CN107184198A - A kind of electrocardiosignal classifying identification method - Google Patents
A kind of electrocardiosignal classifying identification method Download PDFInfo
- Publication number
- CN107184198A CN107184198A CN201710402483.1A CN201710402483A CN107184198A CN 107184198 A CN107184198 A CN 107184198A CN 201710402483 A CN201710402483 A CN 201710402483A CN 107184198 A CN107184198 A CN 107184198A
- Authority
- CN
- China
- Prior art keywords
- layer
- neural networks
- convolutional neural
- sampling
- convolutional
- 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 20
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 67
- 241001269238 Data Species 0.000 claims abstract description 13
- 238000000605 extraction Methods 0.000 claims abstract description 10
- 238000013461 design Methods 0.000 claims abstract description 8
- 230000033764 rhythmic process Effects 0.000 claims abstract description 7
- 238000010276 construction Methods 0.000 claims abstract description 3
- 238000005070 sampling Methods 0.000 claims description 45
- 238000012549 training Methods 0.000 claims description 16
- 210000002569 neuron Anatomy 0.000 claims description 13
- 230000008447 perception Effects 0.000 claims description 6
- 230000000747 cardiac effect Effects 0.000 claims description 5
- 206010003119 arrhythmia Diseases 0.000 claims description 4
- 230000006793 arrhythmia Effects 0.000 claims description 4
- 238000000926 separation method Methods 0.000 claims description 4
- 238000001514 detection method Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 claims description 3
- 238000012986 modification Methods 0.000 claims description 2
- 230000004048 modification Effects 0.000 claims description 2
- 230000001537 neural effect Effects 0.000 claims 1
- 238000012544 monitoring process Methods 0.000 abstract description 4
- 238000004458 analytical method Methods 0.000 abstract description 3
- 230000008859 change Effects 0.000 abstract description 3
- 230000000694 effects Effects 0.000 abstract description 3
- 208000019622 heart disease Diseases 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 11
- 238000012706 support-vector machine Methods 0.000 description 5
- 238000012795 verification Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 230000009466 transformation Effects 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000001934 delay Effects 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000009510 drug design Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 230000000302 ischemic effect Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000002107 myocardial effect Effects 0.000 description 1
- 230000007935 neutral effect Effects 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 230000002035 prolonged effect Effects 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 230000001052 transient effect Effects 0.000 description 1
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/316—Modalities, i.e. specific diagnostic methods
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/316—Modalities, i.e. specific diagnostic methods
- A61B5/318—Heart-related electrical modalities, e.g. electrocardiography [ECG]
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Biophysics (AREA)
- General Health & Medical Sciences (AREA)
- Veterinary Medicine (AREA)
- Public Health (AREA)
- Animal Behavior & Ethology (AREA)
- Surgery (AREA)
- Molecular Biology (AREA)
- Medical Informatics (AREA)
- Heart & Thoracic Surgery (AREA)
- Pathology (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Signal Processing (AREA)
- Psychiatry (AREA)
- Physiology (AREA)
- Mathematical Physics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Fuzzy Systems (AREA)
- Cardiology (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The invention discloses a kind of electrocardiosignal classifying identification method, including following implementation steps, obtain original electrocardiographicdigital figure Wave data of the time of measuring more than 10 seconds, and the extraction of electrocardiogram rhythm and pace of moving things information and PQRST waveforms is carried out according to original electrocardiographicdigital figure Wave data, obtain the digitalized data of electrocardiogram rhythm and pace of moving things information and PQRST waveforms；Design construction convolutional neural networks are simultaneously trained to it, and obtained PQRST Wave datas are inputted from the input for the convolutional neural networks trained, after being classified through convolutional neural networks, categorical data are obtained.The present invention utilizes capability of fitting of the convolutional neural networks to complex nonlinear function, realization is more accurately and efficiently classified to ECG signal, so as to real-time monitoring angiocardiopathy people at highest risk, sub-health population, state of an illness crowd undetermined, electrocardio change when intellectual analysis normal life, work and activity, assist in the state of an illness, or the ecg information of potential heart disease is captured, forewarning function is played to patient.
Description
Technical field
The present invention relates to field of medical technology, more particularly to a kind of recognition methods classified to electrocardiosignal.
Background technology
Hospital's electrocardiographic examination, although data precision is high, but can only record one section of patient in specific and very short time
Ecg wave form, to non-standing arrhythmia cordis, especially usually leaks to transient arrhythmia cordis and of short duration myocardial ischemic attacks
Inspection, delays diagnosis.24 hr Ambulatory EKG Monitorings (DCG) being widely used at present are able to record that prolonged electrocardiogram
(ECG) signal, but DCG does not have the ability of processing data, it is impossible to signal is classified automatically, it is impossible to which automatically identifying has
The ill-condition signal of medical significance, it is necessary to wait monitoring in 24 hours to terminate, can just transfer to doctor to analyze data and make knot
By.
The technology that ECG signal carries out intellectual analysis is being developed always.By analyzing the ECG data in medical features space
Distribution situation understand, can be to be classified compared with high-accuracy to ECG signal, it is necessary to a highly complex non-linear letter
Number, so the name of the game is the fitting to complex nonlinear function.In current prior art to the fitting of this function still
It is not ideal enough.Such as, a kind of prior art using wavelet transformation extract signal characteristic, utilize the multiple dimensioned characteristic of basic function
ECG signal is deployed under different scale and useful information is extracted.However, wavelet transformation is substantially simple integral transformation,
Its nonlinear function approximation ability is very limited.Another prior art uses SVMs (SVM), and SVM can be regarded as
Network structure and 3 layers of radial basis function network of Weight number adaptively adjustment, but it faces nonlinear function approximation that ability is restricted,
When given training sample, the family of functions that SVM can be fitted just entirely defines, but for more complicated nonlinear function, currently
SVM is just helpless.
The content of the invention
The defect that the present invention is directed to prior art is based on convolutional neural networks there is provided one kind, utilizes convolutional neural networks pair
The capability of fitting of complex nonlinear function, makes it be applied to one-dimensional ECG signal by reasonably designing, and realizes to ECG signal more
Effectively, more accurately classifying identification method.
In order to solve the above technical problems, the present invention is adopted the following technical scheme that：A kind of electrocardiosignal classifying identification method, its
It is characterised by：Including following implementation steps,
The original electrocardiographicdigital figure Wave data of a, acquirement time of measuring more than 10 seconds, and according to original electrocardiographicdigital figure Wave data
The extraction of electrocardiogram rhythm and pace of moving things information and PQRST waveforms is carried out, the digitlization number of electrocardiogram rhythm and pace of moving things information and PQRST waveforms is obtained
According to；
B, design construction convolutional neural networks are simultaneously trained to it, and the PQRST Wave datas that step a is obtained are from training
The input input of complete convolutional neural networks, after being classified through convolutional neural networks, categorical data is obtained from its output end.
Wherein step a had the sufficient waveform of quantity using the original electrocardiographicdigital figure Wave data of more than 10 seconds so that extract
Obtained electrocardiogram rhythm and pace of moving things information, PQRST Wave datas are more accurate.Wherein the extraction of PQRST waveforms can effectively reduce non-
The electrocardiogram classification results that the unstable waveform change influence lifting brought of diagnostic element such as human motion, electrode finally gives
Accuracy；
The convolutional neural networks used in step b are the machine learning methods of artificial intelligence field, and convolutional neural networks lead to
Combination low-level feature formation more abstract high-rise expression attribute classification or feature are crossed, to find the distributed nature table of data
Show, realize feature extraction and merging for pattern classification is unified.Convolutional neural networks have been led in the application such as image recognition at present
Domain demonstrates validity, and the accuracy of identification of conventional method can be greatly improved.Convolutional neural networks are imported in the present invention
Electrocardiogram classification field, the characteristics of reasonably classifying with reference to electrocardiogram, and deep learning method is instructed by above step
Practice and carry out waveform separation with deep learning method, the accuracy of electrocardiogram classification results can be increased substantially.
Preferably, the original electrocardiographicdigital figure Wave data that step a is used, i.e. ECG signal are double lead signals, its more single lead
Data have more fully information, can lift the accuracy of the electrocardiogram classification auxiliary information of correlation, compare more multi-lead
Design, double leads are easily realized on portable devices so that the scope of application of the invention is wider.
Preferably, the convolutional neural networks of step b designs are laminated and be designed as convolution down-sampling by convolutional layer and down-sampling
Layer, the output layer that the input layer three convolution down-sampling layers different from neuron number are constituted plus multi-layer perception (MLP) is combined into
For a complete convolutional neural networks；
Wherein convolution down-sampling layer 1 is produced as follows, first by 2 × 13 different convolution kernels of 6 initial values to input
Layer does convolution, obtains the convolutional layer C1 containing 6 characteristic faces, then use 2 × 3 sampling cores 6 sub-samplings of generation to convolutional layer C1
Layer S1；Convolution down-sampling layer 2 is produced as follows, and first 2 characteristic faces adjacent to sub- sample level S1, which are combined, obtains 6 newly
Characteristic face, then does convolution algorithm therewith using 62 × 10 convolution kernels, the convolutional layer C2 containing 36 characteristic faces is produced, to convolution
Layer C2 each characteristic face produces 36 sub- sample level S2 using 2 × 3 sampling cores；Convolution down-sampling layer 3 is produced as follows,
The combination of 3 adjacent surfaces is first carried out to sub- sample level S2 and obtains 12 new feature faces, then using 62 × 6 convolution kernels, produces and contains 72
The convolutional layer C3 of individual characteristic face, 72 sub- sample level S3 are produced to convolutional layer C3 using 2 × 3 sampling cores.
Preferably, training of the step b to convolutional neural networks uses 48 records in MIT-BIH arrhythmia cordis databases
Convolutional neural networks are carried out；Double lead signals are recorded as, 0.5~25Hz bandpass filter is first passed through, filters out High-frequency Interference
And baseline drift；In each lead, centered on R ripples, 100 points of R wavefront is taken, latter 199 points, totally 300 sampled point including R ripples
A complete cardiac is constituted, 25000 samples is extracted and is trained, training set is divided, points 50 groups, every group 500, every time
8 groups are taken to train, 2 groups verify.
Further, detailed process is implemented according to the following steps,
(1) double lead electrocardiogram Wave datas are obtained, and it is 10 seconds to intercept wherein length according to electrocardiogram waveform data
Data are used as original electrocardiographicdigital figure Wave data；
(2) the original electrocardiographicdigital figure Wave data obtained as needed to step (1) carries out denoising, and denoising is used
High-pass filter removes baseline drift noise, and noise jamming is removed using low pass Butterworth filter when noise is too high；
(3) setting convolutional neural networks input layer, hidden layer, the node number of output layer, and setting adjacent layer is each at random
Weight between node；
(4) convolutional neural networks are trained, training data are from the pre-recorded number obtained by means such as physical examinations
According to this and MIT-BIH database datas；Training data are inputted from the input of convolutional neural networks, through convolutional neural networks
After being classified, categorical data is obtained from the output end of convolutional neural networks；The type that convolutional neural networks output end is obtained
The type mark of data and pre-recorded data is compared, the detection error based on output end and true type of waveform, after utilization
The weights of each node in convolutional neural networks are changed to pass-algorithm；
(5) repeat step (4) meets true type of waveform, i.e. convolutional neural networks until convolutional neural networks sentence read result
When structural parameters are restrained, the convolutional neural networks trained are obtained；Convolutional neural networks are in by a large amount of electrocardiographic waves
It is follow-up to the progress of other ECG datas to reach inside waveform separation characteristic element reflection to convolutional neural networks after habit
Correct classification results can be drawn during classification in a short time.
Further, concrete operation step is as follows,
Have 48 records when being trained using MIT-BIH databases, every record is all double leads, by 0.5~
25Hz bandpass filter, filters out High-frequency Interference and baseline drift；In each lead, centered on R ripples, 100 points of R wavefront is taken,
199 points afterwards, totally 300 groups of samples, into a complete cardiac, extract 50000 samples, take 25000 samples including R ripples
It is trained, remaining sample is tested；Training set is divided, divides 50 groups, every group 500, takes 8 groups to train every time, 2 groups are done
Verification；
Three convolution kernels of convolutional neural networks take [2,13] respectively, [2,10] and [2,6]；Three level sampling core be [2,
3], i.e., obscured in 3 consecutive points progress average of an independent lead；Three characteristic face numbers are respectively 6,36 and 72, multilayer sense
The neuron number for knowing the hidden layer of machine is 50, and output layer neuron number is 6；
Double leads constitute 2 × 300 input layer, use 2 × 13 different convolution kernels of 6 initial values, each convolution
Verification input layer does convolution algorithm, can obtain 6 new characteristic faces altogether, each dimension is 2 × 288, this 6 new features
Face is convolutional layer C1；Then convolutional layer C1 is produced using 2 × 3 sampling cores and constitutes sub-sampling layer containing 62 × 96 characteristic faces
S1；
2 characteristic faces adjacent to sub- sample level S1, which are overlapped, takes average combination to obtain 6 new feature faces, ordinate 0-5
S1 6 characteristic faces are represented, the 0th new characteristic face is combined into the 1st with the 0th, records in abscissa, is similarly comprised
Other 5；Then using 62 × 10 convolution kernels, each convolution kernel is produced containing the characteristic face of 36 2 × 87, constitutes convolutional layer
C2, uses 2 × 3 sampling cores to produce 36 2 × 29 characteristic faces composition sub-sampling layer S2 convolutional layer C2 each characteristic face；
The combination of 3 adjacent surfaces is carried out to sub- sample level S2 and obtains 12 new feature faces, then using 62 × 6 convolution kernels, production
The raw convolutional layer C3 containing 72 2 × 24 characteristic faces, the son of 72 2 × 8 characteristic faces is produced to convolutional layer C3 using 2 × 3 sampling cores
Sample level S3；It is the multi-layer perception (MLP) of a basic model from sub-sampling layer S3 to output layer, its middle hidden layer sets 50 god
Through member, it is connected entirely with sub-sampling layer S3；Output layer sets 6 neurons, is connected entirely between hidden layer.
The present invention blends to the feature extraction of ECG signal with pattern classification, it is to avoid complicated and insecure characteristic
According to the problem of extraction.According to existing theory, the intelligent classification of ECG signal is a kind of pattern recognition problem, and conventional solution needs
ECG signal characteristic value constitutive characteristic space is extracted in analysis, then carries out pattern classification to feature space.The extraction of characteristic value is very
Crucial and extremely complex, many documents propose different methods but all do not solve this problem reliably.And convolution is refreshing
Design through network is included in feature extraction in assorting process, it is to avoid the deviation that characteristic extraction procedure occurs.
Convolutional neural networks are enable rationally to handle one-dimensional ECG signal by design.Convolutional neural networks are applied to processing
Two dimensional image signal has excellent pattern classification effect, and its convolution kernel chosen when handling image is generally n × m matrixes,
2 × m convolution kernels are used during present invention processing pair lead ECG signal, due to scheming between the double lead signals of ECG and without being similar to
As the close correlation of adjacent pixel, therefore 2 × m convolution kernels are divided into two 1 × m convolution kernels to two leads in convolution algorithm
Data carry out one-dimensional convolution respectively, operation result still preserves as 2-D data and sends into Multilayer Perception mechanism with two dimensional form
Into output layer produce classification results.
The present invention is based on convolutional neural networks, using capability of fitting of the convolutional neural networks to complex nonlinear function, leads to
Crossing rational design makes it be applied to one-dimensional ECG signal, realizes and ECG signal is more effectively more accurately classified.Accurately dividing
Being capable of real-time monitoring angiocardiopathy people at highest risk, sub-health population, state of an illness crowd undetermined, intelligence on the basis of class ECG signal
Electrocardio change when normal life, work and activity is analyzed, the state of an illness is assisted in, or capture the heart of potential heart disease
Power information, forewarning function is played to patient, brings good social benefit.
Brief description of the drawings
Fig. 1 is the structure chart of exemplary convolution neutral net；
Fig. 2 is that 2 characteristic faces adjacent to sub- sample level S1 are overlapped the combination for taking average combination to obtain 6 new feature faces
Mode coordinate diagram；
Fig. 3 is that the combination coordinate diagram that the combination of 3 adjacent surfaces obtains 12 new feature faces is carried out to sub- sample level S2.
Embodiment
In the present embodiment, the electrocardiosignal classifying identification method is carried out according to the following steps,
(1) double lead electrocardiogram Wave datas are obtained, and it is 10 seconds to intercept wherein length according to electrocardiogram waveform data
Data are used as original electrocardiographicdigital figure Wave data；
(2) as needed, the original electrocardiographicdigital figure Wave data that can be obtained to step (1) carries out denoising, denoising
Baseline drift noise is removed using high-pass filter, noise jamming is removed using low pass Butterworth filter when noise is too high；
(3) setting convolutional neural networks input layer, hidden layer, the node number of output layer, and setting adjacent layer is each at random
Weight between node；The typical structure of the convolutional neural networks is as shown in Figure 1；
(4) convolutional neural networks are trained, training data are from the pre-recorded number obtained by means such as physical examinations
According to this and MIT-BIH database datas.Training data are inputted from the input of convolutional neural networks, through convolutional neural networks
After being classified, categorical data is obtained from the output end of convolutional neural networks.The type that convolutional neural networks output end is obtained
The type mark of data and pre-recorded data is compared, the detection error based on output end and true type of waveform, after utilization
The weights of each node in convolutional neural networks are changed to pass-algorithm；
(5) repeat step (4) meets true type of waveform, i.e. convolutional neural networks until convolutional neural networks sentence read result
When structural parameters are restrained, the convolutional neural networks trained are obtained.Convolutional neural networks are in by a large amount of electrocardiographic waves
After habit, inside waveform separation characteristic element reflection to convolutional neural networks, when subsequently classifying to other ECG datas
Correct classification results can be being drawn in a short time.
Have 48 records when being trained using MIT-BIH databases, every record is all double leads, by 0.5~
25Hz bandpass filter, filters out High-frequency Interference and baseline drift.In each lead, centered on R ripples, 100 points of R wavefront is taken,
199 points afterwards, totally 300 groups of samples, into a complete cardiac, extract 50000 samples, take 25000 samples including R ripples
It is trained, remaining sample is tested.Training set is divided, divides 50 groups, every group 500, takes 8 groups to train every time, 2 groups are done
Verification.
Three convolution kernels of convolutional neural networks take [2,13] respectively, [2,10] and [2,6]；Three level sampling core be [2,
3], i.e., obscured in 3 consecutive points progress average of an independent lead；Three characteristic face numbers are respectively 6,36 and 72, multilayer sense
The neuron number for knowing the hidden layer of machine is 50, and output layer neuron number is 6.
Double leads constitute 2 × 300 input layer, use 2 × 13 different convolution kernels of 6 initial values, each convolution
Verification input layer does convolution algorithm, can obtain 6 new characteristic faces altogether, each dimension is 2 × 288, this 6 new features
Face is convolutional layer C1；Then convolutional layer C1 is produced using 2 × 3 sampling cores and constitutes sub-sampling layer containing 62 × 96 characteristic faces
S1；
2 characteristic faces adjacent to sub- sample level S1, which are overlapped, takes average combination to obtain 6 new feature faces, specific combination
Mode is as shown in Figure 2.
Ordinate 0-5 represents S1 6 characteristic faces in figure, and the 0th new characteristic face is combined into the 1st with the 0th, note
Record is similarly formed other 5 in abscissa.Then using 62 × 10 convolution kernels, each convolution kernel, which is produced, contains 36 2 × 87
Characteristic face, constitute convolutional layer C2,36 2 × 29 characteristic faces produced using 2 × 3 sampling cores to convolutional layer C2 each characteristic face
Constitute sub-sampling layer S2；
The combination of 3 adjacent surfaces is carried out to sub- sample level S2 and obtains 12 new feature faces, specific combination is as shown in Figure 3.
Then using 62 × 6 convolution kernels, the convolutional layer C3 containing 72 2 × 24 characteristic faces is produced, 2 are used to convolutional layer C3
× 3 sampling cores produce the sub-sampling layer S3 of 72 2 × 8 characteristic faces；It is many of basic model from sub-sampling layer S3 to output layer
Layer perceptron, its middle hidden layer sets 50 neurons, is connected entirely with S3；Output layer set 6 neurons, with hidden layer it
Between connect entirely.
The present invention is described in detail above, described above, only the preferred embodiments of the invention, when can not
Limit the application practical range, i.e., it is all to make equivalent changes and modifications according to the application scope, it all should still belong to covering scope of the present invention
It is interior.
Claims (6)
1. a kind of electrocardiosignal classifying identification method, it is characterised in that：Including following implementation steps,
The original electrocardiographicdigital figure Wave data of a, acquirement time of measuring more than 10 seconds, and carried out according to original electrocardiographicdigital figure Wave data
The extraction of electrocardiogram rhythm and pace of moving things information and PQRST waveforms, obtains the digitalized data of electrocardiogram rhythm and pace of moving things information and PQRST waveforms；
B, design construction convolutional neural networks are simultaneously trained to it, and PQRST Wave datas that step a is obtained are from having trained
The input input of convolutional neural networks, after being classified through convolutional neural networks, categorical data is obtained from its output end.
2. electrocardiosignal classifying identification method according to claim 1, it is characterised in that：The original electrocardiographicdigital that step a is used
Figure Wave data, i.e. ECG signal are double lead signals.
3. electrocardiosignal classifying identification method according to claim 1, it is characterised in that：The convolutional Neural of step b designs
Network is laminated and be designed as convolution down-sampling layer by convolutional layer and down-sampling, and three different from neuron number of input layer are rolled up
The output layer that product down-sampling layer is constituted plus multi-layer perception (MLP) is combined into a complete convolutional neural networks；
Wherein convolution down-sampling layer 1 is produced as follows, and first input layer is done by 2 × 13 different convolution kernels of 6 initial values
Convolution, obtains the convolutional layer C1 containing 6 characteristic faces, then use 2 × 36 sub- sample level S1 of sampling cores generation to convolutional layer C1；
Convolution down-sampling layer 2 is produced as follows, and first 2 characteristic faces adjacent to sub- sample level S1, which are combined, obtains 6 new features
Face, then does convolution algorithm therewith using 62 × 10 convolution kernels, the convolutional layer C2 containing 36 characteristic faces is produced, to convolutional layer C2
Each characteristic face using 2 × 3 sampling cores produce 36 sub- sample level S2；Convolution down-sampling layer 3 is produced as follows, first right
Sub-sampling layer S2 carries out the combination of 3 adjacent surfaces and obtains 12 new feature faces, then using 62 × 6 convolution kernels, produces containing 72 spies
The convolutional layer C3 in face is levied, 72 sub- sample level S3 are produced using 2 × 3 sampling cores to convolutional layer C3.
4. electrocardiosignal classifying identification method according to claim 3, it is characterised in that：Step b is to convolutional neural networks
Training convolutional neural networks are carried out using 48 records in MIT-BIH arrhythmia cordis databases；It is recorded as double lead letters
Number, 0.5~25Hz bandpass filter is first passed through, High-frequency Interference and baseline drift is filtered out；In each lead, centered on R ripples,
100 points of R wavefront is taken, totally 300 groups of samples, into a complete cardiac, are extracted 25000 samples and entered including R ripples at latter 199 points
Row training, training set is divided, and is divided 50 groups, every group 500, is taken 8 groups to train every time, 2 groups verify.
5. electrocardiosignal classifying identification method according to claim 1, it is characterised in that：Detailed process is real according to the following steps
Apply,
(1) double lead electrocardiogram Wave datas are obtained, and the data that wherein length is 10 seconds are intercepted according to electrocardiogram waveform data
It is used as original electrocardiographicdigital figure Wave data；
(2) the original electrocardiographicdigital figure Wave data obtained as needed to step (1) carries out denoising, and denoising uses high pass
Wave filter removes baseline drift noise, and noise jamming is removed using low pass Butterworth filter when noise is too high；
(3) setting convolutional neural networks input layer, hidden layer, the node number of output layer, and each node of adjacent layer is set at random
Between weight；
(4) convolutional neural networks are trained, training data from the pre-recorded data obtained by means such as physical examinations with
And MIT-BIH database datas；Training data are inputted from the input of convolutional neural networks, carried out through convolutional neural networks
After classification, categorical data is obtained from the output end of convolutional neural networks；The categorical data that convolutional neural networks output end is obtained
It is compared with the type mark of pre-recorded data, the detection error based on output end and true type of waveform, utilizes backward biography
Pass the weights of each node in algorithm modification convolutional neural networks；
(5) repeat step (4) meets true type of waveform, i.e. convolutional neural networks structure until convolutional neural networks sentence read result
When parameter restrains, the convolutional neural networks trained are obtained；Convolutional neural networks after the study of a large amount of electrocardiographic waves,
Inside waveform separation characteristic element reflection to convolutional neural networks, when subsequently being classified with reaching to other ECG datas
Correct classification results can be drawn in a short time.
6. electrocardiosignal classifying identification method according to claim 5, it is characterised in that：Concrete operation step is as follows,
48 records are had when being trained using MIT-BIH databases, every record is all double leads, by 0.5~25Hz
Bandpass filter, filter out High-frequency Interference and baseline drift；In each lead, centered on R ripples, 100 points of R wavefront, rear 199 are taken
Point, totally 300 groups of samples, into a complete cardiac, extract 50000 samples, take 25000 samples to be instructed including R ripples
Practice, remaining sample is tested；Training set is divided, divides 50 groups, every group 500, takes 8 groups to train every time, 2 groups verify；
Three convolution kernels of convolutional neural networks take [2,13] respectively, [2,10] and [2,6]；Three level sampling core is [2,3], i.e.,
Average is carried out in 3 consecutive points of an independent lead to obscure；Three characteristic face numbers are respectively 6,36 and 72, multi-layer perception (MLP)
The neuron number of hidden layer is 50, and output layer neuron number is 6；
Double leads constitute 2 × 300 input layer, use 2 × 13 different convolution kernels of 6 initial values, each convolution kernel pair
Input layer does convolution algorithm, can obtain 6 new characteristic faces altogether, each dimension is 2 × 288, and this 6 new feature faces are
It is convolutional layer C1；Then convolutional layer C1 is produced using 2 × 3 sampling cores and constitutes sub-sampling layer S1 containing 62 × 96 characteristic faces；
2 characteristic faces adjacent to sub- sample level S1, which are overlapped, takes average combination to obtain 6 new feature faces, and ordinate 0-5 is represented
S1 6 characteristic faces, the 0th new characteristic face is combined into the 0th with the 1st, is recorded in abscissa, is similarly comprised in addition
5；Then using 62 × 10 convolution kernels, each convolution kernel is produced containing the characteristic face of 36 2 × 87, constitutes convolutional layer C2, right
Convolutional layer C2 each characteristic face produces 36 2 × 29 characteristic faces using 2 × 3 sampling cores and constitutes sub-sampling layer S2；
The combination of 3 adjacent surfaces is carried out to sub- sample level S2 and obtains 12 new feature faces, then using 62 × 6 convolution kernels, generation contains
The convolutional layer C3 of 72 2 × 24 characteristic faces, the sub-sampling of 72 2 × 8 characteristic faces is produced to convolutional layer C3 using 2 × 3 sampling cores
Layer S3；It is the multi-layer perception (MLP) of a basic model from sub-sampling layer S3 to output layer, its middle hidden layer sets 50 neurons,
It is connected entirely with sub-sampling layer S3；Output layer sets 6 neurons, is connected entirely between hidden layer.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710402483.1A CN107184198A (en) | 2017-06-01 | 2017-06-01 | A kind of electrocardiosignal classifying identification method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710402483.1A CN107184198A (en) | 2017-06-01 | 2017-06-01 | A kind of electrocardiosignal classifying identification method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107184198A true CN107184198A (en) | 2017-09-22 |
Family
ID=59876299
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710402483.1A Pending CN107184198A (en) | 2017-06-01 | 2017-06-01 | A kind of electrocardiosignal classifying identification method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107184198A (en) |
Cited By (29)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107510452A (en) * | 2017-09-30 | 2017-12-26 | 扬美慧普（北京）科技有限公司 | A kind of ECG detecting method based on multiple dimensioned deep learning neutral net |
| CN108039203A (en) * | 2017-12-04 | 2018-05-15 | 北京医拍智能科技有限公司 | The detecting system of arrhythmia cordis based on deep neural network |
| CN108030488A (en) * | 2017-11-30 | 2018-05-15 | 北京医拍智能科技有限公司 | The detecting system of arrhythmia cordis based on convolutional neural networks |
| CN108090509A (en) * | 2017-12-13 | 2018-05-29 | 四川大学 | A kind of adaptive electrocardiogram sorting technique of data length |
| CN108175402A (en) * | 2017-12-26 | 2018-06-19 | 智慧康源（厦门）科技有限公司 | The intelligent identification Method of electrocardiogram (ECG) data based on residual error network |
| CN108256452A (en) * | 2018-01-06 | 2018-07-06 | 天津大学 | A kind of method of the ECG signal classification of feature based fusion |
| CN108256453A (en) * | 2018-01-06 | 2018-07-06 | 天津大学 | A kind of method based on one-dimensional ECG signal extraction two dimension CNN features |
| CN108403107A (en) * | 2018-02-06 | 2018-08-17 | 北京大学深圳研究生院 | A kind of arrhythmia cordis method of discrimination and system |
| CN108766557A (en) * | 2018-05-12 | 2018-11-06 | 鲁东大学 | Automatic arrhythmia analysis method based on channel signal fused neural network |
| CN108836312A (en) * | 2018-07-13 | 2018-11-20 | 希蓝科技（北京）有限公司 | A kind of method and system of the novel progress clutter rejecting based on artificial intelligence |
| CN108926341A (en) * | 2018-04-20 | 2018-12-04 | 平安科技（深圳）有限公司 | Detection method, device, computer equipment and the storage medium of ECG signal |
| CN108960050A (en) * | 2018-05-25 | 2018-12-07 | 东软集团股份有限公司 | Classification model training method, ECG data classifying method, device and equipment |
| CN108937912A (en) * | 2018-05-12 | 2018-12-07 | 鲁东大学 | A kind of automatic arrhythmia analysis method based on deep neural network |
| CN109036573A (en) * | 2018-07-06 | 2018-12-18 | 苏州大学 | Electrocardiogram based on production confrontation network technology generates and classification method |
| CN109063552A (en) * | 2018-06-22 | 2018-12-21 | 深圳大学 | A kind of multi-lead electrocardiosignal classification method and system |
| CN109044339A (en) * | 2018-08-13 | 2018-12-21 | 重庆工商大学 | A kind of ECG signal diagnostic method of limit convolutional neural networks |
| CN109254107A (en) * | 2018-11-01 | 2019-01-22 | 广州城市职业学院 | The general tea Fast Classification discrimination method of mandarin orange |
| CN109567790A (en) * | 2018-12-24 | 2019-04-05 | 深圳市瑞康宏业科技开发有限公司 | A kind of quick partial analysis method, apparatus of electrocardio scatter plot and computer installation |
| WO2019100560A1 (en) * | 2017-11-27 | 2019-05-31 | 乐普（北京）医疗器械股份有限公司 | Artificial intelligence self-learning-based automatic electrocardiography analysis method and apparatus |
| CN110236520A (en) * | 2019-05-20 | 2019-09-17 | 上海数创医疗科技有限公司 | ECG type recognition methods and device based on double convolutional neural networks |
| CN110522440A (en) * | 2019-08-12 | 2019-12-03 | 广州视源电子科技股份有限公司 | Electrocardiosignal recognition device based on grouping convolution neural network |
| CN110720894A (en) * | 2019-11-08 | 2020-01-24 | 广州视源电子科技股份有限公司 | Atrial flutter detection method, device, equipment and storage medium |
| CN111448614A (en) * | 2017-11-02 | 2020-07-24 | 皇家飞利浦有限公司 | Method and apparatus for analyzing echocardiograms |
| CN111956212A (en) * | 2020-07-29 | 2020-11-20 | 鲁东大学 | Inter-group atrial fibrillation identification method based on frequency domain filtering-multi-mode deep neural network |
| CN112587147A (en) * | 2020-11-11 | 2021-04-02 | 上海数创医疗科技有限公司 | Atrial premature beat target detection method based on convolutional neural network |
| CN112587149A (en) * | 2020-11-11 | 2021-04-02 | 上海数创医疗科技有限公司 | Atrial premature beat target detection device based on convolutional neural network |
| CN113349790A (en) * | 2020-10-16 | 2021-09-07 | 山东大学 | Heart beat classification method and system of 2D convolutional neural network based on S-shaped reconstruction |
| CN115862843A (en) * | 2022-12-12 | 2023-03-28 | 哈尔滨医科大学 | Auxiliary identification system and equipment for cardiac troponin elevation type and cardiovascular disease |
| RU2823433C1 (en) * | 2023-11-21 | 2024-07-23 | Общество С Ограниченной Ответственностью "Компания "Стрим Лабс" | Method of processing and analysing electrocardiogram (ecg) data |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101088456A (en) * | 2007-07-13 | 2007-12-19 | 天津大学 | ECG feature extraction method based on evolutionary wavelet domain Wiener deconvolution |
| CN104970789A (en) * | 2014-04-04 | 2015-10-14 | 中国科学院苏州纳米技术与纳米仿生研究所 | Electrocardiogram classification method and system |
| CN105726018A (en) * | 2016-02-06 | 2016-07-06 | 河北大学 | Automatic atrial fibrillation detection method irrelevant to RR interphase |
| CN105748063A (en) * | 2016-04-25 | 2016-07-13 | 山东大学齐鲁医院 | Intelligent arrhythmia diagnosis method based on multiple-lead and convolutional neural network |
| CN106108889A (en) * | 2016-07-20 | 2016-11-16 | 杨平 | Electrocardiogram classification method based on degree of depth learning algorithm |
| CN106344005A (en) * | 2016-10-28 | 2017-01-25 | 张珈绮 | Mobile ECG (electrocardiogram) monitoring system and monitoring method |
| WO2017072250A1 (en) * | 2015-10-27 | 2017-05-04 | CardioLogs Technologies | An automatic method to delineate or categorize an electrocardiogram |
-
2017
- 2017-06-01 CN CN201710402483.1A patent/CN107184198A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101088456A (en) * | 2007-07-13 | 2007-12-19 | 天津大学 | ECG feature extraction method based on evolutionary wavelet domain Wiener deconvolution |
| CN104970789A (en) * | 2014-04-04 | 2015-10-14 | 中国科学院苏州纳米技术与纳米仿生研究所 | Electrocardiogram classification method and system |
| WO2017072250A1 (en) * | 2015-10-27 | 2017-05-04 | CardioLogs Technologies | An automatic method to delineate or categorize an electrocardiogram |
| CN105726018A (en) * | 2016-02-06 | 2016-07-06 | 河北大学 | Automatic atrial fibrillation detection method irrelevant to RR interphase |
| CN105748063A (en) * | 2016-04-25 | 2016-07-13 | 山东大学齐鲁医院 | Intelligent arrhythmia diagnosis method based on multiple-lead and convolutional neural network |
| CN106108889A (en) * | 2016-07-20 | 2016-11-16 | 杨平 | Electrocardiogram classification method based on degree of depth learning algorithm |
| CN106344005A (en) * | 2016-10-28 | 2017-01-25 | 张珈绮 | Mobile ECG (electrocardiogram) monitoring system and monitoring method |
Cited By (41)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107510452B (en) * | 2017-09-30 | 2019-10-08 | 扬美慧普（北京）科技有限公司 | A kind of ECG detecting method based on multiple dimensioned deep learning neural network |
| CN107510452A (en) * | 2017-09-30 | 2017-12-26 | 扬美慧普（北京）科技有限公司 | A kind of ECG detecting method based on multiple dimensioned deep learning neutral net |
| CN111448614B (en) * | 2017-11-02 | 2024-05-28 | 皇家飞利浦有限公司 | Method and apparatus for analyzing echocardiography |
| CN111448614A (en) * | 2017-11-02 | 2020-07-24 | 皇家飞利浦有限公司 | Method and apparatus for analyzing echocardiograms |
| WO2019100560A1 (en) * | 2017-11-27 | 2019-05-31 | 乐普（北京）医疗器械股份有限公司 | Artificial intelligence self-learning-based automatic electrocardiography analysis method and apparatus |
| US11529103B2 (en) | 2017-11-27 | 2022-12-20 | Shanghai Lepu CloudMed Co., LTD | Artificial intelligence self-learning-based automatic electrocardiography analysis method and apparatus |
| CN108030488A (en) * | 2017-11-30 | 2018-05-15 | 北京医拍智能科技有限公司 | The detecting system of arrhythmia cordis based on convolutional neural networks |
| CN108039203A (en) * | 2017-12-04 | 2018-05-15 | 北京医拍智能科技有限公司 | The detecting system of arrhythmia cordis based on deep neural network |
| CN108090509A (en) * | 2017-12-13 | 2018-05-29 | 四川大学 | A kind of adaptive electrocardiogram sorting technique of data length |
| CN108090509B (en) * | 2017-12-13 | 2021-10-08 | 四川大学 | A data-length adaptive method for electrocardiogram classification |
| CN108175402A (en) * | 2017-12-26 | 2018-06-19 | 智慧康源（厦门）科技有限公司 | The intelligent identification Method of electrocardiogram (ECG) data based on residual error network |
| CN108256453A (en) * | 2018-01-06 | 2018-07-06 | 天津大学 | A kind of method based on one-dimensional ECG signal extraction two dimension CNN features |
| CN108256452A (en) * | 2018-01-06 | 2018-07-06 | 天津大学 | A kind of method of the ECG signal classification of feature based fusion |
| CN108403107A (en) * | 2018-02-06 | 2018-08-17 | 北京大学深圳研究生院 | A kind of arrhythmia cordis method of discrimination and system |
| CN108403107B (en) * | 2018-02-06 | 2019-12-31 | 北京大学深圳研究生院 | Method and system for identifying arrhythmia |
| CN108926341A (en) * | 2018-04-20 | 2018-12-04 | 平安科技（深圳）有限公司 | Detection method, device, computer equipment and the storage medium of ECG signal |
| CN108937912A (en) * | 2018-05-12 | 2018-12-07 | 鲁东大学 | A kind of automatic arrhythmia analysis method based on deep neural network |
| CN108766557A (en) * | 2018-05-12 | 2018-11-06 | 鲁东大学 | Automatic arrhythmia analysis method based on channel signal fused neural network |
| CN108766557B (en) * | 2018-05-12 | 2021-07-20 | 鲁东大学 | Automatic arrhythmia analysis method based on channel signal fusion neural network |
| CN108937912B (en) * | 2018-05-12 | 2021-05-11 | 鲁东大学 | Automatic arrhythmia analysis method based on deep neural network |
| CN108960050A (en) * | 2018-05-25 | 2018-12-07 | 东软集团股份有限公司 | Classification model training method, ECG data classifying method, device and equipment |
| CN109063552B (en) * | 2018-06-22 | 2021-03-30 | 深圳大学 | Multi-lead electrocardiosignal classification method and system |
| CN109063552A (en) * | 2018-06-22 | 2018-12-21 | 深圳大学 | A kind of multi-lead electrocardiosignal classification method and system |
| CN109036573A (en) * | 2018-07-06 | 2018-12-18 | 苏州大学 | Electrocardiogram based on production confrontation network technology generates and classification method |
| CN108836312A (en) * | 2018-07-13 | 2018-11-20 | 希蓝科技（北京）有限公司 | A kind of method and system of the novel progress clutter rejecting based on artificial intelligence |
| CN109044339A (en) * | 2018-08-13 | 2018-12-21 | 重庆工商大学 | A kind of ECG signal diagnostic method of limit convolutional neural networks |
| CN109254107B (en) * | 2018-11-01 | 2020-09-04 | 广州城市职业学院 | Rapid classification and identification method for citrus Pu' er tea |
| CN109254107A (en) * | 2018-11-01 | 2019-01-22 | 广州城市职业学院 | The general tea Fast Classification discrimination method of mandarin orange |
| CN109567790A (en) * | 2018-12-24 | 2019-04-05 | 深圳市瑞康宏业科技开发有限公司 | A kind of quick partial analysis method, apparatus of electrocardio scatter plot and computer installation |
| CN110236520B (en) * | 2019-05-20 | 2020-09-29 | 上海数创医疗科技有限公司 | Electrocardiogram type recognition device based on double convolution neural network |
| CN110236520A (en) * | 2019-05-20 | 2019-09-17 | 上海数创医疗科技有限公司 | ECG type recognition methods and device based on double convolutional neural networks |
| CN110522440A (en) * | 2019-08-12 | 2019-12-03 | 广州视源电子科技股份有限公司 | Electrocardiosignal recognition device based on grouping convolution neural network |
| CN110720894A (en) * | 2019-11-08 | 2020-01-24 | 广州视源电子科技股份有限公司 | Atrial flutter detection method, device, equipment and storage medium |
| CN111956212A (en) * | 2020-07-29 | 2020-11-20 | 鲁东大学 | Inter-group atrial fibrillation identification method based on frequency domain filtering-multi-mode deep neural network |
| CN111956212B (en) * | 2020-07-29 | 2023-08-01 | 鲁东大学 | Inter-group atrial fibrillation recognition method based on frequency domain filtering-multi-mode deep neural network |
| CN113349790A (en) * | 2020-10-16 | 2021-09-07 | 山东大学 | Heart beat classification method and system of 2D convolutional neural network based on S-shaped reconstruction |
| CN112587149A (en) * | 2020-11-11 | 2021-04-02 | 上海数创医疗科技有限公司 | Atrial premature beat target detection device based on convolutional neural network |
| CN112587147A (en) * | 2020-11-11 | 2021-04-02 | 上海数创医疗科技有限公司 | Atrial premature beat target detection method based on convolutional neural network |
| CN115862843A (en) * | 2022-12-12 | 2023-03-28 | 哈尔滨医科大学 | Auxiliary identification system and equipment for cardiac troponin elevation type and cardiovascular disease |
| CN115862843B (en) * | 2022-12-12 | 2024-02-02 | 哈尔滨医科大学 | Auxiliary identification system and equipment for myocardial troponin elevation type and cardiovascular diseases |
| RU2823433C1 (en) * | 2023-11-21 | 2024-07-23 | Общество С Ограниченной Ответственностью "Компания "Стрим Лабс" | Method of processing and analysing electrocardiogram (ecg) data |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107184198A (en) | A kind of electrocardiosignal classifying identification method | |
| Wang et al. | Arrhythmia classification algorithm based on multi-head self-attention mechanism | |
| CN107951485B (en) | Dynamic electrocardiogram analysis method and device based on artificial intelligence self-learning | |
| US12186086B2 (en) | Systems and methods for reduced lead electrocardiogram diagnosis using deep neural networks and rule-based systems | |
| CN104970789B (en) | Electrocardiogram sorting technique and system | |
| CN106108889B (en) | Electrocardiogram classification method based on deep learning algorithm | |
| CN114224342B (en) | A multi-channel EEG signal emotion recognition method based on spatiotemporal fusion feature network | |
| CN109222963A (en) | A kind of anomalous ecg method for identifying and classifying based on convolutional neural networks | |
| CN109893118A (en) | A kind of electrocardiosignal classification diagnosis method based on deep learning | |
| CN108256453A (en) | A kind of method based on one-dimensional ECG signal extraction two dimension CNN features | |
| CN110313894A (en) | Arrhythmia cordis sorting algorithm based on convolutional neural networks | |
| CN110522412A (en) | A Method for Classifying EEG Signals Based on Multiscale Brain Functional Networks | |
| CN113901893A (en) | Recognition and classification of ECG signals based on multiple cascaded deep neural networks | |
| Abibullaev et al. | A brute-force CNN model selection for accurate classification of sensorimotor rhythms in BCIs | |
| Tseng et al. | Sliding large kernel of deep learning algorithm for mobile electrocardiogram diagnosis | |
| CN115881308A (en) | Method for constructing heart disease classification model based on spatiotemporal joint detection | |
| Wang et al. | Multiscale residual network based on channel spatial attention mechanism for multilabel ECG classification | |
| Gupta et al. | Pre-processing based ECG signal analysis using emerging tools | |
| Alagarsamy et al. | Performing the classification of pulsation cardiac beats automatically by using CNN with various dimensions of kernels | |
| Zhao et al. | PVC recognition for wearable ECGs using modified frequency slice wavelet transform and convolutional neural network | |
| Krishnakumar et al. | Detection of arrhythmia and congestive heart failure through classification of ECG signals using deep learning neural network | |
| Shi et al. | Automated heartbeat classification based on convolutional neural network with multiple kernel sizes | |
| Nursalim et al. | Classification of electrocardiogram signal using deep learning models | |
| Kan et al. | A Novel PSO‐Based Optimized Lightweight Convolution Neural Network for Movements Recognizing from Multichannel Surface Electromyogram | |
| CN114081502B (en) | A non-invasive heart disease diagnosis method and device based on machine 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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170922 |
|
| RJ01 | Rejection of invention patent application after publication |