameters are input into the convolutional neural network to obtain the detection model of the ECG signal, and the detection model of the trained ECG signal can be used to detect the unknown ECG signal, predict the result, thereby replacing the manual detection, reducing the manpower, Material cost; and the detection speed is obviously improved, and the detection efficiency is improved.
  • the detecting unit 20 is configured to input a second ECG signal to be detected into the detection model of the ECG signal for calculation, and output a signal rhythm type of the second ECG signal.
  • the detection model of the ECG signal is the detection model obtained by the training unit 10 being trained.
  • the medical device can be used to collect the model.
  • the second ECG signal, the detecting unit 20 inputs the second ECG signal to be detected into the detection model of the ECG signal, and the detection model of the ECG signal outputs the prediction result of the second ECG signal, that is, the second prediction is predicted.
  • the type of signal rhythm of the ECG signal thereby determining whether the patient has an arrhythmia based on the predicted result.
  • the machine automatically detects the type of signal rhythm in the patient's ECG signal, thereby judging whether the patient has arrhythmia according to the signal rhythm type, improving the diagnosis speed of diagnosis of arrhythmia, saving a lot of manpower and material cost; meanwhile, the diagnostic accuracy can reach Even more than experts.
  • the apparatus for detecting an ECG signal further includes:
  • the normalization unit 101 is configured to perform normalization processing on the first ECG signal.
  • the normalization process is a commonly used method of convolutional neural networks, and is mainly used to normalize the first ECG signal to a specified range.
  • the normalization unit 101 normalizes the first ECG signal to a mean of 0 variance as an input to the convolutional neural network; when the convolutional neural network outputs, it can output a prediction of the first ECG signal every 1 s. As a result, all of the prediction results combine to correspond to the entire first ECG signal sequence.
  • the normalization process of the above-described normalization unit 101 is to subtract the mean of all ECG signal data for each second ECG signal data and divide by the standard deviation of all ECG signal data. It should be understood that, when the above-described detecting unit 20 detects the signal rhythm type of the second ECG signal, the second ECG signal is subjected to standardization processing in accordance with the same processing procedure.
  • the detecting unit 20 is specifically configured to:
  • the second ECG signal to be detected is input to an input layer in the detection model of the ECG signal, and is outputted through a 32-layer convolution layer and then output through the output layer to obtain a signal rhythm type of the second ECG signal.
  • the second ECG signal is input to the detection model of the ECG signal for calculation to output the signal rhythm type of the second ECG signal.
  • the convolutional neural network in the training unit 10 is the same as the network structure of the detection model of the ECG signal in this step, and the difference is only in the unknown of the training parameters. Therefore, the training process of the training unit 10 is consistent with the calculation process of the detecting unit 20, The loss function used is consistent, the optimization method of the network model is consistent, and the training process of the training unit 10 is not described again, which can refer to the specific calculation process of the detecting unit 20.
  • the detection model of the convolutional neural network and the ECG signal in this embodiment is a 34-layer convolutional neural network, and the network structure specifically includes an input layer, an output layer, and 16 residual blocks, wherein each residual block includes 2 Convolutional layer.
  • the size of each convolutional layer filter is: filter length is 16 x 1, filter number is 64k, k starts at 1, and increments by 1 every 4 residual blocks k.
  • filter length is 16 x 1
  • filter number is 64k
  • k starts at 1
  • the ECG signal feature in the network structure it is processed by a common method in the neural network, for example, every other residual block, and downsampled by a coefficient of 2.
  • the detecting unit 20 includes:
  • the second ECG signal is sequentially subjected to batch normalization, linear correction, and deep learning.
  • the loss function used in the detection model of the convolutional neural network and the ECG signal is a cross entropy loss function
  • the optimization method of the convolutional neural network is the Adam method
  • the cross entropy loss function is used to measure the volume.
  • the Adam method dynamically adjusts the learning rate for each parameter based on the first-order moment estimate and the second-order moment estimate of the gradient of each parameter according to the loss function. The learning rate will gradually decrease after Loss (loss function) no longer decreases. The reason why loss does not decrease is that the learning rate is too large, so the way to reduce the learning rate is to reduce the loss.
  • Optimized using the Adam method The optimization method can also use SGD (random steepest descent method), Momentum (momentum optimization) and other methods, but after experiment comparison, it is found that using Adam method is the best.
  • the apparatus for detecting an ECG signal further includes:
  • a verification unit 30 configured to input a third ECG signal of a known signal rhythm type to a detection model of the ECG signal, and verify whether a signal rhythm type output by the detection model of the ECG signal is related to the third ECG signal The signal rhythm type is the same.
  • a training set and a test set are set, and the training set and the test set ratio may be set to 3:1; wherein the training set includes the first ECG signal, and the test set includes the third ECG signal, which are all The source of the third ECG signal is the same as the source of the first ECG signal, and is not described here.
  • the verification unit 30 After the training unit 10 trains the detection model of the ECG signal, in order to verify the detection accuracy of the detection model of the ECG signal, the verification unit 30 inputs the third ECG signal of the known signal rhythm type for calculation, and determines the prediction of the output.
  • the result (signal rhythm type) is the same as the signal rhythm type of the third ECG signal; if the same, the detection model of the above ECG signal has a good training effect.
  • the detecting unit 20 detects the second ECG signal of the unknown signal rhythm type using the detection model of the ECG signal described above. In the verification process, the verification unit 30 inputs the third ECG signal into the detection model of the ECG signal for calculation, which is consistent with the specific implementation of the detection unit 20, and is not described herein again.
  • the verification unit 30 inputs the third ECG signal into the detection model for calculation to obtain a convergence.
  • Matrix confusion matrix is also called error matrix, which is a standard format for accuracy evaluation).
  • the detection model has obtained correct prediction classification results in arrhythmia diseases.
  • the history diagnosis database is used. Search/match similar/similar arrhythmia cases so that the doctor can refer to the diagnosis, analyze the similar cases, and analyze the arrhythmia disease conveniently.
  • the historical diagnosis database stores a large number of historical diagnostic records of the hospital, including patient information, cases, specific information of arrhythmia, etc.; the information of the patient having arrhythmia detected by the detection model in the present application is also deposited into In the database.
  • the detecting unit 20 matches the detected patient similar to/similar to the ECG signal in the database of historical diagnoses before detecting the second ECG signal of the patient to be detected fed into the detection model of the ECG signal.
  • Information, the detected patient information includes arrhythmia results of historical diagnosis, and finally the result of detecting the patient to be detected by using the detection model is compared with the arrhythmia result of the detected patient to determine the difference, and if the detection result is different If it is large, one of them may be misdiagnosed; it may be tested again or diagnosed by a specialist doctor for correction.
  • the apparatus for detecting an ECG signal inputs a first ECG signal of a known signal rhythm type into a convolutional neural network for training, and training corresponding training parameters, which will result in
  • the convolutional neural network of the training parameter is used as a detection model of the ECG signal;
  • the second ECG signal to be detected is input into the detection model of the ECG signal for calculation, and the signal rhythm type of the second ECG signal is output;
  • the training is completed.
  • the ECG signal detection model can replace the medical expert's automatic detection of the signal rhythm type in the patient's ECG signal, saving a lot of manpower and material cost; at the same time, the accuracy of detecting the signal rhythm type in the patient's ECG signal can reach or exceed the expert.
  • the computer device may be a server, and its internal structure may be as shown in FIG. 6.
  • the computer device includes a processor, memory, network interface, and database connected by a system bus. Among them, the computer designed processor is used to provide calculation and control capabilities.
  • the memory of the computer device includes a non-volatile storage medium, an internal memory.
  • the non-volatile storage medium stores an operating system, computer readable instructions, and a database.
  • the internal memory provides an environment for operation of an operating system and computer readable instructions in a non-volatile storage medium.
  • the database of the computer device is used to store data such as a convolutional neural network.
  • the network interface of the computer device is used to communicate with an external terminal via a network connection.
  • the computer readable instructions are executed by a processor to implement a method of detecting an ECG signal.
  • the above processor executes the step of detecting the ECG signal by inputting the first ECG signal of the known signal rhythm type into the convolutional neural network for training, training the corresponding training parameter, and obtaining the convolutional nerve of the training parameter
  • the network acts as a detection model for ECG signals
  • the method before the step of the processor inputting the first ECG signal of the known signal rhythm type into the convolutional neural network for training and training the corresponding training parameter, the method includes:
  • the processor inputs the second ECG signal to be detected into the detection model of the ECG signal for calculation, and outputs a signal rhythm type of the second ECG signal, including:
  • the second ECG signal to be detected is input to an input layer in the detection model of the ECG signal, and is outputted through a 32-layer convolution layer and then output through the output layer to obtain a signal rhythm type of the second ECG signal.
  • the loss function used by the convolutional neural network is a cross entropy loss function
  • the optimization method of the convolutional neural network is the Adam method
  • the processor includes:
  • the processor inputs the first ECG signal of the known signal rhythm type into the convolutional neural network for training, trains the corresponding training parameters, and derives the convolutional neural network of the training parameters as the ECG signal.
  • the steps of the test model include:
  • a third ECG signal of a known signal rhythm type is input to a detection model of the ECG signal for verification, and it is verified whether the signal rhythm type output by the detection model of the ECG signal is the same as the signal rhythm type of the third ECG signal.
  • the processor inputs the first ECG signal of the known signal rhythm type into the convolutional neural network for training, and the corresponding training parameters are trained, including:
  • a first ECG signal of a known signal rhythm type is input into the convolutional neural network for training, and the output training result is a signal rhythm type known to the first ECG signal to derive a corresponding training parameter.
  • FIG. 6 is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation of the computer device to which the solution of the present application is applied.
  • An embodiment of the present application further provides a computer non-volatile readable storage medium, where computer readable instructions are stored, and when the computer readable instructions are executed by the processor, a method for detecting an ECG signal is implemented, specifically: The first ECG signal of the known signal rhythm type is input into the convolutional neural network for training, and the corresponding training parameters are trained, and the convolutional neural network of the training parameters is obtained as the detection model of the ECG signal;
  • the method before the step of the processor inputting the first ECG signal of the known signal rhythm type into the convolutional neural network for training and training the corresponding training parameter, the method includes:
  • the processor inputs the second ECG signal to be detected into the detection model of the ECG signal for calculation, and outputs a signal rhythm type of the second ECG signal, including:
  • the second ECG signal to be detected is input to an input layer in the detection model of the ECG signal, and is outputted through a 32-layer convolution layer and then output through the output layer to obtain a signal rhythm type of the second ECG signal.
  • the loss function used by the convolutional neural network is a cross entropy loss function
  • the optimization method of the convolutional neural network is the Adam method
  • the processor includes:
  • the processor inputs the first ECG signal of the known signal rhythm type into the convolutional neural network for training, trains the corresponding training parameters, and derives the convolutional neural network of the training parameters as the ECG signal.
  • the steps of the test model include:
  • a third ECG signal of a known signal rhythm type is input to a detection model of the ECG signal for verification, and it is verified whether the signal rhythm type output by the detection model of the ECG signal is the same as the signal rhythm type of the third ECG signal.
  • the processor inputs the first ECG signal of the known signal rhythm type into the convolutional neural network for training, and the corresponding training parameters are trained, including:
  • a first ECG signal of a known signal rhythm type is input into the convolutional neural network for training, and the output training result is a signal rhythm type known to the first ECG signal to derive a corresponding training parameter.
  • the method, device, computer device and storage medium for detecting ECG signals input the first ECG signal of the known signal rhythm type into the convolutional neural network for training, training Corresponding training parameters, a convolutional neural network of the training parameters is obtained as a detection model of the ECG signal; a second ECG signal to be detected is input into the detection model of the ECG signal for calculation, and the second ECG is obtained by the output.
  • the signal rhythm type of the signal; the trained ECG signal detection model can replace the medical expert's automatic detection of the signal rhythm type in the patient's ECG signal, saving a lot of manpower and material cost; meanwhile, it detects the signal rhythm type in the patient ECG signal. Accuracy can reach or exceed experts.
  • Non-volatile memory can include read only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory.
  • Volatile memory can include random access memory (RAM) or external cache memory.
  • RAM is available in a variety of formats, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual-speed SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronization.
  • SRAM static RAM
  • DRAM dynamic RAM
  • SDRAM synchronous DRAM
  • SSRSDRAM dual-speed SDRAM
  • ESDRAM enhanced SDRAM
  • SLDRAM Link (Synchlink) DRAM
  • SLDRAM Memory Bus
  • RDRAM Direct RAM
  • DRAM Direct Memory Bus Dynamic RAM
  • RDRAM Memory Bus Dynamic RAM

Landscapes

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

Abstract

一种ECG信号的检测方法、装置、计算机设备和存储介质。方法包括：将已知信号节奏类型的第一ECG信号输入至卷积神经网络中进行训练，训练出对应的训练参数，将得出训练参数的卷积神经网络作为ECG信号的检测模型（S1）；将待检测的第二ECG信号输入至所述ECG信号的检测模型中进行计算，输出得到所述第二ECG信号的信号节奏类型（S2）。通过该方法，提升了检测信号节奏类型的准确率。

Description

ECG信号的检测方法、装置、计算机设备和存储介质
本申请要求于2018年4月20日提交中国专利局、申请号为201810362786X，发明名称为“ECG信号的检测方法、装置、计算机设备和存储介质”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
本申请涉及计算机技术领域，特别涉及一种ECG信号的检测方法、装置、计算机设备和存储介质。
 背景技术
心律失常指的是心脏电传导系统异常所引起的各种症状，包含心跳不规则、过快、或过慢的表现总称，是心血管疾病中重要的一组疾病。它可单独发病，亦可与其它心血管疾病伴发。心律失常的患者比例常年位于各种心血管疾病的前两位，严重危害我国居民的健康。因此，快速、准确的心律失常检测技术显得极为重要。
由于心电图信号（ECG信号，Electrocardiogram）的多样性以及噪声的存在，机器检测存在较大的误差，目前，主流的心律失常检测技术是通过专家对ECG信号进行人工诊断，其诊断速度低、诊断效率低，其耗费大量的人力和物力成本。
 技术问题
本申请的主要目的为提供一种ECG信号的检测方法、装置、计算机设备和存储介质，克服人工诊断心律失常时诊断速度低、诊断效率低的缺陷。
 技术解决方案
为实现上述目的，本申请提供一种ECG信号的检测方法，包括以下步骤：
将已知信号节奏类型的第一ECG信号输入至卷积神经网络中进行训练，训练出对应的训练参数，将得出训练参数的卷积神经网络作为ECG信号的检测模型；
将待检测的第二ECG信号输入至所述ECG信号的检测模型中进行计算，输出得到所述第二ECG信号的信号节奏类型。
本申请还提供了一种ECG信号的检测装置，包括：
训练单元，用于将已知信号节奏类型的第一ECG信号输入至卷积神经网络中进行训练，训练出对应的训练参数，将得出训练参数的卷积神经网络作为ECG信号的检测模型；
检测单元，用于将待检测的第二ECG信号输入至所述ECG信号的检测模型中进行计算，输出得到所述第二ECG信号的信号节奏类型。
本申请还提供一种计算机设备，包括存储器和处理器，所述存储器中存储有计算机可读指令，所述处理器执行所述计算机可读指令时实现上述方法的步骤。
本申请还提供一种计算机非易失性可读存储介质，其上存储有计算机可读指令，所述计算机可读指令被处理器执行时实现上述方法的步骤。
 有益效果
本申请中提供的ECG信号的检测方法、装置、计算机设备和存储介质，将已知信号节奏类型的第一ECG信号输入至卷积神经网络中进行训练，训练出对应的训练参数，将得出训练参数的卷积神经网络作为ECG信号的检测模型；将待检测的第二ECG信号输入至所述ECG信号的检测模型中进行计算，输出得到所述第二ECG信号的信号节奏类型；训练完成的ECG信号检测模型可替代医学专家全自动检测患者ECG信号中的信号节奏类型，节约大量的人力和物力成本；同时，其检测患者ECG信号中的信号节奏类型的准确率可以达到甚至超过专家。
 附图说明
图1 是本申请一实施例中ECG信号的检测方法步骤示意图；
图2 是本申请另一实施例中ECG信号的检测方法步骤示意图；
图3 是本申请一实施例中ECG信号的检测装置结构框图；
图4 是本申请另一实施例中ECG信号的检测装置结构框图；
图5 是本申请又一实施例中ECG信号的检测装置结构框图；
图6 为本申请一实施例的计算机设备的结构示意框图。
 本发明的最佳实施方式
参照图1，本申请实施例中提供了一种ECG信号的检测方法，包括以下步骤：
步骤S1，将已知信号节奏类型的第一ECG信号输入至卷积神经网络中进行训练，训练出对应的训练参数，将得出训练参数的卷积神经网络作为ECG信号的检测模型；
在本步骤S1中，ECG信号通常包括14种信号节奏类型，其中12种异常节奏，一种窦节奏以及一种噪声节奏，信号节奏表达的是ECG信号的状态信息，当检测到ECG信号为上述12种异常节奏的任意一种信号节奏类型时，则可以判定该ECG信号的来源患者为心律异常。因此，检测ECG信号的信号节奏类型成为检测患者是否患有心律异常的科学手段。
本实施例中，第一ECG信号的信号节奏类型为预先已知。上述卷积神经网络（Convolutional Neural Networks，CNN）具体可以为Sequence to sequence CNN模型，该模型是一种深度学习网络，用于一维时间序列模型，正好适用于心电图时间序列数据（即心电图信号数据）；因此，用于对心电图信号进行训练时更加准确、快速。
具体地，上述第一ECG信号的来源可以是从历史检测数据库中已经检测好的ECG信号中选取的，也可以是通过临时采集并标记好信号节奏类型。例如，在具体的一个实施例中，寻找尽量多的目标病人佩戴心脏监测设备连续对其心律监测多日，收集多个持续n秒（一般小于60秒）频率为200Hz的ECG信号；收集到上述ECG信号之后，通过专家（医生）对上述ECG信号划分区域，专家根据每个区域的ECG信号特征判断信号节奏类型，并使用标注工具进行标注，即标注每个区域的ECG信号的具体信号节奏类型。将上述标注有信号节奏类型的ECG信号作为第一ECG信号。
具体地，将上述第一ECG信号输入至卷积神经网络中，根据未知的训练参数训练输出未知的训练结果，该训练结果与训练参数相关联，不同的训练参数得出不同的训练结果；训练时，将上述第一ECG信号输入至卷积神经网络中，我们希望得到的预期输出的训练结果即第一ECG信号的信号节奏类型。而训练之前已经知道第一ECG信号的信号节奏类型，只需将第一ECG信号输入至卷积神经网络中进行训练，并使得输出的训练结果为所述第一ECG信号已知的信号节奏类型，或者使得输出的训练结果趋近于所述第一ECG信号已知的信号节奏类型；则可以得出卷积神经网络用在本实施例中对ECG信号的信号节奏类型进行检测时的训练参数。将训练参数输入至卷积神经网络中则得到ECG信号的检测模型，该训练完成的ECG信号的检测模型则可以用于对未知的ECG信号进行检测，预测结果，从而替代人工检测，降低人力、物力成本；且检测速度明显提升，检测效率得到提高。
步骤S2，将待检测的第二ECG信号输入至所述ECG信号的检测模型中进行计算，输出得到所述第二ECG信号的信号节奏类型。
在本步骤S2中，所述ECG信号的检测模型即为上述步骤S1中训练完成得到的检测模型，此时，若有新的患者需要检测是否患有心律失常，则可以通过医疗设备采集其第二ECG信号，或者接收到医疗设备采集的第二ECG信号时，将待检测的第二ECG信号输入至上述ECG信号的检测模型中计算，该ECG信号的检测模型输出的则是对第二ECG信号的预测结果，即预测出第二ECG信号的信号节奏类型，从而根据预测结果判断患者是否患有心律失常。通过机器全自动检测患者ECG信号中的信号节奏类型，从而根据信号节奏类型判断患者是否存在心律失常，提高诊断心律失常的诊断速度，节约大量的人力和物力成本；同时，其诊断准确率可以达到甚至超过专家。
参照图2，在一实施例中，上述将已知信号节奏类型的第一ECG信号输入至卷积神经网络中进行训练，训练出对应的训练参数的步骤之前，包括：
步骤S101，对所述第一ECG信号进行标准化处理。
在本步骤中，标准化处理为卷积神经网络常用方法，主要用于将第一ECG信号标准化至指定范围内。本实施例中，将第一ECG信号标准化至均值为0方差为1，作为卷积神经网络的输入；卷积神经网络输出时，则可以每1s输出一个对第一ECG信号的预测结果，所有的预测结果合起来对应整个第一ECG信号序列。具体地，上述标准化处理过程为对每一秒ECG信号数据减去所有ECG信号数据的均值再除以所有ECG信号数据的标准差。应当理解的是，在上述步骤S2中检测第二ECG信号的信号节奏类型时，对第二ECG信号按照同样的处理过程进行标准化处理。
在一实施例中，上述将待检测的第二ECG信号输入至所述ECG信号的检测模型中进行计算，输出得到所述第二ECG信号的信号节奏类型的步骤S2，包括：
将待检测的第二ECG信号输入至所述ECG信号的检测模型中的输入层，经过32层卷积层卷积后再经输出层输出，得到所述第二ECG信号的信号节奏类型。
本实施例中，将第二ECG信号输入至ECG信号的检测模型中进行计算，以输出第二ECG信号的信号节奏类型。上述步骤S1中的卷积神经网络与本步骤中的ECG信号的检测模型的网络结构相同，区别仅在于训练参数的未知，因此，步骤S1中的训练过程与本步骤S2中的计算过程一致，使用的损失函数一致，网络模型的优化方法一致，对步骤S1中的训练过程则不再进行赘述，其可参照步骤S2中的具体计算过程。
本实施例中的卷积神经网络、ECG信号的检测模型为34层卷积神经网络，网络结构中具体包括一个输入层、一个输出层以及16个残差块，其中每个残差块包括2个卷积层。每个卷积层滤波（filter）的尺寸为：滤波长度为16 x 1，滤波个数为64k，k开始为1，每隔4个残差块k增加1。该网络结构中为了减小ECG信号特征（feature）的尺寸，则采用神经网络中常用方法进行处理，例如每隔一个残差块，以系数2进行下采样。
为了加速网络参数优化收敛，每个卷积层卷积之前需要依次进行批标准化（Batch Normalization）、线性修正（Relu，Rectified Linear Unit, 线性整流函数,又称线性修正单元）和深度学习（Dropout）。
因此，所述经过32层卷积层卷积后再经输出层输出的步骤之前，包括：
对所述第二ECG信号依次进行批标准化、线性修正以及深度学习。
在本实施例中，上述卷积神经网络、ECG信号的检测模型中使用的损失函数为交叉熵损失函数，所述卷积神经网络的优化方法为Adam方法；交叉熵损失函数是用来衡量卷积神经网络（CNN）的预测值与实际值的一种方式。与二次代价函数相比，它能更有效地促进CNN的训练。Adam方法是根据损失函数对每个参数的梯度的一阶矩估计和二阶矩估计动态调整针对于每个参数的学习速率。学习速率在Loss（损失函数）不再减小之后会逐渐减小，loss不减小的原因是learning rate（学习速率）过大，所以常用减小学习速率的方式是使loss继续减小，因此使用Adam方法进行优化。优化方法也可以用SGD（随机最速下降法）、Momentum（动量优化）等方法，但是经过实验对比发现使用Adam方法效果最好。
在一实施例中，上述将已知信号节奏类型的第一ECG信号输入至卷积神经网络中进行训练，训练出对应的训练参数，将得出训练参数的卷积神经网络作为ECG信号的检测模型的步骤S1之后，包括：
S102，将已知信号节奏类型的第三ECG信号输入至所述ECG信号的检测模型进行验证，验证所述ECG信号的检测模型输出的信号节奏类型是否与所述第三ECG信号的信号节奏类型相同。
本实施例中，设置有训练集以及测试集，训练集与测试集比例可设置为3:1；其中，训练集包括上述第一ECG信号，测试集包括上述第三ECG信号，其均是已知信号节奏类型的ECG信号，上述第三ECG信号的来源与第一ECG信号的来源相同，在此不再进行赘述。
在上述步骤S1中训练得出ECG信号的检测模型之后，为了验证该ECG信号的检测模型的检测准确性，将已知信号节奏类型的第三ECG信号输入进行计算，判断输出的预测结果（信号节奏类型）是否与第三ECG信号的信号节奏类型相同；若相同，则上述ECG信号的检测模型训练效果好。验证之后，再使用上述ECG信号的检测模型对未知信号节奏类型的第二ECG信号进行检测。该验证过程中，第三ECG信号输入至ECG信号的检测模型中进行计算的过程与上述步骤S2的具体实现一致，在此不再进行赘述。在一具体实施例中，将第三ECG信号输入检测模型中进行计算得到confusion matrix（混淆矩阵也称误差矩阵，是表示精度评价的一种标准格式），检测模型在心律失常疾病上均得到了正确的预测分类结果。
在另一实施例中，上述步骤S2中，检测第二ECG信号中的异常信号节奏类型后（即判断为心律失常），根据该心律失常的信号节奏类型关键词/特征，去历史诊断的数据库中检索/匹配出相似/相近的心律失常案例，以便医生参考进行诊断，对相似案例进行分析，还可以方便对该心律失常疾病进行分析。历史诊断的数据库中存储有医院大量历史诊断记录，其中包括患者的信息、病例、心律失常的具体信息等；利用本申请中的检测模型检测出的患者患有心律失常的信息同样被存入至数据库中。
在又一实施例中，在将待检测患者的第二ECG信号喂入ECG信号的检测模型中检测之前，在历史诊断的数据库中匹配与该ECG信号相似/相近的已检测的患者信息，该已检测患者信息中包括历史诊断的心律失常结果，最后利用检测模型对待检测患者进行检测的结果出来之后与已检测患者的心律失常结果进行对比，判断差异性，检测结果若差异性过大，则可能其中一个出现误诊；可再次进行检测，或者通过专家医生进行诊断，以便进行纠正。
综上所述，为本申请实施例中提供的ECG信号的检测方法，将已知信号节奏类型的第一ECG信号输入至卷积神经网络中进行训练，训练出对应的训练参数，将得出训练参数的卷积神经网络作为ECG信号的检测模型；将待检测的第二ECG信号输入至所述ECG信号的检测模型中进行计算，输出得到所述第二ECG信号的信号节奏类型；训练完成的ECG信号检测模型可替代医学专家，全自动检测患者ECG信号中的信号节奏类型，节约大量的人力和物力成本；同时，其检测患者ECG信号中的信号节奏类型的准确率可以达到甚至超过专家。
参照图3，本申请实施例中还提供了一种ECG信号的检测装置，包括：
训练单元10，用于将已知信号节奏类型的第一ECG信号输入至卷积神经网络中进行训练，训练出对应的训练参数，将得出训练参数的卷积神经网络作为ECG信号的检测模型；
在本实施例中，ECG信号通常包括14种信号节奏类型，其中12种异常节奏，一种窦节奏以及一种噪声节奏，信号节奏表达的是ECG信号的状态信息，当检测到ECG信号为上述12种异常节奏的任意一种信号节奏类型时，则可以判定该ECG信号的来源患者为心律异常。因此，检测ECG信号的信号节奏类型成为检测患者是否患有心律异常的科学手段。
本实施例中，第一ECG信号的信号节奏类型为预先已知。上述卷积神经网络（Convolutional Neural Networks，CNN）具体可以为Sequence to sequence CNN模型，该模型是一种深度学习网络，用于一维时间序列模型，正好适用于心电图时间序列数据（即心电图信号数据）；因此，用于对心电图信号进行训练时更加准确、快速。
具体地，上述第一ECG信号的来源可以是从历史检测数据库中已经检测好的ECG信号中选取的，也可以是通过临时采集并标记好信号节奏类型。例如，在具体的一个实施例中，寻找尽量多的目标病人佩戴心脏监测设备连续对其心律监测多日，收集多个持续n秒（一般小于60秒）频率为200Hz的ECG信号；收集到上述ECG信号之后，通过专家（医生）对上述ECG信号划分区域，专家根据每个区域的ECG信号特征判断信号节奏类型，并使用标注工具进行标注，即标注每个区域的ECG信号的具体信号节奏类型。将上述标注有信号节奏类型的ECG信号作为第一ECG信号。
具体地，训练单元10将上述第一ECG信号输入至卷积神经网络中，根据未知的训练参数训练输出未知的训练结果，该训练结果与训练参数相关联，不同的训练参数得出不同的训练结果；训练时，将上述第一ECG信号输入至卷积神经网络中，我们希望得到的预期输出的训练结果即第一ECG信号的信号节奏类型。而训练之前已经知道第一ECG信号的信号节奏类型，只需将第一ECG信号输入至卷积神经网络中进行训练，并使得输出的训练结果为所述第一ECG信号已知的信号节奏类型，或者使得输出的训练结果趋近于所述第一ECG信号已知的信号节奏类型，则可以得出卷积神经网络用在本实施例中对ECG信号的信号节奏类型进行检测时的训练参数。将训练参数输入至卷积神经网络中则得到ECG信号的检测模型，该训练完成的ECG信号的检测模型则可以用于对未知的ECG信号进行检测，预测结果，从而替代人工检测，降低人力、物力成本；且检测速度明显提升，检测效率得到提高。
检测单元20，用于将待检测的第二ECG信号输入至所述ECG信号的检测模型中进行计算，输出得到所述第二ECG信号的信号节奏类型。
在本实施例中，所述ECG信号的检测模型即为上述训练单元10训练完成得到的检测模型，此时，若有新的患者需要检测是否患有心律失常，则可以使用医疗设备采集其第二ECG信号，检测单元20将待检测的第二ECG信号输入至上述ECG信号的检测模型中计算，该ECG信号的检测模型输出的则是对第二ECG信号的预测结果，即预测出第二ECG信号的信号节奏类型，从而根据预测结果判断患者是否患有心律失常。通过机器全自动检测患者ECG信号中的信号节奏类型，从而根据信号节奏类型判断患者是否存在心律失常，提高诊断心律失常的诊断速度，节约大量的人力和物力成本；同时，其诊断准确率可以达到甚至超过专家。
参照图4，在一实施例中，上述ECG信号的检测装置还包括：
标准化单元101，用于对所述第一ECG信号进行标准化处理。
在本实施例中，标准化处理为卷积神经网络常用方法，主要用于将第一ECG信号标准化至指定范围内。本实施例中，标准化单元101将第一ECG信号标准化至均值为0方差为1，作为卷积神经网络的输入；卷积神经网络输出时，则可以每1s输出一个对第一ECG信号的预测结果，所有的预测结果合起来对应整个第一ECG信号序列。具体地，上述标准化单元101的标准化处理过程为对每一秒ECG信号数据减去所有ECG信号数据的均值再除以所有ECG信号数据的标准差。应当理解的是，在上述检测单元20检测第二ECG信号的信号节奏类型时，对第二ECG信号按照同样的处理过程进行标准化处理。
在一实施例中，上述检测单元20具体用于：
将待检测的第二ECG信号输入至所述ECG信号的检测模型中的输入层，经过32层卷积层卷积后再经输出层输出，得到所述第二ECG信号的信号节奏类型。
本实施例中，将第二ECG信号输入至ECG信号的检测模型中进行计算，以输出第二ECG信号的信号节奏类型。上述训练单元10中的卷积神经网络与本步骤中的ECG信号的检测模型的网络结构相同，区别仅在于训练参数的未知，因此，训练单元10的训练过程与检测单元20的计算过程一致，使用的损失函数一致，网络模型的优化方法一致，对训练单元10的训练过程则不再进行赘述，其可参照检测单元20的具体计算过程。
本实施例中的卷积神经网络、ECG信号的检测模型为34层卷积神经网络，网络结构中具体包括一个输入层、一个输出层以及16个残差块，其中每个残差块包括2个卷积层。每个卷积层滤波（filter）的尺寸为：滤波长度为16 x 1，滤波个数为64k，k开始为1，每隔4个残差块k增加1。该网络结构中为了减小ECG信号特征（feature）的尺寸，则采用神经网络中常用方法进行处理，例如每隔一个残差块，以系数2进行下采样。
为了加速网络参数优化收敛，每个卷积层卷积之前需要依次进行批标准化（Batch Normalization）、线性修正（Relu，Rectified Linear Unit, 线性整流函数,又称线性修正单元）和深度学习（Dropout）。
因此，所述检测单元20经过32层卷积层卷积后再经输出层输出之前，包括：
将第二ECG信号依次进行批标准化、线性修正以及深度学习。
在本实施例中，上述卷积神经网络、ECG信号的检测模型中使用的损失函数为交叉熵损失函数，所述卷积神经网络的优化方法为Adam方法；交叉熵损失函数是用来衡量卷积神经网络（CNN）的预测值与实际值的一种方式。与二次代价函数相比，它能更有效地促进CNN的训练。Adam方法是根据损失函数对每个参数的梯度的一阶矩估计和二阶矩估计动态调整针对于每个参数的学习速率。学习速率在Loss（损失函数）不再减小之后会逐渐减小，loss不减小的原因是learning rate（学习速率）过大，所以常用减小学习速率的方式是使loss继续减小，因此使用Adam方法进行优化。优化方法也可以用SGD（随机最速下降法）、Momentum（动量优化）等方法，但是经过实验对比发现使用Adam方法效果最好。
参照图5，在一实施例中，上述ECG信号的检测装置还包括：
验证单元30，用于将已知信号节奏类型的第三ECG信号输入至所述ECG信号的检测模型进行验证，验证所述ECG信号的检测模型输出的信号节奏类型是否与所述第三ECG信号的信号节奏类型相同。
本实施例中，设置有训练集以及测试集，训练集与测试集比例可设置为3:1；其中，训练集包括上述第一ECG信号，测试集包括上述第三ECG信号，其均是已知信号节奏类型的ECG信号，上述第三ECG信号的来源与第一ECG信号的来源相同，在此不再进行赘述。
在上述训练单元10训练得出ECG信号的检测模型之后，为了验证该ECG信号的检测模型的检测准确性，验证单元30将已知信号节奏类型的第三ECG信号输入进行计算，判断输出的预测结果（信号节奏类型）是否与第三ECG信号的信号节奏类型相同；若相同，则上述ECG信号的检测模型训练效果好。验证单元30验证之后，检测单元20再使用上述ECG信号的检测模型对未知信号节奏类型的第二ECG信号进行检测。该验证过程中，验证单元30将第三ECG信号输入至ECG信号的检测模型中进行计算的过程与上述检测单元20的具体实现一致，在此不再进行赘述。在一具体实施例中，验证单元30将第三ECG信号输入检测模型中进行计算得到confusion matrix（混淆矩阵也称误差矩阵，是表示精度评价的一种标准格式），检测模型在心律失常疾病上均得到了正确的预测分类结果。
在另一实施例中，上述检测单元20检测出第二ECG信号中的异常信号节奏类型后（即判断为心律失常），根据该心律失常的信号节奏类型关键词/特征，去历史诊断的数据库中检索/匹配出相似/相近的心律失常案例，以便医生参考进行诊断，对相似案例进行分析，还可以方便对该心律失常疾病进行分析。历史诊断的数据库中存储有医院大量历史诊断记录，其中包括患者的信息、病例、心律失常的具体信息等；利用本申请中的检测模型检测出的患者患有心律失常的信息同样被存入至数据库中。
在又一实施例中，检测单元20在将待检测患者的第二ECG信号喂入ECG信号的检测模型中检测之前，在历史诊断的数据库中匹配与该ECG信号相似/相近的已检测的患者信息，该已检测患者信息中包括历史诊断的心律失常结果，最后利用检测模型对待检测患者进行检测的结果出来之后与已检测患者的心律失常结果进行对比，判断差异性，检测结果若差异性过大，则可能其中一个出现误诊；可再次进行检测，或者通过专家医生进行诊断，以便进行纠正。
综上所述，为本申请实施例中提供的ECG信号的检测装置，将已知信号节奏类型的第一ECG信号输入至卷积神经网络中进行训练，训练出对应的训练参数，将得出训练参数的卷积神经网络作为ECG信号的检测模型；将待检测的第二ECG信号输入至所述ECG信号的检测模型中进行计算，输出得到所述第二ECG信号的信号节奏类型；训练完成的ECG信号检测模型可替代医学专家全自动检测患者ECG信号中的信号节奏类型，节约大量的人力和物力成本；同时，其检测患者ECG信号中的信号节奏类型的准确率可以达到甚至超过专家。
参照图6，本申请实施例中还提供一种计算机设备，该计算机设备可以是服务器，其内部结构可以如图6所示。该计算机设备包括通过系统总线连接的处理器、存储器、网络接口和数据库。其中，该计算机设计的处理器用于提供计算和控制能力。该计算机设备的存储器包括非易失性存储介质、内存储器。该非易失性存储介质存储有操作系统、计算机可读指令和数据库。该内存储器为非易失性存储介质中的操作系统和计算机可读指令的运行提供环境。该计算机设备的数据库用于存储卷积神经网络等数据。该计算机设备的网络接口用于与外部的终端通过网络连接通信。该计算机可读指令被处理器执行时以实现一种ECG信号的检测方法。
上述处理器执行上述ECG信号的检测方法的步骤：将已知信号节奏类型的第一ECG信号输入至卷积神经网络中进行训练，训练出对应的训练参数，将得出训练参数的卷积神经网络作为ECG信号的检测模型；
将待检测的第二ECG信号输入至所述ECG信号的检测模型中进行计算，输出得到所述第二ECG信号的信号节奏类型。
在一实施例中，上述处理器将已知信号节奏类型的第一ECG信号输入至卷积神经网络中进行训练，训练出对应的训练参数的步骤之前，包括：
对所述第一ECG信号进行标准化处理。
在一实施例中，上述处理器将待检测的第二ECG信号输入至所述ECG信号的检测模型中进行计算，输出得到所述第二ECG信号的信号节奏类型的步骤，包括：
将待检测的第二ECG信号输入至所述ECG信号的检测模型中的输入层，经过32层卷积层卷积后再经输出层输出，得到所述第二ECG信号的信号节奏类型。
在一实施例中，上述卷积神经网络使用的损失函数为交叉熵损失函数，所述卷积神经网络的优化方法为Adam方法。
在一实施例中，上述处理器经过32层卷积层卷积后再经输出层输出的步骤之前，包括：
对所述第二ECG信号依次进行批标准化、线性修正以及深度学习。
在一实施例中，上述处理器将已知信号节奏类型的第一ECG信号输入至卷积神经网络中进行训练，训练出对应的训练参数，将得出训练参数的卷积神经网络作为ECG信号的检测模型的步骤之后，包括：
将已知信号节奏类型的第三ECG信号输入至所述ECG信号的检测模型进行验证，验证所述ECG信号的检测模型输出的信号节奏类型是否与所述第三ECG信号的信号节奏类型相同。
在一实施例中，上述处理器将已知信号节奏类型的第一ECG信号输入至卷积神经网络中进行训练，训练出对应的训练参数的步骤，包括：
将已知信号节奏类型的第一ECG信号输入至卷积神经网络中进行训练，并使得输出的训练结果为所述第一ECG信号已知的信号节奏类型，以得出对应的训练参数。
本领域技术人员可以理解，图6中示出的结构，仅仅是与本申请方案相关的部分结构的框图，并不构成对本申请方案所应用于其上的计算机设备的限定。
本申请一实施例还提供一种计算机非易失性可读存储介质，其上存储有计算机可读指令，计算机可读指令被处理器执行时实现一种ECG信号的检测方法，具体为：将已知信号节奏类型的第一ECG信号输入至卷积神经网络中进行训练，训练出对应的训练参数，将得出训练参数的卷积神经网络作为ECG信号的检测模型；
将待检测的第二ECG信号输入至所述ECG信号的检测模型中进行计算，输出得到所述第二ECG信号的信号节奏类型。
在一实施例中，上述处理器将已知信号节奏类型的第一ECG信号输入至卷积神经网络中进行训练，训练出对应的训练参数的步骤之前，包括：
对所述第一ECG信号进行标准化处理。
在一实施例中，上述处理器将待检测的第二ECG信号输入至所述ECG信号的检测模型中进行计算，输出得到所述第二ECG信号的信号节奏类型的步骤，包括：
将待检测的第二ECG信号输入至所述ECG信号的检测模型中的输入层，经过32层卷积层卷积后再经输出层输出，得到所述第二ECG信号的信号节奏类型。
在一实施例中，上述卷积神经网络使用的损失函数为交叉熵损失函数，所述卷积神经网络的优化方法为Adam方法。
在一实施例中，上述处理器经过32层卷积层卷积后再经输出层输出的步骤之前，包括：
对所述第二ECG信号依次进行批标准化、线性修正以及深度学习。
在一实施例中，上述处理器将已知信号节奏类型的第一ECG信号输入至卷积神经网络中进行训练，训练出对应的训练参数，将得出训练参数的卷积神经网络作为ECG信号的检测模型的步骤之后，包括：
将已知信号节奏类型的第三ECG信号输入至所述ECG信号的检测模型进行验证，验证所述ECG信号的检测模型输出的信号节奏类型是否与所述第三ECG信号的信号节奏类型相同。
在一实施例中，上述处理器将已知信号节奏类型的第一ECG信号输入至卷积神经网络中进行训练，训练出对应的训练参数的步骤，包括：
将已知信号节奏类型的第一ECG信号输入至卷积神经网络中进行训练，并使得输出的训练结果为所述第一ECG信号已知的信号节奏类型，以得出对应的训练参数。
综上所述，为本申请实施例中提供的ECG信号的检测方法、装置、计算机设备和存储介质，将已知信号节奏类型的第一ECG信号输入至卷积神经网络中进行训练，训练出对应的训练参数，将得出训练参数的卷积神经网络作为ECG信号的检测模型；将待检测的第二ECG信号输入至所述ECG信号的检测模型中进行计算，输出得到所述第二ECG信号的信号节奏类型；训练完成的ECG信号检测模型可替代医学专家全自动检测患者ECG信号中的信号节奏类型，节约大量的人力和物力成本；同时，其检测患者ECG信号中的信号节奏类型的准确率可以达到甚至超过专家。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机可读指令来指令相关的硬件来完成，所述的计算机可读指令可存储与一非易失性计算机可读取存储介质中，该计算机可读指令在执行时，可包括如上述各方法的实施例的流程。其中，本申请所提供的和实施例中所使用的对存储器、存储、数据库或其它介质的任何引用，均可包括非易失性和/或易失性存储器。非易失性存储器可以包括只读存储器（ROM）、可编程ROM（PROM）、电可编程ROM（EPROM）、电可擦除可编程ROM（EEPROM）或闪存。易失性存储器可包括随机存取存储器（RAM）或者外部高速缓冲存储器。作为说明而非局限，RAM通过多种形式可得，诸如静态RAM（SRAM）、动态RAM（DRAM）、同步DRAM（SDRAM）、双速据率SDRAM（SSRSDRAM）、增强型SDRAM（ESDRAM）、同步链路（Synchlink）DRAM（SLDRAM）、存储器总线（Rambus）直接RAM（RDRAM）、直接存储器总线动态RAM（DRDRAM）、以及存储器总线动态RAM（RDRAM）等。
需要说明的是，在本文中，术语“包括”、“包含”或者其任何其它变体意在涵盖非排他性的包含，从而使得包括一系列要素的过程、装置、物品或者方法不仅包括那些要素，而且还包括没有明确列出的其它要素，或者是还包括为这种过程、装置、物品或者方法所固有的要素。在没有更多限制的情况下，由语句“包括一个……”限定的要素，并不排除在包括该要素的过程、装置、物品或者方法中还存在另外的相同要素。
以上所述仅为本申请的优选实施例，并非因此限制本申请的专利范围，凡是利用本申请说明书及附图内容所作的等效结构或等效流程变换，或直接或间接运用在其它相关的技术领域，均同理包括在本申请的专利保护范围内。

Claims (20)

  1. 一种ECG信号的检测方法，其特征在于，包括以下步骤：
    将已知信号节奏类型的第一ECG信号输入至卷积神经网络中进行训练，训练出对应的训练参数，将得出训练参数的卷积神经网络作为ECG信号的检测模型；
    将待检测的第二ECG信号输入至所述ECG信号的检测模型中进行计算，输出得到所述第二ECG信号的信号节奏类型。
  2. 根据权利要求1所述的ECG信号的检测方法，其特征在于，所述将已知信号节奏类型的第一ECG信号输入至卷积神经网络中进行训练，训练出对应的训练参数的步骤之前，包括：
    对所述第一ECG信号进行标准化处理。
  3. 根据权利要求1所述的ECG信号的检测方法，其特征在于，所述将待检测的第二ECG信号输入至所述ECG信号的检测模型中进行计算，输出得到所述第二ECG信号的信号节奏类型的步骤，包括：
    将待检测的第二ECG信号输入至所述ECG信号的检测模型中的输入层，经过32层卷积层卷积后再经输出层输出，得到所述第二ECG信号的信号节奏类型。
  4. 根据权利要求1所述的ECG信号的检测方法，其特征在于，所述卷积神经网络使用的损失函数为交叉熵损失函数，所述卷积神经网络的优化方法为Adam方法。
  5. 根据权利要求3所述的ECG信号的检测方法，其特征在于，所述经过32层卷积层卷积后再经输出层输出的步骤之前，包括：
    对所述第二ECG信号依次进行批标准化、线性修正以及深度学习。
  6. 根据权利要求1所述的ECG信号的检测方法，其特征在于，所述将已知信号节奏类型的第一ECG信号输入至卷积神经网络中进行训练，训练出对应的训练参数，将得出训练参数的卷积神经网络作为ECG信号的检测模型的步骤之后，包括：
    将已知信号节奏类型的第三ECG信号输入至所述ECG信号的检测模型进行验证，验证所述ECG信号的检测模型输出的信号节奏类型是否与所述第三ECG信号的信号节奏类型相同。
  7. 根据权利要求1所述的ECG信号的检测方法，其特征在于，所述将已知信号节奏类型的第一ECG信号输入至卷积神经网络中进行训练，训练出对应的训练参数的步骤，包括：
    将已知信号节奏类型的第一ECG信号输入至卷积神经网络中进行训练，并使得输出的训练结果为所述第一ECG信号已知的信号节奏类型，以得出对应的训练参数。
  8. 一种ECG信号的检测装置，其特征在于，包括：
    训练单元，用于将已知信号节奏类型的第一ECG信号输入至卷积神经网络中进行训练，训练出对应的训练参数，将得出训练参数的卷积神经网络作为ECG信号的检测模型；
    检测单元，用于将待检测的第二ECG信号输入至所述ECG信号的检测模型中进行计算，输出得到所述第二ECG信号的信号节奏类型。
  9. 根据权利要求8所述的ECG信号的检测装置，其特征在于，
    标准化单元，用于对所述第一ECG信号进行标准化处理。
  10. 根据权利要求8所述的ECG信号的检测装置，其特征在于，所述检测单元具体用于：
    将待检测的第二ECG信号输入至所述ECG信号的检测模型中的输入层，经过32层卷积层卷积后再经输出层输出，得到所述第二ECG信号的信号节奏类型。
  11. 根据权利要求8所述的ECG信号的检测装置，其特征在于，所述卷积神经网络使用的损失函数为交叉熵损失函数，所述卷积神经网络的优化方法为Adam方法。
  12. 根据权利要求10所述的ECG信号的检测装置，其特征在于，所述检测单元经过32层卷积层卷积后再经输出层输出之前，包括：
    对所述第二ECG信号依次进行批标准化、线性修正以及深度学习。
  13. 根据权利要求8所述的ECG信号的检测装置，其特征在于，还包括：
    验证单元，用于将已知信号节奏类型的第三ECG信号输入至所述ECG信号的检测模型进行验证，验证所述ECG信号的检测模型输出的信号节奏类型是否与所述第三ECG信号的信号节奏类型相同。
  14. 根据权利要求8所述的ECG信号的检测装置，其特征在于，所述训练单元具体用于：
    将已知信号节奏类型的第一ECG信号输入至卷积神经网络中进行训练，并使得输出的训练结果为所述第一ECG信号已知的信号节奏类型，以得出对应的训练参数。
  15. 一种计算机设备，包括存储器和处理器，所述存储器中存储有计算机可读指令，其特征在于，所述处理器执行所述计算机可读指令时实现ECG信号的检测方法，所述方法包括：
    将已知信号节奏类型的第一ECG信号输入至卷积神经网络中进行训练，训练出对应的训练参数，将得出训练参数的卷积神经网络作为ECG信号的检测模型；
    将待检测的第二ECG信号输入至所述ECG信号的检测模型中进行计算，输出得到所述第二ECG信号的信号节奏类型。
  16. 根据权利要求15所述的计算机设备，其特征在于，所述处理器将已知信号节奏类型的第一ECG信号输入至卷积神经网络中进行训练，训练出对应的训练参数的步骤之前，包括：
    对所述第一ECG信号进行标准化处理。
  17. 根据权利要求15所述的计算机设备，其特征在于，所述处理器将待检测的第二ECG信号输入至所述ECG信号的检测模型中进行计算，输出得到所述第二ECG信号的信号节奏类型的步骤，包括：
    将待检测的第二ECG信号输入至所述ECG信号的检测模型中的输入层，经过32层卷积层卷积后再经输出层输出，得到所述第二ECG信号的信号节奏类型。
  18. 一种计算机非易失性可读存储介质，其上存储有计算机可读指令，其特征在于，所述计算机可读指令被处理器执行时实现ECG信号的检测方法，所述方法包括：
    将已知信号节奏类型的第一ECG信号输入至卷积神经网络中进行训练，训练出对应的训练参数，将得出训练参数的卷积神经网络作为ECG信号的检测模型；
    将待检测的第二ECG信号输入至所述ECG信号的检测模型中进行计算，输出得到所述第二ECG信号的信号节奏类型。
  19. 根据权利要求18所述的计算机非易失性可读存储介质，其特征在于，所述处理器将已知信号节奏类型的第一ECG信号输入至卷积神经网络中进行训练，训练出对应的训练参数的步骤之前，包括：
    对所述第一ECG信号进行标准化处理。
  20. 根据权利要求18所述的计算机非易失性可读存储介质，其特征在于，所述处理器将待检测的第二ECG信号输入至所述ECG信号的检测模型中进行计算，输出得到所述第二ECG信号的信号节奏类型的步骤，包括：
    将待检测的第二ECG信号输入至所述ECG信号的检测模型中的输入层，经过32层卷积层卷积后再经输出层输出，得到所述第二ECG信号的信号节奏类型。
PCT/CN2018/095494 2018-04-20 2018-07-12 Ecg信号的检测方法、装置、计算机设备和存储介质 Ceased WO2019200746A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201810362786.X 2018-04-20
CN201810362786.XA CN108926341A (zh) 2018-04-20 2018-04-20 Ecg信号的检测方法、装置、计算机设备和存储介质

Publications (1)

Publication Number Publication Date
WO2019200746A1 true WO2019200746A1 (zh) 2019-10-24

Family

ID=64449367

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/095494 Ceased WO2019200746A1 (zh) 2018-04-20 2018-07-12 Ecg信号的检测方法、装置、计算机设备和存储介质

Country Status (2)

Country Link
CN (1) CN108926341A (zh)
WO (1) WO2019200746A1 (zh)

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110507313B (zh) * 2019-08-30 2023-06-06 武汉中旗生物医疗电子有限公司 一种腔内心电信号重建方法及装置
CN111048111B (zh) * 2019-12-25 2023-07-04 广州酷狗计算机科技有限公司 检测音频的节奏点的方法、装置、设备及可读存储介质
CN113456084A (zh) * 2021-05-31 2021-10-01 山西云时代智慧城市技术发展有限公司 一种基于ResNet-Xgboost模型预测心电波异常类型的方法
CN120995247B (zh) * 2025-10-21 2025-12-23 天津工业大学 一种ecg信号分类系统、方法、设备、介质及产品

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105748063A (zh) * 2016-04-25 2016-07-13 山东大学齐鲁医院 基于多导联和卷积神经网络的心律失常智能诊断方法
CN106214145A (zh) * 2016-07-20 2016-12-14 杨平 一种基于深度学习算法的心电图分类方法
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
CN107767413A (zh) * 2017-09-20 2018-03-06 华南理工大学 一种基于卷积神经网络的图像深度估计方法
CN107822622A (zh) * 2017-09-22 2018-03-23 成都比特律动科技有限责任公司 基于深度卷积神经网络的心电图诊断方法和系统

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR102501838B1 (ko) * 2016-02-01 2023-02-23 삼성전자주식회사 심전도 신호를 이용한 인증 장치 및 방법
CN106372390B (zh) * 2016-08-25 2019-04-02 汤一平 一种基于深度卷积神经网络的预防肺癌自助健康云服务系统
EP3312748A1 (en) * 2016-10-19 2018-04-25 Fujitsu Limited Method for aiding a diagnosis, program and apparatus
CN107184198A (zh) * 2017-06-01 2017-09-22 广州城市职业学院 一种心电信号分类识别方法
CN107256393B (zh) * 2017-06-05 2020-04-24 四川大学 基于深度学习的一维生理信号的特征提取与状态识别
CN107495962B (zh) * 2017-09-18 2020-05-05 北京大学 一种单导联脑电的睡眠自动分期方法
CN107736894A (zh) * 2017-09-24 2018-02-27 天津大学 一种基于深度学习的心电信号情绪识别方法
CN107510452B (zh) * 2017-09-30 2019-10-08 扬美慧普（北京）科技有限公司 一种基于多尺度深度学习神经网络的心电检测方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105748063A (zh) * 2016-04-25 2016-07-13 山东大学齐鲁医院 基于多导联和卷积神经网络的心律失常智能诊断方法
CN106214145A (zh) * 2016-07-20 2016-12-14 杨平 一种基于深度学习算法的心电图分类方法
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
CN107767413A (zh) * 2017-09-20 2018-03-06 华南理工大学 一种基于卷积神经网络的图像深度估计方法
CN107822622A (zh) * 2017-09-22 2018-03-23 成都比特律动科技有限责任公司 基于深度卷积神经网络的心电图诊断方法和系统

Also Published As

Publication number Publication date
CN108926341A (zh) 2018-12-04

Similar Documents

Publication Publication Date Title
WO2019200746A1 (zh) Ecg信号的检测方法、装置、计算机设备和存储介质
JP3987675B2 (ja) 臨床検査システム
CN112801224B (zh) 糖尿病的分型概率预测方法、装置、设备及存储介质
US20240274297A1 (en) Surgical warning
Mistry et al. The smart analysis of machine learning-based diagnostics model of cardiovascular diseases in patients
CN111755086B (zh) 数据异常检测方法及装置
US20220102010A1 (en) Systems and methods for modelling a human subject
CN117594208A (zh) 一种医用耗材的质控方法、装置、设备及存储介质
CN118522396B (zh) 一种临床诊疗数据录入方法及系统
KR20220149093A (ko) 인공 지능을 이용한 혈액 투석 중 저혈압 예측 장치 및 그 방법
CN114203306A (zh) 医疗事件预测模型训练方法、医疗事件预测方法及装置
US20090222248A1 (en) Method and system for determining a combined risk
CN116206774B (zh) 一种结合大数据自动匹配护理治疗方案的方法及系统
CN118072943A (zh) 一种基于人工智能医疗数据分析的辅助系统
CN120183734A (zh) 一种基于机器学习的危险性腹痛早期识别方法
CN119153116A (zh) 血吸虫性肝纤维化患者早期诊断数据管理系统
CN120015319A (zh) 基于多数据融合的脑部重症护理分析系统及方法
KR102893317B1 (ko) 심전도를 이용한 건강 상태의 예측 방법, 프로그램 및 장치
CN118553372A (zh) 基于人工智能的医疗数据管理系统
CN119274821A (zh) 基于增强现实的护理信息共享系统及方法
CN116168841A (zh) 基于XGBoost的呼吸机仿真预测系统及其实现方法
US20220051809A1 (en) System of predicting risks with biomedical data, method of predicting risks with biomedical data, and non-transient state computer-readable storage medium
CN111986804A (zh) 基于体温数据进行模型训练的方法、装置和计算机设备
CN112397179A (zh) 一种医用心电图智能分析系统及分析方法
CN119314700B (zh) 一种基于预测模型的脑卒中重症监护分析方法及系统

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: 18914949

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM1205 DATED 22.02.2021)

122 Ep: pct application non-entry in european phase

Ref document number: 18914949

Country of ref document: EP

Kind code of ref document: A1