CN114004258B - Semi-supervised electrocardiographic abnormality detection method - Google Patents
Semi-supervised electrocardiographic abnormality detection method Download PDFInfo
- Publication number
- CN114004258B CN114004258B CN202111293464.2A CN202111293464A CN114004258B CN 114004258 B CN114004258 B CN 114004258B CN 202111293464 A CN202111293464 A CN 202111293464A CN 114004258 B CN114004258 B CN 114004258B
- Authority
- CN
- China
- Prior art keywords
- electrocardiographic
- generator
- discriminator
- threshold
- training
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 37
- 230000005856 abnormality Effects 0.000 title claims abstract description 19
- 238000012549 training Methods 0.000 claims abstract description 51
- 238000000034 method Methods 0.000 claims abstract description 35
- 230000002159 abnormal effect Effects 0.000 claims abstract description 21
- 230000009466 transformation Effects 0.000 claims abstract description 9
- 230000009467 reduction Effects 0.000 claims abstract description 6
- 239000013598 vector Substances 0.000 claims description 14
- 238000000354 decomposition reaction Methods 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 6
- 230000003993 interaction Effects 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 239000007787 solid Substances 0.000 claims description 3
- 230000006870 function Effects 0.000 description 8
- 238000013528 artificial neural network Methods 0.000 description 7
- 238000010586 diagram Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000002547 anomalous effect Effects 0.000 description 2
- 238000013459 approach Methods 0.000 description 2
- 230000002457 bidirectional effect Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000002565 electrocardiography Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 241000212376 Ammi Species 0.000 description 1
- 235000007034 Carum copticum Nutrition 0.000 description 1
- 230000008485 antagonism Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000008033 biological extinction Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000007635 classification algorithm Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 230000008014 freezing Effects 0.000 description 1
- 238000007710 freezing Methods 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 238000002955 isolation Methods 0.000 description 1
- 238000003064 k means clustering Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
- G06F2218/06—Denoising by applying a scale-space analysis, e.g. using wavelet analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Signal Processing (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The invention discloses a semi-supervised electrocardiographic abnormality detection method, which comprises the following steps: carrying out noise reduction treatment on the electrocardiosignals by using wavelet threshold transformation, randomly selecting a plurality of normal electrocardiosignals as a training set, and randomly selecting a plurality of normal electrocardiosignals and abnormal electrocardiosignals for a threshold optimizing data set; constructing an AD-ECGGAN network model, wherein the AD-ECGGAN network model comprises a generator and a discriminator; a fixed generator for extracting n real electrocardiographic samples from the training set, generating n false electrocardiographic samples in the generator by using defined noise distribution, and training a discriminator by the real electrocardiographic samples and the false electrocardiographic samples; the AD-ECGGAN model for detecting the electrocardiosignal abnormality improves the network structure for generating an countermeasure network, and provides a novel training method, so that the classifier for detecting the electrocardiosignal abnormality can be trained under the condition of lacking an abnormal electrocardiosignal sample, a professional doctor is not required to label abnormal data, and labor and time cost are greatly saved.
Description
Technical Field
The invention relates to the technical field of intelligent medical treatment and health, in particular to a semi-supervised electrocardio abnormality detection method based on an antagonism network.
Background
Electrocardiogram measurement plays an important auxiliary role in modern diagnostics. The development of artificial intelligence greatly improves the accuracy of electrocardiosignal classification and anomaly detection. However, most existing detection methods belong to supervised learning. Supervised learning relies on a large number of accurately labeled datasets. When analyzing medical signals such as electrocardiograms, the acquired samples need to be marked by special doctors, and the process consumes a great deal of time and labor cost. And the electrocardiosignals are used as medical data and have privacy and sensitivity. A large number of accurately labeled electrocardiographic datasets are often difficult to obtain. In addition, the number of positive samples in the medical data is much greater than the number of negative samples, and such unbalanced data distribution can also affect the accuracy of classification and anomaly detection. The above reasons cause that the existing supervised electrocardiographic abnormality detection method is difficult to effectively land.
An anomaly of data refers to a significant deviation of one or a series of observations from the overall distribution of the data. Anomaly detection can be categorized from whether a dataset is annotated into supervised learning, semi-supervised learning, and unsupervised learning. The supervised learning requires a large number of marked data sets and is easy to cause overfitting to the training set, so that the generalization capability of the model is not strong, the current accuracy of the unsupervised learning is not ideal, and the unsupervised learning which only needs positive samples and does not need negative samples is particularly suitable for medical data. Anomaly detection can be divided methodically into statistical learning based, classical machine learning based and neural network (deep learning) based. Template matching based on statistical learning, a moving average method, an exponential smoothing method, a preset confidence interval setting and the like. The classical machine learning-based method comprises K-means clustering, density-based clustering, a local anomaly factor method, forest isolation, a single-class support vector machine, extreme gradient lifting and the like. The neural network is based on convolutional neural network, residual neural network, wavelet network, LSTM network, automatic encoder, etc. The above methods can be generalized to statistics, regression, clustering, and reconstruction.
With the development of deep neural networks, methods for abnormality detection through neural networks are increasingly used. The initial detection method using the neural network is to learn normal data and then use a classification algorithm to distinguish normal data from abnormal data. The encoder and decoder based methods are effective in such methods, and the idea is to learn an encoding-decoding model with normal data, so that the model can reconstruct normal data but cannot reconstruct abnormal data. However, this approach is very susceptible to noise and requires various constraints to be placed on the model. One idea of anomaly detection is to consider a positive sample as a class and a negative sample as a class, and perform two classifications. The actual situation is that positive samples are easily available in large numbers, while negative samples are difficult to obtain.
Disclosure of Invention
The invention aims to provide a semi-supervised electrocardiographic anomaly detection method, which trains an electrocardiographic anomaly detection model under the condition of only having a normal electrocardiographic sample, has higher model detection efficiency compared with a supervised model, and can feed back results in real time.
In order to achieve the above purpose, the technical scheme of the application is as follows: a semi-supervised electrocardiographic anomaly detection method, comprising:
step 1: carrying out noise reduction treatment on the electrocardiosignals by using wavelet threshold transformation, randomly selecting a plurality of normal electrocardiosignals as a training set, and randomly selecting a plurality of normal electrocardiosignals and abnormal electrocardiosignals for a threshold optimizing data set;
step 2: constructing an AD-ECGGAN network model (an electrocardiographic abnormality detection countermeasure network model), wherein the AD-ECGGAN network model comprises a generator and a discriminator;
step 3: a fixed generator for extracting n real electrocardiographic samples from the training set, generating n false electrocardiographic samples in the generator by using defined noise distribution, and training a discriminator by the real electrocardiographic samples and the false electrocardiographic samples; setting a k-time updating discriminator for each cycle, and updating the 1-time generator into one round; printing the loss of the generator and the discriminator and the accuracy of the discriminator in each round;
step 4: judging whether the network is converged through the loss of the generator and the discriminator and the accuracy of the discriminator, and if so, turning to step 5; if not, continuing the countermeasure training in the step 3;
step 5: stopping training the generator, continuing to train the discriminator by using the abnormal electrocardiograph data generated by the generator and the real electrocardiograph data in the training set until the accuracy is not improved any more, and storing a discriminator model at the moment;
step 6: calling the stored discriminator model, and finding an optimal threshold theta in a threshold interval of [0.4-0.6] by using the threshold optimizing data set;
step 7: and detecting the electrocardio abnormality by using the optimal threshold value theta of the discriminator model.
Further, the wavelet threshold transformation is used for carrying out noise reduction processing on the electrocardiosignal, and the method concretely comprises the following steps:
inputting electrocardiosignals into a wavelet function to perform multi-scale decomposition; and after the wavelet coefficients of all scales are obtained, threshold decomposition is carried out, and then the electrocardiosignal is reconstructed through inverse transformation.
Further, the threshold decomposition rule is a fixed threshold method:
the wavelet function selects a soft threshold estimation method, and the formula is as follows:
wherein N is the length value of the high-frequency coefficient of each layer of wavelet; λ is a threshold; w is a high frequency wavelet coefficient.
Further, the objective function expression of the AD-ECGGAN network model is as follows:
where G represents the generator, D represents the arbiter, and z represents random noise.
Further, a modified BILSTM layer is added to the generator, and a small batch of arbiter layers is added to the arbiter.
Further, the implementation method of the improved BILSTM layer is as follows:
k t ＝f(w 1 x t +w 2 k t-1 )
k′ t ＝f(w 3 x t +w 5 k′ t+1 )
o t ＝g(w 4 k t +w 6 k′ t )
wherein x is t Representing the current input, k t Representing the current output, k, in forward propagation t-1 Representing the previous output, k ', of the forward propagation' t Representing the current output, k ', in backward propagation' t+1 Represents the last output of backward propagation, o t Represents the final output, w 1 -w 6 The weight of the threshold unit; calculating from time 1 to time t during forward propagation, and obtaining forward output of each time; reversely calculating from time t to time 1 during backward propagation to obtain backward output of each time; and combining the calculation results of the forward layer and the backward layer at the corresponding time at each time to obtain final output, so that the final output covers the information of the bidirectional input sequence.
Furthermore, the small batch of discriminators obtains the difference information between feature images in a small batch of samples as the additional output of the next layer in the discriminators so as to achieve the purpose of information interaction between samples, specifically:
wherein sample x i A layer in the discriminatorIs given by (a)Will f (x) i ) Multiplying by a tensorObtain tensor->Then the L1 distance of the M row vector between each sample is obtainedThen all c b (x i ,x j ) Added to obtain o (x) i ) b B o (x i ) b Merging to obtain a vector o (x) i ) The method comprises the steps of carrying out a first treatment on the surface of the Finally, o (x i ) And f (x) i ) And combining the two vectors into one vector to be used as the input of the next layer of the discriminator.
Further, the loss function of the generator and the arbiter is KL divergence (relative entropy) as follows:
where p (x) represents the distribution of true solid electrical data and q (x) generates the distribution of electrocardiographic data.
As a further step, the method of stopping the training generator is to freeze parameters of the generator, specifically: setting a flag=0 at the beginning of training, and updating the flag=1 after judging that the network converges; judging whether the flag is 0 at the beginning of each round of countermeasure training, and if the flag is 0, performing the countermeasure training; if the parameter is 1, the generator parameter is frozen, and the training of the discriminator is continued.
As a further step, the optimal threshold θ traverses between [0.4,0.6], and the output value of the discriminant model is o, o > θ represents the predicted normal electrocardiograph, o < θ represents the predicted abnormal electrocardiograph, and in order to find the θ with the highest accuracy, the calculation formula is as follows:
TP is true positive, which indicates that the classification prediction is correct and is judged to be positive; TN is true negative, meaning that the classification prediction is correct and is determined to be negative; FP is false positive, indicating that the classification prediction is wrong and is determined to be positive; FN is false negative, indicating that classification is mispredicted and is determined to be negative.
By adopting the technical scheme, the invention can obtain the following technical effects: the AD-ECGGAN model for detecting the electrocardiosignal abnormality improves the network structure for generating an countermeasure network, and provides a novel training method, so that the classifier for detecting the electrocardiosignal abnormality can be trained under the condition of lacking an abnormal electrocardiosignal sample, a professional doctor is not required to label abnormal data, and labor and time cost are greatly saved. The accuracy of the proposed model reaches 94.33%, and the accuracy under AMMI classification reaches 96.61%. Moreover, because of the semi-supervised model, the model can detect any abnormal electrocardiography, including rare, rarely recorded abnormal electrocardiography. Compared with a supervised model, the model detection efficiency is higher, real-time feedback results can be achieved, and the method can be applied to wearable medical equipment such as single leads and two leads. The single view generating capability and the effect of generating electrocardiographic data by the improved network are obviously improved. By the aid of the single-view training method, any existing countermeasure network model can simply change the original codes by using the training method provided by the invention, so that the discriminator model has the capability of anomaly detection. In addition, the concept and the method of the invention can be simply migrated and popularized to the abnormality detection of other time sequence data, including but not limited to log abnormality detection, audio abnormality detection, video abnormality detection and the like.
Drawings
FIG. 1 is a flow chart of a semi-supervised electrocardiographic anomaly detection method;
FIG. 2 is a diagram of the structure of an AD-ECGGAN network model;
FIG. 3 is a schematic diagram of BILSTM layer model;
FIG. 4 is a diagram of the generation of reactive network generator and arbiter training loss;
fig. 5 is a graph of the resulting electrocardiographic effect.
Detailed Description
The following description of the embodiments of the present invention will be made more complete and should be understood by reference to the figures of the accompanying drawings. All other embodiments, which can be made by a person skilled in the art without making any inventive effort, are intended to fall within the scope of the present invention.
The invention aims to solve the technical problem that an electrocardiographic abnormality detection model is trained under the condition of only having a normal electrocardiographic sample, and provides an AD-ECGGAN network model, which comprises the following steps:
generating an abnormal electrocardiographic sample by improving the generation countermeasure network;
providing a new method for judging Nash equilibrium and a new training method, wherein a discriminator for generating an countermeasure network is used for anomaly detection;
the method specifically comprises the following steps:
step 1: carrying out noise reduction processing on the electrocardiosignal by using wavelet threshold transformation, randomly selecting a plurality of normal electrocardiosignals as training sets for the processed data, and randomly selecting a plurality of normal electrocardiosignals and abnormal electrocardiosignals for the threshold optimizing data set;
specifically, an electrocardiosignal is input into a wavelet function to carry out multi-scale decomposition; after the wavelet coefficients of each scale are obtained, threshold decomposition is carried out, and then electrocardiosignals are reconstructed through inverse transformation;
the threshold decomposition rule is a fixed threshold method:
the wavelet function selects a soft threshold estimation method, and the formula is as follows:
wherein N is the length value of the high-frequency coefficient of each layer of wavelet; λ is a threshold; w is a high frequency wavelet coefficient.
Step 2: constructing an AD-ECGGAN network model, wherein the AD-ECGGAN network model comprises a generator and a discriminator;
specifically, the AD-ECGGAN model is a new network model provided according to the electrocardiosignal characteristics, improves the structures of a generator and a discriminator network, and provides a new training method;
the objective function expression of the AD-ECGGAN model is:
where G represents the generator, D represents the arbiter, and z represents random noise. The generator receives a random noise z, maximizes the fitting of the random noise following a simple distribution, makes it follow the distribution of real data as much as possible, and generates false electrocardiographic samples. The discriminator is responsible for judging whether the input data is real data, giving the real sample as large a value as possible, and giving the generated sample as small a value as possible. During training, the generator and the discriminator are mutually game, the false electrocardio samples generated by the generator are gradually close to real data, and the judging capability of the false electrocardio samples is gradually enhanced. When the arbiter cannot determine whether the data is real data or distributed data, the generator has learned the distribution rule of the real data well, and the network converges at this time.
A modified BILSTM layer is added to the generator to make it more suitable for electrocardiographic signals. And a small batch of discriminator layers are added to the discriminators, so that the training stability is effectively improved.
BILSTM is a special recurrent neural network with memory units and threshold limiting mechanisms added to alleviate the gradient extinction and gradient explosion problems of RNNs in long sequence input tasks. BILSTM includes an input layer, a forward LSTM layer, a reverse LSTM layer, and an output layer. The improved BILSTM concretely comprises the following steps:
k t ＝f(w 1 x t +w 2 k t-1 )
k′ t ＝f(w 3 x t +w 5 k′ t+1 )
o t ＝g(w 4 k t +w 6 k′ t )
wherein x is t Representing the current input, k t Representing the current output, k, in forward propagation t-1 Representing the previous output, k ', of the forward propagation' t Representing the current output, k ', in backward propagation' t+1 Represents the last output of backward propagation, o t Represents the final output, w 1 -w 6 Is the weight of the threshold unit. And calculating from time 1 to time t in forward propagation, and acquiring the forward output of each time. And (3) reversely calculating from time t to time 1 during backward propagation, and obtaining backward output of each time. And combining the calculation results of the forward layer and the backward layer at the corresponding time at each time to obtain a final output, wherein the final output covers the information of the bidirectional input sequence.
The small batch of discriminant layers can effectively prevent the occurrence of mode collapse and improve the training stability. The small batch of discriminators layer obtains the difference information between feature images in a small batch of samples as the additional output of the next layer in the discriminators, so as to achieve the purpose of information interaction between samples.
Wherein sample x i The feature vector of a certain layer in the discriminator isWill f (x) i ) Multiplying by a tensorObtain tensor->Then calculate the L1 distance of M row vector between each sample to obtainThen all c b (x i ,x j ) Added to obtain o (x) i ) b B o (x i ) b Merging to obtain a vector o (x) i ). Finally, o (x i ) And f (x) i ) And combining the two vectors into one vector to be used as the input of the next layer of the discriminator.
Step 3: a fixed generator for extracting n real electrocardiographic samples from the training set, generating n false electrocardiographic samples in the generator by using defined noise distribution, and training a discriminator by the real electrocardiographic samples and the false electrocardiographic samples; setting a k-time updating discriminator for each cycle, and updating the 1-time generator into one round; printing the loss of the generator and the discriminator and the accuracy of the discriminator in each round;
specifically, the AD-ECGGAN network model is trained by the countermeasure mode.
In the invention, the generator generates abnormal data in the process of generating the electrocardio, and the discriminator distinguishes the boundary between the real normal data and the generated abnormal data. When zero and game between the generator and the arbiter approach Nash equilibrium, the training generator (i.e., the parameters of the freeze generator) is stopped, and then the arbiter continues to be trained until the arbiter accuracy is no longer improved. At the beginning of training, the generator cannot generate a sufficient amount of potentially anomalous data, and the arbiter can only separate the generated data from the true data by rough boundaries. After several iterations, the generator may generate more and more potential anomaly data that appears within or near the true data, at which point the arbiter may accurately describe the boundaries of the true data. The generator effectively improves the accuracy of the arbiter by generating potentially anomalous data.
Step 4: judging whether the network is converged through the loss of the generator and the discriminator and the accuracy of the discriminator, and if so, turning to step 5; if not, continuing the countermeasure training in the step 3;
specifically, the loss function of the generator and the discriminator is KL divergence (relative entropy) as follows:
where p (x) represents the true solid electrical distribution and q (x) generates the electrocardiographic distribution.
Judging whether the network is converged or not, namely judging whether the countermeasure games of the generator and the arbiter reach Nash equilibrium, wherein the Nash equilibrium point of zero and the game is 1/2. The specific method is to judge whether the accuracy of the discriminant is stabilized at 50+/-0.2% in the latest 5 rounds of training and whether the loss fluctuation of the generator and the discriminant is stabilized at +/-0.2%.
Step 5: stopping training the generator, continuing to train the discriminator by using the abnormal electrocardiograph data generated by the generator and the real electrocardiograph data in the training set until the accuracy is not improved any more, and storing a discriminator model at the moment;
specifically, the method for stopping the training generator is a parameter of the freezing generator, specifically: the flag=0 is set at the beginning of training, and flag=1 is updated when it is judged that the network converges. Judging whether the flag is 0 at the beginning of each round of countermeasure training, and if the flag is 0, performing the countermeasure training; if the parameter is 1, the generator parameter is frozen, and the training of the discriminator is continued. The continued training of the arbiter requires attention to stop immediately when accuracy is no longer improving, rather than stopping when the arbiter loss is no longer decreasing, the purpose of this operation is to prevent overfitting.
Step 6: calling the stored discriminator model, and finding an optimal threshold theta in a threshold interval of [0.4-0.6] by using the threshold optimizing data set;
step 7: and detecting the electrocardio abnormality by using the optimal threshold value theta of the discriminator model.
Specifically, the optimal threshold θ traverses between [0.4,0.6], and the output value of the discriminator model is set as o, o > θ represents the predicted normal electrocardiograph, o < θ represents the predicted abnormal electrocardiograph, and in order to find the θ with the highest accuracy, the calculation formula is as follows:
TP is true positive, which indicates that the classification prediction is correct and is judged to be positive; TN is true negative, meaning that the classification prediction is correct and is determined to be negative; FP is false positive, indicating that the classification prediction is wrong and is determined to be positive; FN is false negative, indicating that classification is mispredicted and is determined to be negative.
The above description is only of the preferred embodiments of the present invention and is not intended to limit the present invention, and it should be noted that it is possible for those skilled in the art to make several improvements and modifications without departing from the technical principle of the present invention, and these improvements and modifications should also be regarded as the protection scope of the present invention.
Claims (10)
1. A semi-supervised electrocardiographic anomaly detection method, comprising:
step 1: carrying out noise reduction treatment on the electrocardiosignals by using wavelet threshold transformation, randomly selecting a plurality of normal electrocardiosignals as a training set, and randomly selecting a plurality of normal electrocardiosignals and abnormal electrocardiosignals for a threshold optimizing data set;
step 2: constructing an AD-ECGGAN network model, wherein the AD-ECGGAN network model comprises a generator and a discriminator;
step 3: a fixed generator for extracting n real electrocardiographic samples from the training set, generating n false electrocardiographic samples in the generator by using defined noise distribution, and training a discriminator by the real electrocardiographic samples and the false electrocardiographic samples; setting a k-time updating discriminator for each cycle, and updating the 1-time generator into one round; printing the loss of the generator and the discriminator and the accuracy of the discriminator in each round;
step 4: judging whether the network is converged through the loss of the generator and the discriminator and the accuracy of the discriminator, and if so, turning to step 5; if not, continuing the countermeasure training in the step 3;
step 5: stopping training the generator, continuing to train the discriminator by using the abnormal electrocardiograph data generated by the generator and the real electrocardiograph data in the training set until the accuracy is not improved any more, and storing a discriminator model at the moment;
step 6: calling the stored discriminator model, and finding an optimal threshold theta in a threshold interval of [0.4-0.6] by using the threshold optimizing data set;
step 7: and detecting the electrocardio abnormality by using the optimal threshold value theta of the discriminator model.
2. The semi-supervised electrocardiographic anomaly detection method of claim 1, wherein the electrocardiographic signal is subjected to noise reduction processing by using wavelet threshold transformation, specifically comprising the following steps:
inputting electrocardiosignals into a wavelet function to perform multi-scale decomposition; and after the wavelet coefficients of all scales are obtained, threshold decomposition is carried out, and then the electrocardiosignal is reconstructed through inverse transformation.
3. The semi-supervised electrocardiographic anomaly detection method of claim 2, wherein the threshold decomposition rule is a fixed threshold method:
the wavelet function selects a soft threshold estimation method, and the formula is as follows:
wherein N is the length value of the high-frequency coefficient of each layer of wavelet; λ is a threshold; w is a high frequency wavelet coefficient.
4. The semi-supervised electrocardiographic anomaly detection method of claim 1, wherein the objective function expression of the AD-ECGGAN network model is:
where G represents the generator, D represents the arbiter, and z represents random noise.
5. A semi-supervised electrocardiographic anomaly detection method as claimed in claim 1, wherein a modified BILSTM layer is added to the generator and a small batch of discriminant layers is added to the discriminant.
6. The semi-supervised electrocardiographic anomaly detection method of claim 5, wherein the improved BILSTM layer implementation method is as follows:
k t ＝f(w 1 x t +w 2 k t-1 )
k′ t ＝f(w 3 x t +w 5 k′ t+1 )
o t ＝g(w 4 k t +w 6 k′ t )
wherein x is t Representing the current input, k t Representing the current output, k, in forward propagation t-1 Representing the previous output, k ', of the forward propagation' t Representing the current output, k ', in backward propagation' t+1 Represents the last output of backward propagation, o t Represents the final output, w 1 -w 6 The weight of the threshold unit; calculating from time 1 to time t during forward propagation, and obtaining forward output of each time; backward propagation counter from time t to time 1Calculating, namely acquiring backward output of each moment; and combining the calculation results of the forward layer and the backward layer at the corresponding moments to obtain final output.
7. The semi-supervised electrocardiographic anomaly detection method of claim 5, wherein the small batch of discriminators obtains difference information between feature maps in a small batch of samples as additional output of a next layer in the discriminators to achieve the purpose of information interaction between samples, specifically:
wherein sample x i The feature vector of a certain layer in the discriminator isWill f (x) i ) Multiplying by a tensor->Obtain tensor->Then the L1 distance of the M row vector between each sample is obtainedThen all c b (x i ,x j ) Added to obtain o (x) i ) b B o (x i ) b Merging to obtain a vector o (x) i ) The method comprises the steps of carrying out a first treatment on the surface of the Finally, o (x i ) And f (x) i ) And combining the two vectors into one vector to be used as the input of the next layer of the discriminator.
8. The semi-supervised electrocardiographic anomaly detection method of claim 1, wherein the loss function of the generator and the arbiter is KL divergence, with the formula:
where p (x) represents the distribution of true solid electrical data and q (x) generates the distribution of electrocardiographic data.
9. The semi-supervised electrocardiographic anomaly detection method of claim 1, wherein the method of stopping the training generator is a parameter of a freeze generator, specifically: setting a flag=0 at the beginning of training, and updating the flag=1 after judging that the network converges; judging whether the flag is 0 at the beginning of each round of countermeasure training, and if the flag is 0, performing the countermeasure training; if the parameter is 1, the generator parameter is frozen, and the training of the discriminator is continued.
10. The semi-supervised electrocardiographic anomaly detection method according to claim 1, wherein the optimal threshold θ traverses between [0.4,0.6], and the output value of the discriminant model is o, o > θ represents the predicted normal electrocardiograph, o < θ represents the predicted abnormal electrocardiograph, and in order to find the θ with the highest accuracy, the calculation formula is:
TP is true positive, which indicates that the classification prediction is correct and is judged to be positive; TN is true negative, meaning that the classification prediction is correct and is determined to be negative; FP is false positive, indicating that the classification prediction is wrong and is determined to be positive; FN is false negative, indicating that classification is mispredicted and is determined to be negative.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111293464.2A CN114004258B (en) | 2021-11-03 | 2021-11-03 | Semi-supervised electrocardiographic abnormality detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111293464.2A CN114004258B (en) | 2021-11-03 | 2021-11-03 | Semi-supervised electrocardiographic abnormality detection method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114004258A CN114004258A (en) | 2022-02-01 |
| CN114004258B true CN114004258B (en) | 2024-04-02 |
Family
ID=79927188
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111293464.2A Active CN114004258B (en) | 2021-11-03 | 2021-11-03 | Semi-supervised electrocardiographic abnormality detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114004258B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115381438B (en) * | 2022-08-24 | 2023-04-18 | 华南师范大学 | Method and device for reconstructing vital sign signals, computer equipment and storage medium |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112991579A (en) * | 2021-01-14 | 2021-06-18 | 北京航空航天大学 | Helicopter mobile part abnormity detection method based on generation countermeasure network |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8364254B2 (en) * | 2009-01-28 | 2013-01-29 | Brainscope Company, Inc. | Method and device for probabilistic objective assessment of brain function |
-
2021
- 2021-11-03 CN CN202111293464.2A patent/CN114004258B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112991579A (en) * | 2021-01-14 | 2021-06-18 | 北京航空航天大学 | Helicopter mobile part abnormity detection method based on generation countermeasure network |
Non-Patent Citations (1)
| Title |
|---|
| 陈国炜 ; 刘磊 ; 郭嘉逸 ; 潘宗序 ; 胡文龙 ; .基于生成对抗网络的半监督遥感图像飞机检测.中国科学院大学学报.2020,(04),全文. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114004258A (en) | 2022-02-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109948647B (en) | Electrocardiogram classification method and system based on depth residual error network | |
| US11727279B2 (en) | Method and apparatus for performing anomaly detection using neural network | |
| Yang et al. | Classification of acoustic physiological signals based on deep learning neural networks with augmented features | |
| CN111772619B (en) | Heart beat identification method based on deep learning, terminal equipment and storage medium | |
| Nejad et al. | A new enhanced learning approach to automatic image classification based on Salp Swarm Algorithm | |
| CN110558971A (en) | Method for generating countermeasure network electrocardiogram abnormity detection based on single target and multiple targets | |
| Wang et al. | Multi-branching temporal convolutional network for sepsis prediction | |
| CN116861354A (en) | Abnormality identification and mode distinguishing method and system for monitoring data of power transformer | |
| Zhou et al. | A feature extraction method for multivariate time series classification using temporal patterns | |
| CN114004258B (en) | Semi-supervised electrocardiographic abnormality detection method | |
| Murthy et al. | ANN, SVM and KNN classifiers for prognosis of cardiac ischemia-a comparison | |
| JP2019105871A (en) | Abnormality candidate extraction program, abnormality candidate extraction method and abnormality candidate extraction apparatus | |
| CN107862375A (en) | A kind of two stage equipment fault diagnosis method | |
| CN108665001B (en) | Cross-tested idle state detection method based on deep belief network | |
| CN110613445B (en) | DWNN framework-based electrocardiosignal identification method | |
| CN116595465A (en) | High-dimensional sparse data outlier detection method and system based on self-encoder and data enhancement | |
| CN116712099A (en) | Fetal heart state detection method based on multi-mode data, electronic equipment and storage medium | |
| CN113066544B (en) | FVEP characteristic point detection method based on CAA-Net and LightGBM | |
| CN116028858A (en) | Pre-labeled self-supervision neural network learning heart beat classification method and system | |
| Meng et al. | Biosignal Classification Based on Multi-Feature Multi-Dimensional WaveNet-LSTM Models. | |
| Mitrokhin et al. | Deep learning approach for QRS wave detection in ECG monitoring | |
| Sonawane et al. | Prediction of heart disease by optimized distance and density-based clustering | |
| Zhu et al. | Semi-supervised cardiac MRI image of the left ventricle segmentation algorithm based on contrastive learning | |
| Roy et al. | Identification and classification of human mental stress using physiological data: A low-power hybrid approach | |
| Muthulakshmi et al. | Prediction of Heart Disease using Ensemble Learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |