CN113143204A - Electrocardiosignal quality evaluation method, computer device and storage medium - Google Patents
Electrocardiosignal quality evaluation method, computer device and storage medium Download PDFInfo
- Publication number
- CN113143204A CN113143204A CN202110118684.5A CN202110118684A CN113143204A CN 113143204 A CN113143204 A CN 113143204A CN 202110118684 A CN202110118684 A CN 202110118684A CN 113143204 A CN113143204 A CN 113143204A
- Authority
- CN
- China
- Prior art keywords
- neural network
- data set
- quality
- electrocardiosignals
- layer
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 43
- 238000013441 quality evaluation Methods 0.000 title claims abstract description 23
- 238000013528 artificial neural network Methods 0.000 claims abstract description 111
- 238000012549 training Methods 0.000 claims abstract description 39
- 238000012360 testing method Methods 0.000 claims abstract description 19
- 238000010845 search algorithm Methods 0.000 claims abstract description 14
- 238000005457 optimization Methods 0.000 claims abstract description 8
- 239000010410 layer Substances 0.000 claims description 39
- 230000006870 function Effects 0.000 claims description 16
- 238000011176 pooling Methods 0.000 claims description 13
- 238000001303 quality assessment method Methods 0.000 claims description 8
- 238000000605 extraction Methods 0.000 claims description 4
- 238000004458 analytical method Methods 0.000 claims description 3
- 230000000747 cardiac effect Effects 0.000 claims description 3
- 239000002356 single layer Substances 0.000 claims description 3
- 238000013500 data storage Methods 0.000 abstract description 3
- 238000012545 processing Methods 0.000 abstract description 3
- 238000004590 computer program Methods 0.000 description 8
- 238000012544 monitoring process Methods 0.000 description 7
- 230000008569 process Effects 0.000 description 7
- 238000010586 diagram Methods 0.000 description 5
- 238000005070 sampling Methods 0.000 description 4
- 208000024172 Cardiovascular disease Diseases 0.000 description 3
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 description 3
- 238000001514 detection method Methods 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 3
- 230000009471 action Effects 0.000 description 2
- 230000004913 activation Effects 0.000 description 2
- 206010003119 arrhythmia Diseases 0.000 description 2
- 230000006793 arrhythmia Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000003745 diagnosis Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 208000019622 heart disease Diseases 0.000 description 2
- 230000007774 longterm Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000033764 rhythmic process Effects 0.000 description 2
- 230000035945 sensitivity Effects 0.000 description 2
- 208000020446 Cardiac disease Diseases 0.000 description 1
- 230000003321 amplification Effects 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000005189 cardiac health Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 239000000470 constituent Substances 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 230000034994 death Effects 0.000 description 1
- 231100000517 death Toxicity 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000013209 evaluation strategy Methods 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012806 monitoring device Methods 0.000 description 1
- 238000003199 nucleic acid amplification method Methods 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 230000001575 pathological effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7221—Determining signal validity, reliability or quality
-
- 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
Abstract
The invention discloses an electrocardiosignal quality evaluation method, a computer device and a storage medium, wherein the electrocardiosignal quality evaluation method comprises the steps of obtaining a training data set, inputting the training data set into an automatic architecture search algorithm for neural network search, obtaining a neural network of a Mobile invested Residual network block with different super parameters, carrying out fine adjustment optimization binary training on the neural network, obtaining a test data set, inputting the test data set into the neural network, obtaining a quality evaluation result output by the neural network and the like. The method has the advantages of smaller required data storage space and higher operation speed, the used neural network is a lightweight neural network, the requirement on the computing capability of a computer is low, the method is applicable to mobile terminals such as wearable equipment or mobile phones and the like, but is more robust, and the quality unacceptable data of the type can be more reliably and effectively identified. The invention is widely applied to the technical field of electrocardiosignal processing.
Description
Technical Field
The invention relates to the technical field of electrocardiosignal processing, in particular to an electrocardiosignal quality evaluation method, a computer device and a storage medium.
Background
Cardiovascular diseases account for about one third of the worldwide deaths, and early detection of cardiovascular diseases is therefore of great importance. ECG is the most widely and effectively used means for medical staff and related experts to diagnose heart health, and diagnosis of heart diseases is performed by drawing a plurality of potential changes from a plurality of points on the body surface through an electrocardiograph in each cardiac cycle of the heart, based on the waveform and rhythm of the electrocardiogram. For medical personnel and cardiologists, long-term monitoring of patient electrocardiographic waveforms and rhythms is difficult to achieve, and wearable ECG monitoring devices for automatic arrhythmia detection can greatly assist physicians in cardiac disease diagnosis.
The burden of medical workers can be effectively reduced by monitoring the electrocardio events at the mobile end and the cloud end by utilizing the neural network, long-time electrocardio monitoring can be provided for middle-aged and elderly people, the risk brought by cardiovascular diseases is reduced, and the necessary premise of carrying out arrhythmia automatic detection is to carry out quality evaluation on the electrocardio signals. In modern related scientific research work, researchers mostly carry out quality evaluation on electrocardiosignals based on reference characteristics such as time domain characteristics or frequency domain characteristics of QRS wave groups, R-R intervals and the like, but the existing methods are not robust enough, and are easy to misjudge pathological signals into signals with unqualified quality, which is an unacceptable phenomenon in clinical application. In recent years, the rapid development of the neural network makes the neural network an effective tool, the accuracy of the electrocardio quality evaluation performed by the neural network can reach a higher level, but due to the limitation of computing power, the general neural network can only be deployed at the cloud end and is difficult to operate on terminals with lower computing power, such as a mobile terminal, and therefore, the application of the neural network for the electrocardio quality evaluation is limited.
Disclosure of Invention
In view of at least one of the above technical problems, it is an object of the present invention to provide an electrocardiographic signal quality evaluation method, a computer device, and a storage medium.
In one aspect, an embodiment of the present invention provides a method for evaluating quality of a dynamic electrocardiographic signal based on neural network architecture search, including:
acquiring a training data set; the training data set comprises a plurality of sections of electrocardiosignals and quality marks corresponding to each section of electrocardiosignals; the quality mark is used for marking the corresponding electrocardiosignal as acceptable or unacceptable;
inputting part or all of the training data set into an automatic architecture search algorithm to search a neural network, and obtaining the neural network of a Mobile invoked Residual network block with different hyper-parameters;
performing fine-tuning optimization two-classification training on the neural network;
acquiring a test data set; the test data set comprises a plurality of segments of electrocardiosignals;
and inputting part or all of the test data set into the neural network, and acquiring a quality evaluation result of the electrocardiosignals in the test data set, which is output by the neural network.
Further, the acquiring the training data set includes:
acquiring original dynamic electrocardiogram data through wearable equipment;
segmenting the original dynamic electrocardio data according to fixed time length to obtain a plurality of sections of electrocardiosignals;
carrying out noise analysis on each section of the electrocardiosignal;
when the electrocardiosignals comprise power frequency interference noise, myoelectricity interference noise and/or baseline drift noise, setting the quality marks of the electrocardiosignals to be unacceptable, otherwise, setting the quality marks of the electrocardiosignals to be acceptable;
the training data set is formatted as HDF5 format.
Further, the Mobile inversed Residual network block is configured to use the received input data as a low-dimensional feature map, expand the low-dimensional feature map into a high-dimensional feature map by using 1 × 1 point convolution, perform local layer feature extraction on the high-dimensional feature map by using k × k deep convolution, and map the extracted features into a low-dimensional space by using 1 × 1 linear point convolution.
Further, in the step of inputting part or all of the training data set into an automatic architecture search algorithm for neural network search, the input of each layer of the neural network is the output of the previous layer, and the output of the ith layer of the neural network is the output of the previous layerWhereinIs an input of the i-th layer, bjIs a mask determined by probability, bjCan be expressed as
bj=[1,0,...,0]×p1+[0,1,...,0]×p2+...+[0,0,...,1]×pN;
pk,pq≠0;pk+pq=1;p i0, i ≠ k, q; i is 0,.. times, N, k, q is not more than N; where N is the number of candidate operations, which may be represented as the set o ═ { convk,n,s,poolingk_p,s_p,identityn_i}; wherein conv is convolution operation, k is the size of a convolution kernel, n is the number of the convolution kernels, and s is the step length of convolution; posing is pooling operation, k _ p is the size of the pooling kernel, and s _ p is the step size of pooling; identity is the full connect operation and n _ i is the number of full connect outputs.
Further, in the step of inputting part or all of the training data set into an automatic architecture search algorithm for neural network search, the loss function of the neural network isWherein LossCEFor the cross entropy of the tape label to be smooth,for regularization term, T (latency) is the operation delay of the whole neural network, λ1And λ2Are coefficients.
Further, the air conditioner is provided with a fan,wherein y issoftIs a smoothed true one-hot encoded tag,is a predictive value of the neural network,orEpsilon is the error;
whereinFor the latency of the jth operation at layer i in the neural network,and M and N are the number of candidate operations for the structural parameter of the ith operation of the jth layer in the neural network.
Further, in the step of performing fine tuning optimization two-class training on the neural network, the weight parameter of the gradient descent update network may be represented as:
wherein wiThe updated network parameter, wi-1For the network parameter updated last time, α is the learning rate.
Further, the method for evaluating the quality of the dynamic electrocardiosignal further comprises the following steps:
before inputting part or all of the training data set into an automatic architecture search algorithm for searching the neural network, fixing a first layer input layer of the neural network as a single-layer long-short term memory unit (LSTM).
In another aspect, the present invention further includes a computer device, including a memory and a processor, where the memory is used to store at least one program, and the processor is used to load the at least one program to perform the method for evaluating the quality of a dynamic electrocardiographic signal based on a neural network architecture search in the embodiment.
In another aspect, the present invention further includes a storage medium, in which a program executable by a processor is stored, and the program executable by the processor is used for executing the method for evaluating the quality of a dynamic electrocardiographic signal based on neural network architecture search in the embodiment.
The invention has the beneficial effects that: the neural network trained and used in the embodiment is a neural network with different hyper-parameters and a Mobile invested Residual network block, the received electrocardiosignals can be one-dimensional signals, and compared with the prior art which can only receive two-dimensional electrocardiosignals, the required data storage space is smaller and the operation speed is higher; the neural network in the embodiment is a neural network with different super-parameter Mobile invoked redundant network blocks, has few parameters, is a light-weight neural network, has low requirement on the computing capability of a computer, and can be suitable for Mobile terminals such as wearable equipment or Mobile phones; compared with the traditional electrocardiosignal quality evaluation algorithm, the neural network in the embodiment is more robust to the problem of unacceptable signal quality caused by the falling of short-time leads, and can more reliably and effectively identify the type of unacceptable quality data; after training and fine adjustment, quality evaluation can be carried out according to different indexes such as accuracy, sensitivity, specificity, positive predicted value, negative predicted value, AUC, F1 score and the like.
Drawings
FIG. 1 is a flow chart of a method for evaluating quality of a dynamic electrocardiographic signal based on neural network architecture search according to an embodiment;
FIG. 2 is a schematic diagram of electrocardiographic data with acceptable quality marks in an embodiment;
FIGS. 3 and 4 are schematic diagrams of electrocardiographic data with quality marks of unacceptable quality in an embodiment;
FIGS. 5, 6 and 7 are schematic diagrams of a Mobile invoked redundant network block based on different hyper-parameters in an embodiment;
FIG. 8 is a schematic diagram of a neural network search in an embodiment;
FIG. 9 is a block diagram of a neural network in an embodiment;
fig. 10 and 11 are structural views of LSTM used in the embodiment.
Detailed Description
In this embodiment, referring to fig. 1, the method for evaluating the quality of a dynamic electrocardiographic signal based on neural network architecture search includes the following steps:
s1, acquiring a training data set;
s2, inputting part or all of the training data set into an automatic architecture search algorithm to search a neural network, and obtaining the neural network of the Mobile invested Residual network block with different super parameters;
s3, carrying out fine-tuning optimization two-classification training on the neural network;
s4, acquiring a test data set; the test data set comprises a plurality of segments of electrocardiosignals;
and S5, inputting part or all of the test data set into a neural network, and obtaining a quality evaluation result of the electrocardiosignals in the test data set, which is output by the neural network.
In this embodiment, the step S1, that is, the step of acquiring the training data set, includes:
s101, acquiring original dynamic electrocardiogram data through wearable equipment;
s102, segmenting original dynamic electrocardio data according to fixed time length to obtain multiple sections of electrocardio signals;
s103, carrying out noise analysis on each section of electrocardiosignal;
s104, when the electrocardiosignal comprises power frequency interference noise, myoelectricity interference noise and/or baseline drift noise, setting the quality mark of the electrocardiosignal to be unacceptable, otherwise, setting the quality mark of the electrocardiosignal to be acceptable;
and S105, setting the format of the training data set to be in the HDF5 format.
In the step S101, the original electrocardiographic signal is obtained through the wearable device, and the time course is long, the electrocardiographic data under various daily behaviors can be included, the noise types in the data are richer and more diverse, the diversity of the data is increased, the training of the neural network is facilitated, and the actual situation of remote electrocardiographic event monitoring is better met. The wearable device may use a sampling frequency of 500Hz when acquiring the raw dynamic electrocardiographic data.
In step S102, segmenting the electrocardiogram data and making the segmented electrocardiogram data into a general data set of a neural network; in the actual long-term dynamic electrocardiographic monitoring, in order to obtain a monitoring result in real time, monitoring information should be output every other short time period, so that in this example, the length of the time segment is set to be a common value of 10 seconds, and the electrocardiographic signals in the short time period can also reduce the influence of baseline drift and filter motion artifacts. The sampling rate of the electrocardiosignals is set to be 500 Hz.
10 seconds of electrocardiographic data also have two types of tags: acceptable quality due to having clinical diagnostic value and unacceptable quality due to not having clinical diagnostic value. Specifically, the quality index of the examined electrocardiographic signal may be evaluated in steps S103 and S104 by an expert group consisting of 1 evaluator, 1 auditor, 2 preliminary auditor, and 1 final auditor, and the electrocardiographic signal may be classified as acceptable or unacceptable. For example, the quality index would be acceptable for the electrocardiographic data shown in FIG. 2 and unacceptable for the electrocardiographic data shown in FIGS. 3 and 4.
In step S105, the format of the training data set is set to the hierachical dataformat, that is, the HDF5 format.
In step S2, the training data set is input to an automatic architecture search algorithm for neural network search, and a neural network with excellent performance of each index is obtained. Network architecture search is a mainstream direction of automatic machine learning, and comprises three aspects of a search space, a search strategy and a performance evaluation strategy. In this embodiment, the search policy is set as a gradient descent method, and the performance evaluation policy is set as a whole network evaluation for weight sharing. In this embodiment, the parameters of the chain architecture search space include:
A1. a maximum number of layers of the neural network;
A2. operation of each layer: convolution, pooling, and the like;
A3. computing the relevant hyper-parameters: the size k of the filter, the number c of filters, the step size s, the amplification rate e, etc.;
A4. activation function: tanh, Sigmoid, Sotfmax, ReLU, etc.
In step S2, the search space based on the network blocks may be network blocks built in different connection manners, such as an inclusion block, a Residual block, a Mobile invoked Residual block, and the like, and also may be a combination of network blocks of the same type with different hyper-parameters. Considering that the present invention may have a wider practical application scenario, the embodiment selects the search space of the Mobile invoked redundant network block based on different super parameters, and the principles of the used Mobile invoked redundant network block based on different super parameters are shown in fig. 5, 6 and 7, which save more storage space and computational resources than the chain architecture search space, and perform better classification problems than other types of network blocks under the condition of saving storage space.
Referring to fig. 5, 6 and 7, a Mobile invested Residual network block in a neural network firstly amplifies an input low-dimensional feature map into a high-dimensional feature map by using 1 × 1 point convolution, then performs local layer feature extraction on the high-dimensional feature map by using k × k deep convolution, and maps the extracted features into a low-dimensional space by using 1 × 1 linear point convolution, which is an "amplification-feature extraction-reduction" process. Wherein the depth separable convolutional layer functions as feature extraction; the dot convolution has the function of expanding or reducing the number of channels; the ReLU6 is an activation function, and has the function of increasing the nonlinear relation among the layers of the network, so that the network has nonlinear expression and can complete various complex tasks; linear is a fully connected operation, which functions to integrate the extracted features; hopping connections are used to implement Residual, which can slow down the problem of gradient disappearance and also enable the network to make use of the previous information.
The search method of Mobile invoked redundant network blocks (abbreviated as MB) based on different hyper-parameters is to selectively group the network blocks with different hyper-parameters into a network with optimal performance to a data set, and the search process is as shown in fig. 8. In step S2, i.e., the step of inputting part or all of the training data set into the automatic architecture search algorithm for neural network search, the neural network is searched for at each layerThe input is the output of the previous layer, the output of the ith layer of the neural network isWhereinIs an input of the i-th layer, bjIs a mask determined by probability, which has only two non-zero elements in the network searching process, so that only two candidate operations o remain at each layer of the networkjTo save computational resources and memory space.
bjCan be expressed as:
bj=[1,0,...,0]×p1+[0,1,...,0]×p2+...+[0,0,...,1]×pN;
pk,pq≠0;pk+pq=1;p i0, i ≠ k, q; i is 0,.. times, N, k, q is not more than N; where N is the number of candidate operations, which may be represented as the set o ═ { convk,n,s,poolingk_p,s_p,identityn_i}; wherein conv is convolution operation, k is the size of a convolution kernel, n is the number of the convolution kernels, and s is the step length of convolution; posing is pooling operation, k _ p is the size of the pooling kernel, and s _ p is the step size of pooling; identity is the full connect operation and n _ i is the number of full connect outputs.
In step S2, i.e., the step of inputting part or all of the training data set into the automatic architecture search algorithm for neural network search, the loss function of the neural network is
Wherein LossCEFor the cross entropy with label smoothness, the problem caused by one-hot coding can be relieved: the model was easily over-fitted, i.e. performed well in the training samples, but poorly in the unseen samples. The cross entropy can be expressed asWherein y issoftIs a smoothed true one-hot encoded tag,the predicted value of the neural network is, in the binary problem, the original tag is 0 or 1, and the tag after the corresponding one-hot coding can be expressed as: [1,0]Or [0,1 ]]One-hot encoding maps discrete features into Euclidean space, making the distance calculation between features more reasonable, but the problem is as described above. Therefore, the corresponding smoothed true one-hot encoding tag can be expressed as:orWhere ε is the error. After the label is smooth, the tolerance of the label to the wrong data is improved, and the problem of model overfitting can be relieved.
Representation of loss functionFor the regularization term, λ1And λ2Are coefficients. T (latency) is the operation delay of the entire neural network, i.e., the time from input to output, and is added to the loss function, so that the scale of the network can be constrained to be light and fast.
In this embodiment, T (latency) can be expressed asWhereinFor the latency of the jth operation at layer i in the neural network,the structural parameters of the ith operation of the jth layer in the neural network are M and N are the number of candidate operations.
In step S3, i.e., the step of training the neural network in a fine tuning optimization binary mode, the neural network (also referred to as a binary model) obtained in step S2 is further subjected to fine tuning optimization. The structure of the neural network obtained by the step S2 is shown in fig. 9, and the neural network has a 17-layer structure, and can be observed to have a structure with a wide head and a wide tail, and a narrow middle, similarly to a Mobile invested Residual block (MB). The neural network shown in fig. 9, except that the first layer is a single convolution operation and the last layer is a single fully-connected operation, each of the remaining layers is MB, and the output can be expressed as:
xt=MB(xt-1)=Linear(Conv1×1(ReLU6(Dwise(ReLU6(Conv1×1(xt-1))))))+xt-1。
in the neural network shown in fig. 9, the output of each layer is the input of the next layer. If the step length of the deep convolution is 2, the output of the current layer does not need to be added with the output x of the previous layert-1。
In the embodiment, the cross entropy with smooth label is selected as the loss function, so that the problem that the model is easy to over-fit due to one-hot coding can be solved, and the problem can be expressed asWherein y issoftIs a true one-hot tag after smoothing,is a predicted value of the network. The weight parameter of the gradient descent update network can be expressed as:
wherein wiThe updated network parameter, wi-1For the network parameter updated last time, α is the learning rate.
After steps S1-S3 are executed, step S4 is executed to obtain a test data set including multiple segments of electrocardiographic signals, where the format of the electrocardiographic signals in the test data set may be the same as the format of the electrocardiographic signals in the training data set, that is, the format is also obtained by sampling the electrocardiographic signals in the wearable device at a sampling frequency of 500Hz, the duration of each segment of electrocardiographic signals is 10S, and the data format of the test data set is hierarchial dataformat, that is, HDF5 format.
And step S5 is executed, the electrocardiosignal segments in the test data set are input into the neural network after fine adjustment obtained by the step S1-S3 for secondary classification, and the output result of the neural network indicates that the electrocardiosignal segments input into the neural network are acceptable or unacceptable in quality, so that the quality evaluation of the electrocardiosignals in the test data set is realized.
The neural network trained and used by the dynamic electrocardiosignal quality evaluation method based on neural network architecture search in the embodiment is a neural network with different hyper-parameters, the received electrocardiosignals can be one-dimensional signals, and compared with the prior art which can only receive two-dimensional electrocardiosignals, the needed data storage space is smaller and the operation speed is higher; the neural network in the embodiment is a neural network with different super-parameter Mobile invoked redundant network blocks, has a small number of parameters, is a lightweight neural network, has low requirement on the computing capability of a computer, and can be suitable for Mobile terminals such as wearable equipment or Mobile phones; compared with the traditional electrocardiosignal quality evaluation algorithm, the neural network in the embodiment is more robust to the problem of unacceptable signal quality caused by the falling of short-time leads, and can more reliably and effectively identify the type of unacceptable quality data; after training and fine adjustment, quality evaluation can be carried out according to different indexes such as accuracy, sensitivity, specificity, positive predicted value, negative predicted value, AUC, F1 score and the like.
Because the dynamic electrocardiosignal quality assessment method based on the neural network architecture search in the embodiment can be applied to wearable equipment, the wearable equipment can locally operate the dynamic electrocardiosignal quality assessment method based on the neural network architecture search in the embodiment after acquiring heartbeat signals, so that the quality of the heartbeat signals can be rapidly assessed to be acceptable or unacceptable, if the heartbeat signals are unacceptable, the wearable equipment can immediately prompt personnel to adjust the wearing posture and perform reacquisition, thereby being beneficial to rapidly obtaining the acceptable heartbeat signals, compared with the technical scheme that the wearable equipment needs to upload data to be processed to an upper computer or a cloud server, the upper computer or the cloud server processes the data, and then returns the processing result to the wearable equipment for alarm prompt, the dynamic electrocardiosignal quality assessment method based on the neural network architecture search in the embodiment has smaller time delay, thereby providing a better use experience.
In this embodiment, in step S2, before inputting part or all of the training data set into the automatic architecture search algorithm for neural network search, the first layer input layer of the neural network is further fixed as a single-layer long-short term memory unit LSTM, so as to further implement the dynamic evaluation function based on the technical effects of steps S1-S5.
The LSTM used in this embodiment is an efficient form of the cyclic convolution network RNN, and in the sequence model, obtaining the previous information helps the model to perform the current task more efficiently, while the RNN can obtain the previous information through the loop, but when the time interval is increased, the RNN's ability to learn the previous information is greatly reduced, and the LSTM can solve the problem. Referring to fig. 10 and 11, the LSTM includes the following three important structures a 1-A3:
A1. forget the door: the decision of which information to discard from the cell, i.e. the selective "forgetting" of previous information stored in the memory cell, the "remembering" of important information, and the "forgetting" of unimportant information, can be expressed as:
ft=σ(Wf·[ht-1,xt]+bf);
wherein h ist-1For the output of the previous node, σ is sigmoid function, the input variable can be mapped between (0,1), so the weight parameter W at the forgetting gatefUnder the action of the action, after the input and the weight parameter are operated, the offset is added, if the input and the weight parameter are in a larger negative value, the sigmoid function sets the weight parameter to be a number close to 0, the forgetting door is closed, the information is prevented from passing through, namely, the previously reserved information is forgottenInformation; if the positive value is larger, the sigmoid function will set it to a number close to 1, and the gate is forgotten to open, and the information is almost completely passed, i.e. "remembers" the information that was previously retained.
A2. An input gate: determining which new information to save from the cell state, i.e. selectively "remembering" the input, the principle, like forgetting, can be expressed as:
it=σ(Wi·[ht-1,xt]+bi);
And updating the state of the cell, namely updating the stored content by combining the output of the forgetting gate and the output information after selective memory, wherein the updated stored content of the node is the information selectively discarded by the forgetting gate of the next node, and the updating of the state of the cell can be represented as:
wherein C ist-1Updated information for the last node cell stored in the cell.
A4. An output gate: determining the output component in the cell state can be expressed as:
ot=σ(Wo·[ht-1,xt]+bo);
ht=ot⊙tanh(Ct) An exclusive-OR indicates an exclusive-OR operation;
wherein h istIs the final output of the node.
The LSTM controls the transmission state by the gating state, and can memorize information that needs to be memorized for a long time, leave unimportant information, dynamically process sequence data, and perform context.
The dynamic electrocardiographic signal quality evaluation method based on neural network architecture search in the present embodiment may be implemented by writing a computer program for implementing the dynamic electrocardiographic signal quality evaluation method based on neural network architecture search in the present embodiment, writing the computer program into a computer device or a storage medium, and executing the dynamic electrocardiographic signal quality evaluation method based on neural network architecture search in the present embodiment when the computer program is read and run.
It should be noted that, unless otherwise specified, when a feature is referred to as being "fixed" or "connected" to another feature, it may be directly fixed or connected to the other feature or indirectly fixed or connected to the other feature. Furthermore, the descriptions of upper, lower, left, right, etc. used in the present disclosure are only relative to the mutual positional relationship of the constituent parts of the present disclosure in the drawings. As used in this disclosure, the singular forms "a", "an", and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. In addition, unless defined otherwise, all technical and scientific terms used in this example have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in the description of the embodiments herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used in this embodiment, the term "and/or" includes any combination of one or more of the associated listed items.
It will be understood that, although the terms first, second, third, etc. may be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element of the same type from another. For example, a first element could be termed a second element, and, similarly, a second element could be termed a first element, without departing from the scope of the present disclosure. The use of any and all examples, or exemplary language ("e.g.," such as "or the like") provided with this embodiment is intended merely to better illuminate embodiments of the invention and does not pose a limitation on the scope of the invention unless otherwise claimed.
It should be recognized that embodiments of the present invention can be realized and implemented by computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer readable memory. The methods may be implemented in a computer program using standard programming techniques, including a non-transitory computer-readable storage medium configured with the computer program, where the storage medium so configured causes a computer to operate in a specific and predefined manner, according to the methods and figures described in the detailed description. Each program may be implemented in a high level procedural or object oriented programming language to communicate with a computer system. However, the program(s) can be implemented in assembly or machine language, if desired. In any case, the language may be a compiled or interpreted language. Furthermore, the program can be run on a programmed application specific integrated circuit for this purpose.
Further, operations of processes described in this embodiment can be performed in any suitable order unless otherwise indicated herein or otherwise clearly contradicted by context. The processes described in this embodiment (or variations and/or combinations thereof) may be performed under the control of one or more computer systems configured with executable instructions, and may be implemented as code (e.g., executable instructions, one or more computer programs, or one or more applications) collectively executed on one or more processors, by hardware, or combinations thereof. The computer program includes a plurality of instructions executable by one or more processors.
Further, the method may be implemented in any type of computing platform operatively connected to a suitable interface, including but not limited to a personal computer, mini computer, mainframe, workstation, networked or distributed computing environment, separate or integrated computer platform, or in communication with a charged particle tool or other imaging device, and the like. Aspects of the invention may be embodied in machine-readable code stored on a non-transitory storage medium or device, whether removable or integrated into a computing platform, such as a hard disk, optically read and/or write storage medium, RAM, ROM, or the like, such that it may be read by a programmable computer, which when read by the storage medium or device, is operative to configure and operate the computer to perform the procedures described herein. Further, the machine-readable code, or portions thereof, may be transmitted over a wired or wireless network. The invention described in this embodiment includes these and other different types of non-transitory computer-readable storage media when such media include instructions or programs that implement the steps described above in conjunction with a microprocessor or other data processor. The invention also includes the computer itself when programmed according to the methods and techniques described herein.
A computer program can be applied to input data to perform the functions described in the present embodiment to convert the input data to generate output data that is stored to a non-volatile memory. The output information may also be applied to one or more output devices, such as a display. In a preferred embodiment of the invention, the transformed data represents physical and tangible objects, including particular visual depictions of physical and tangible objects produced on a display.
The above description is only a preferred embodiment of the present invention, and the present invention is not limited to the above embodiment, and any modifications, equivalent substitutions, improvements, etc. within the spirit and principle of the present invention should be included in the protection scope of the present invention as long as the technical effects of the present invention are achieved by the same means. The invention is capable of other modifications and variations in its technical solution and/or its implementation, within the scope of protection of the invention.
Claims (10)
1. A dynamic electrocardiosignal quality assessment method based on neural network architecture search is characterized by comprising the following steps:
acquiring a training data set; the training data set comprises a plurality of sections of electrocardiosignals and quality marks corresponding to each section of electrocardiosignals; the quality mark is used for marking the corresponding electrocardiosignal as acceptable or unacceptable;
inputting part or all of the training data set into an automatic architecture search algorithm to search a neural network, and obtaining the neural network of a Mobile invoked Residual network block with different hyper-parameters;
performing fine-tuning optimization two-classification training on the neural network;
acquiring a test data set; the test data set comprises a plurality of segments of electrocardiosignals;
and inputting part or all of the test data set into the neural network, and acquiring a quality evaluation result of the electrocardiosignals in the test data set, which is output by the neural network.
2. The method according to claim 1, wherein the obtaining of the training data set comprises:
acquiring original dynamic electrocardiogram data through wearable equipment;
segmenting the original dynamic electrocardio data according to fixed time length to obtain a plurality of sections of electrocardiosignals;
carrying out noise analysis on each section of the electrocardiosignal;
when the electrocardiosignals comprise power frequency interference noise, myoelectricity interference noise and/or baseline drift noise, setting the quality marks of the electrocardiosignals to be unacceptable, otherwise, setting the quality marks of the electrocardiosignals to be acceptable;
the training data set is formatted as HDF5 format.
3. The method as claimed in claim 1, wherein the Mobile invoked Residual network block is configured to use the received input data as a low-dimensional feature map, expand the low-dimensional feature map into a high-dimensional feature map by using 1 × 1 point convolution, perform local-level feature extraction on the high-dimensional feature map by using k × k deep convolution, and map the extracted features into a low-dimensional space by using 1 × 1 linear point convolution.
4. The method according to claim 1, wherein part or all of the training data set is inputted to the neural network architecture search based dynamic electrocardiosignal quality assessment methodIn the step of searching the neural network in the automatic architecture search algorithm, the input of each layer of the neural network is the output of the previous layer, and the output of the ith layer of the neural network isWhereinIs an input of the i-th layer, bjIs a mask determined by probability, bjCan be expressed as
bj=[1,0,...,0]×p1+[0,1,...,0]×p2+...+[0,0,...,1]×pN;
pk,pq≠0;pk+pq=1;pi0, i ≠ k, q; i is 0,.. times, N, k, q is not more than N; where N is the number of candidate operations, which may be represented as the set o ═ { convk,n,s,poolingk_p,s_p,identityn_i}; wherein conv is convolution operation, k is the size of a convolution kernel, n is the number of the convolution kernels, and s is the step length of convolution; posing is pooling operation, k _ p is the size of the pooling kernel, and s _ p is the step size of pooling; identity is the full connect operation and n _ i is the number of full connect outputs.
5. The method according to claim 1, wherein in the step of inputting part or all of the training data set into an automatic architecture search algorithm for neural network search, the loss function of the neural network isWherein LossCEFor the cross entropy of the tape label to be smooth,for regularization term, T (latency) is the operation delay of the whole neural network, λ1And λ2Are coefficients.
6. The method for evaluating the quality of dynamic electrocardiosignals based on neural network architecture search according to claim 5, wherein the method comprises the following steps:wherein y issoftIs a smoothed true one-hot encoded tag,is a predictive value of the neural network,orEpsilon is the error;
7. The method for evaluating the quality of dynamic electrocardiographic signals based on neural network architecture search according to claim 6, wherein in the step of performing fine tuning optimization classification training on the neural network, the weight parameters of the gradient descent update network can be represented as:
wherein wiThe updated network parameter, wi-1For the network parameter updated last time, α is the learning rate.
8. The dynamic electrocardiosignal quality assessment method based on neural network architecture search according to any one of claims 1 to 7, further comprising:
before inputting part or all of the training data set into an automatic architecture search algorithm for searching the neural network, fixing a first layer input layer of the neural network as a single-layer long-short term memory unit (LSTM).
9. A computer device comprising a memory for storing at least one program and a processor for loading the at least one program to perform the method for dynamic cardiac electrical signal quality assessment based on neural network architecture search of any one of claims 1-8.
10. A storage medium having stored therein a program executable by a processor, wherein the program executable by the processor is adapted to perform a method for dynamic cardiac electrical signal quality assessment based on neural network architecture search according to any one of claims 1-8 when executed by the processor.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110118684.5A CN113143204A (en) | 2021-01-28 | 2021-01-28 | Electrocardiosignal quality evaluation method, computer device and storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110118684.5A CN113143204A (en) | 2021-01-28 | 2021-01-28 | Electrocardiosignal quality evaluation method, computer device and storage medium |
Publications (1)
Publication Number | Publication Date |
---|---|
CN113143204A true CN113143204A (en) | 2021-07-23 |
Family
ID=76878943
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110118684.5A Pending CN113143204A (en) | 2021-01-28 | 2021-01-28 | Electrocardiosignal quality evaluation method, computer device and storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113143204A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113855037A (en) * | 2021-10-15 | 2021-12-31 | 南方医科大学 | Atrial fibrillation identification method and device based on Transformer |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105595990A (en) * | 2016-01-27 | 2016-05-25 | 浙江铭众科技有限公司 | Intelligent terminal device for evaluating and distinguishing quality of electrocardiosignal |
CN109858362A (en) * | 2018-12-28 | 2019-06-07 | 浙江工业大学 | A kind of mobile terminal method for detecting human face based on inversion residual error structure and angle associated losses function |
DE102018221413A1 (en) * | 2017-12-14 | 2019-06-19 | Robert Bosch Gmbh | EFFECTIVE MODULE DESIGN FOR DEEP CONVOLUTIONAL NEURONAL NETWORKS USING A SEARCH |
CN110309758A (en) * | 2019-06-26 | 2019-10-08 | 广州视源电子科技股份有限公司 | Electrocardiosignal feature extracting method, device, computer equipment and storage medium |
-
2021
- 2021-01-28 CN CN202110118684.5A patent/CN113143204A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105595990A (en) * | 2016-01-27 | 2016-05-25 | 浙江铭众科技有限公司 | Intelligent terminal device for evaluating and distinguishing quality of electrocardiosignal |
DE102018221413A1 (en) * | 2017-12-14 | 2019-06-19 | Robert Bosch Gmbh | EFFECTIVE MODULE DESIGN FOR DEEP CONVOLUTIONAL NEURONAL NETWORKS USING A SEARCH |
CN109858362A (en) * | 2018-12-28 | 2019-06-07 | 浙江工业大学 | A kind of mobile terminal method for detecting human face based on inversion residual error structure and angle associated losses function |
CN110309758A (en) * | 2019-06-26 | 2019-10-08 | 广州视源电子科技股份有限公司 | Electrocardiosignal feature extracting method, device, computer equipment and storage medium |
Non-Patent Citations (5)
Title |
---|
HAN CAI ET AL.: "PROXYLESSNAS: DIRECT NEURAL ARCHITECTURE SEARCH ON TARGET TASK AND HARDWARE", 《HTTP://ARIXIV.ORG/PDF/1812.00332.PDF》 * |
SEUNGWOO HAN ET AL.: "Optimal DNN architecture search using Bayesian Optimization Hyperband for arrhythmia detection", 《2020 IEEE WIRELESS POWER TRANSFER CONFERENCE》 * |
刘忠雨: "《移动平台深度神经网络实战:原理、架构与优化》", 31 January 2020, 机械工业出版社 * |
张丹丹: "《基于深度神经网络技术的高分遥感图像处理及应用》", 31 August 2020, 中国宇航出版社 * |
赖杰伟等: "基于DenseNet的心电数据自动诊断算法", 《南方医科大学学报》 * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113855037A (en) * | 2021-10-15 | 2021-12-31 | 南方医科大学 | Atrial fibrillation identification method and device based on Transformer |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Yuan et al. | Muvan: A multi-view attention network for multivariate temporal data | |
Çınar et al. | Classification of normal sinus rhythm, abnormal arrhythmia and congestive heart failure ECG signals using LSTM and hybrid CNN-SVM deep neural networks | |
US10758139B2 (en) | Automatic method to delineate or categorize an electrocardiogram | |
Liu et al. | MFB-CBRNN: A hybrid network for MI detection using 12-lead ECGs | |
Xu et al. | Interpretation of electrocardiogram (ECG) rhythm by combined CNN and BiLSTM | |
Rad et al. | Applying deep learning to stereotypical motor movement detection in autism spectrum disorders | |
US20190133468A1 (en) | Method and apparatus for high accuracy photoplethysmogram based atrial fibrillation detection using wearable device | |
CN110623658B (en) | Signal processing method, signal processing apparatus, medical device, and storage medium | |
CN111772619B (en) | Heart beat identification method based on deep learning, terminal equipment and storage medium | |
CN109614885A (en) | A kind of EEG signals Fast Classification recognition methods based on LSTM | |
Liu et al. | Arrhythmias classification by integrating stacked bidirectional LSTM and two-dimensional CNN | |
Zhou et al. | Arrhythmia recognition and classification through deep learning-based approach | |
Ullah et al. | An end-to-end cardiac arrhythmia recognition method with an effective densenet model on imbalanced datasets using ecg signal | |
Prakash et al. | A system for automatic cardiac arrhythmia recognition using electrocardiogram signal | |
Banerjee et al. | A hybrid CNN-LSTM architecture for detection of coronary artery disease from ECG | |
Ragav et al. | Scalable deep learning for stress and affect detection on resource-constrained devices | |
Ukil et al. | Resource constrained CVD classification using single lead ECG on wearable and implantable devices | |
Fei et al. | Machine and deep learning algorithms for wearable health monitoring | |
CN116864140A (en) | Intracardiac branch of academic or vocational study postoperative care monitoring data processing method and system thereof | |
CN113143204A (en) | Electrocardiosignal quality evaluation method, computer device and storage medium | |
Haroon | ECG arrhythmia classification Using deep convolution neural networks in transfer learning | |
CN115470832A (en) | Electrocardiosignal data processing method based on block chain | |
Liu et al. | Ensemble learning-based atrial fibrillation detection from single lead ECG wave for wireless body sensor network | |
CN110786847B (en) | Electrocardiogram signal library building method and analysis method | |
Balcı | A hybrid attention-based LSTM-XGBoost model for detection of ECG-based atrial fibrillation |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210723 |