CN109645980A - A kind of rhythm abnormality classification method based on depth migration study - Google Patents
A kind of rhythm abnormality classification method based on depth migration study Download PDFInfo
- Publication number
- CN109645980A CN109645980A CN201811353814.8A CN201811353814A CN109645980A CN 109645980 A CN109645980 A CN 109645980A CN 201811353814 A CN201811353814 A CN 201811353814A CN 109645980 A CN109645980 A CN 109645980A
- Authority
- CN
- China
- Prior art keywords
- feature
- data
- adaptation layer
- network
- fusion
- 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
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/316—Modalities, i.e. specific diagnostic methods
- A61B5/318—Heart-related electrical modalities, e.g. electrocardiography [ECG]
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—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]
- A61B5/346—Analysis of electrocardiograms
- A61B5/349—Detecting specific parameters of the electrocardiograph cycle
- A61B5/352—Detecting R peaks, e.g. for synchronising diagnostic apparatus; Estimating R-R interval
-
- 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
- A61B5/7203—Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal
-
- 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
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Biophysics (AREA)
- General Health & Medical Sciences (AREA)
- Cardiology (AREA)
- Veterinary Medicine (AREA)
- Public Health (AREA)
- Animal Behavior & Ethology (AREA)
- Surgery (AREA)
- Molecular Biology (AREA)
- Medical Informatics (AREA)
- Pathology (AREA)
- Biomedical Technology (AREA)
- Heart & Thoracic Surgery (AREA)
- Artificial Intelligence (AREA)
- Signal Processing (AREA)
- Psychiatry (AREA)
- Physiology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Fuzzy Systems (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The invention discloses a kind of rhythm abnormality classification methods based on depth migration study, the described method includes: treated ECG signal to be passed through to the subtended network configuration of 1D-CNN and LSTM, the space characteristics of signal and temporal characteristics are combined and extract ECG signal feature;By the connection type of Multiscale Fusion, small scale features up-sampling is interpolated into and carries out Fusion Features on the scale of adjacent characteristic pattern equal sizes;The feature of source domain and target numeric field data is subjected to feature difference adjustment by adaptation layer, the feature that source domain data are exported from adaptation layer is calculated and is lost by the MMD between the adaptation layer source domain exported and target domain characterization by full articulamentum and Softmax classifier calculated Classification Loss;Last combining classification loss and the common adjustment network parameter of MMD loss.The present invention improves the accuracy rate of a variety of arrhythmia cordis identifications of not same area ECG data, meets the needs in practical application.
Description
Technical field
The present invention relates to area of pattern recognition, more particularly to one kind to carry out feature extraction, feature to not same area electrocardiosignal
Fusion, feature migration, network migration and the classification method for solving a variety of rhythm abnormalities.
Background technique
Rhythm abnormality be the human heart electric signal rhythm and pace of moving things or conduction be interfered caused by the not normal phenomenon of cardiac function, such as
Fruit patient does not have found in time, it is dead to cause heart failure even all standing, therefore rhythm abnormality diagnosis is highly important.
Currently, most of doctors diagnose the rhythm of the heart shape of patient using the mode of observation electrocardiogram (electrocardiogram, ECG)
State, but observation electrocardiogram is uninteresting time-consuming for a long time, is easy to cause mistake.In recent years, the area of computer aided of rhythm abnormality
Diagnostic system appears in academic research and clinical trial, it can efficiently and accurately position abnormal problem, by numerous doctors
Shield personnel favor.
The ECG identifying system of clinical application is divided into three parts: data prediction, signal characteristic abstraction and signal point
Class.What wherein data prediction and Modulation recognition were each responsible for is the ECG signal denoising by sensor acquisition and carries out to signal
It distinguishes, research method is more mature.And since the effect of signal characteristic abstraction directly determines classification performance, and it is more multiple
Miscellaneous, different scholars proposes different solutions from different angles, these methods divide temporal analysis, statistics side
Method, signal converter technique.
Wherein, time-domain analysis be mainly used for extract ECG signal morphological feature, such as two adjacent waves the peak separation R,
QRS mixed recharge[1], duration of T wave etc..This mode intuitively can rapidly extract the change information of waveform, but cannot
Enough extract the potential minutia of signal profound level.Statistical method, such as High-order Cumulant, are able to suppress Gaussian noise,
It is usually used in extracting the implicit nonlinear characteristic of signal.Based on the method for signal transformation, such as Short Time Fourier Transform, wavelet transformation, S
Transformation etc. analyzes electrocardiosignal by time-space domain and frequency domain information.
Deep learning is also increasingly used for electrocardiosignal due to its powerful nonlinear fitting and layer-by-layer characteristic present ability
Classification field.SayantanG etc. proposes that the depth confidence network being made of Boltzmann machine (RBM) extracts ECG feature.
A.Rahhal etc. extracts feature using denoising self-encoding encoder (DAE), and net, network are finely tuned in conjunction with methods of actively studying.Ren Xiaoxia is visited
Dropout depth convolutional neural networks are studied carefully in the classification performance of electrocardiosignal ST wave band.
Current existing ECG recognizer is faced with two challenges: in feature extraction, conventional method is for multi-class classification
Performance is relatively low, and although some deep learning algorithms strengthen character representation, but have ignored the stream between feature timing and information
The general character.Secondly as the physiological function of test patient, motion state, medicining condition, the difference for testing environment, electrocardiosignal are difficult
To be ensured of from same data distribution.So most of electrocardiogram (ECG) datas are from not same area, data under clinical scene
Between there is domain differences to make the effect of model training and test bad, this is that the classification of current rhythm abnormality needs what is solved to ask
Topic.
Summary of the invention
The present invention provides a kind of rhythm abnormality classification method based on depth migration study, the present invention improves not same area
The accuracy rate of a variety of arrhythmia cordis identifications of ECG data, described below:
A kind of rhythm abnormality classification method based on depth migration study, which comprises
The subtended network configuration that treated ECG signal is passed through to 1D-CNN and LSTM, by the space characteristics of signal and when
Between feature combine extract ECG signal feature;
By the connection type of Multiscale Fusion, small scale features up-sampling is interpolated into and adjacent characteristic pattern equal sizes
Scale on carry out Fusion Features;
The feature of source domain and target numeric field data is subjected to feature difference adjustment by adaptation layer, source domain data are from adaptive
The feature of layer output calculates the source exported by adaptation layer by full articulamentum and Softmax classifier calculated Classification Loss
MMD loss between domain and target domain characterization;Last combining classification loss and the common adjustment network parameter of MMD loss.
Wherein, small scale features up-sampling is interpolated into and adjacent feature by the connection type by Multiscale Fusion
Fusion Features are carried out on the scale of figure equal sizes specifically:
The feature arest neighbors method of zonule block is up-sampled 2 times and obtains the feature big with big region unit etc., then with greatly
The feature of region unit carries out the stacking on channel space;
It is polymerize with 1 × 1 convolution on the level of channel, increases the feature mobility of interchannel;Same another zonule block
Multi-scale feature fusion, the spatial information of feature-rich are carried out again with the feature of heap poststack.
Further, the depth migration network that the method proposes is mirror-image structure, utilizes identical feature extraction structure
Extract different interpatient electrocardiosignal features.
Wherein, the method also includes:
Adaptation layer is added in a network, adjusts source domain electrocardiogram (ECG) data and aiming field electrocardio by adaptation layer when training network
Difference between data.
The beneficial effect of the technical scheme provided by the present invention is that:
1, the present invention proposes a kind of new rhythm abnormality classification method for different numeric field datas, utilizes in terms of feature extraction
The cascade structure of 1D-CNN (one-dimensional convolutional neural networks) and LSTM (long memory network in short-term) extract local time in deep learning
Sequence characteristics;
2, the present invention devises the connection type of multi-scale feature fusion in terms of Fusion Features, by different size of feature
Figure carries out polymerization and convolution operation on a passage, to obtain fusion feature;
3, difference between the feature of the invention for reducing different numeric field datas using transfer learning thought, the algorithm is in training process
In, network can Fast Learning feature and with cracking speed convergence, and improve a variety of rhythms of the heart of not same area ECG data and lose
The other accuracy rate of common sense.
Detailed description of the invention
Fig. 1 is a kind of flow chart of rhythm abnormality classification method based on depth migration study;
Fig. 2 is the schematic diagram of the connection block diagram of multi-scale feature fusion;
Fig. 3 is the schematic diagram of the depth migration network architecture.
Specific embodiment
To make the object, technical solutions and advantages of the present invention clearer, embodiment of the present invention is made below further
Ground detailed description.
The embodiment of the present invention proposes a kind of depth migration learning network, for not same area rhythm abnormality classification task: first
It first passes through the long short-term memory cascade structure of one-dimensional convolution sum and extracts electrocardiosignal temporal aspect.To make full use of different characteristic space
On information, improve the ability to express of feature, the embodiment of the present invention proposes the connection type of Multiscale Fusion, and small scale is special
Sign up-sampling is interpolated into the big scale such as adjacent characteristic pattern and carries out Fusion Features.
For data distribution difference problem, the depth migration network that the embodiment of the present invention proposes is mirror-image structure, that is, is utilized
The different interpatient electrocardiosignal features of identical feature extraction structure extraction.Meter is exported to it by way of adaptation layer is added
Largest Mean poor (maximum mean discrepancy, MMD) loss is calculated, to adjust property field distributional difference.
Common optimization network finally is lost using Classification Loss and Largest Mean difference, resolvability is big between training class,
The small feature of otherness between domain.
The embodiment of the present invention proposes a kind of rhythm abnormality sorter network based on depth migration study, will extract feature, spy
Sign fusion, the processing of property field difference and classification ensemble to together, experiments verify that, different patients of this method in same database
Between and disparate databases between all show higher electrocardiosignal classification performance.
Embodiment 1
Referring to Fig. 1, the rhythm abnormality classification method provided in an embodiment of the present invention based on depth migration network, including it is following
Step:
101: pretreatment:
ECG original signal is pre-processed, eliminate noise to the interference of electrocardiosignal and detects R wave, extracts two sections of the peak R
Data be heartbeat segment and to carry out z-score standardization.
Wherein, the step of standardization is known to those skilled in the art, and the embodiment of the present invention does not repeat them here this.
102: feature extraction:
The subtended network configuration that treated ECG signal is passed through to 1D-CNN and LSTM, by the space characteristics of signal and when
Between feature combine extract ECG signal feature.
103: Fusion Features:
Different size of characteristic pattern is reached into equal sizes by way of up-sampling, and is stacked on feature channel, then
The feature on channel layer is further merged by convolution operation.
104: transfer learning:
For the data of not same area, using the thought of model migration and parameter migration in transfer learning, it is special to reduce not same area
Difference between sign makes it meet same data distribution as far as possible, is conducive to next classification task.
105: more classification: utilizing the rhythm abnormality waveform of softmax identification plurality of classes.
In conclusion the embodiment of the present invention proposes the connection type of Multiscale Fusion, small scale features are up-sampled and are inserted
It is worth the big scale such as adjacent characteristic pattern and carries out Fusion Features, improves the standard of a variety of arrhythmia cordis identifications of not same area ECG data
True rate.
Embodiment 2
The scheme in embodiment 1 is further introduced below with reference to specific calculation formula, Fig. 1-Fig. 3, is detailed in
It is described below:
201:ECG signal data comes from the collected data of electrocardiograph or public database;
202: pretreatment:
In terms of noise remove, baseline drift is removed using median filtering method, it is dry using low-pass filter removal power line
It disturbs and high-frequency noise.The detection of R wave is using amplitude threshold, wavelet threshold etc..
203: feature extraction:
The position R that is detected according to pretreatment takes 197 after taking 90 sampled point signals and R wave position before R wave position
A sampled point signal, the ECG segment of totally 288 sampled points.Devise one-dimensional convolution (1D-CNN) and long short-term memory (LSTM)
The feature extraction block of cascade structure extracts ECG signal feature in spatial domain and time-domain.
Wherein, convolutional network is widely used in two dimensional image, extracts depth abstract characteristics.And for ECG signal, this hair
Bright embodiment extracts feature using one-dimensional convolution.Convolution operation is widely used in the signal processing, is meant that unit is rung
Weighted superposition on Ying Yi function.For one-dimensional ECG signal, the one-dimensional convolution checking signal of different size and parameter is designed
Superposition is weighted to extract new feature, different filters can be regarded as to a certain extent to signal filtering operation.Input
Signal setting are as follows: xi=[x1,x2,...xn], n is the number of specimen sample point, one-dimensional convolution operation calculating process:
Wherein,Indicate the output of corresponding j-th of the convolution kernel of l layers of convolutional layer, MjIndicate the receiving of Current neural member
Domain,Indicate l layers of corresponding i-th of the weight coefficient of j-th of convolution kernel,Indicate that l j-th of convolution kernel of layer is corresponding
Bias coefficient.
Wherein, f indicates activation primitive, extracts nonlinear characteristic, uses ReLU function here:
F (x)=max (0, x)
The maximum value of x feature vector adjacent element is calculated using maximum pond layer (max-pooling layer) to obtain
Sparse features vector, and improve the characterization ability of translation invariance.
For the timing for excavating ECG feature, the embodiment of the present invention extracts timing spy with long short-term memory (LSTM) network
Sign, the feature extracted in one-dimensional convolution process be broken down into sequence components and be fed to duplicate length in short-term memory unit to carry out
Time analysis.LSTM network structure[2]As follows: input gate learns the information being stored in memory unit C;Forget a selection memory
Forget in unit or the information content of re -training;When out gate study exports stored information.
The embodiment of the present invention proposes three feature extraction blocks (Block), respectively extract electrocardiosignal shallow-layer information and
The deep information.The convolution kernel that the deconvolution parameter of 1st Block is 16 1 × 5, step-length 1, filling up is 2;2nd Block's
The convolution kernel that deconvolution parameter is 32 1 × 5, step-length 1, filling up is 1;The volume that the deconvolution parameter of 3rd Block is 64 1 × 3
Product core, step-length 1, filling up is 1.Wherein all pond layer core sizes are 1 × 2, step-length 2.
204: Fusion Features:
In order to increase feature rich degree, the embodiment of the present invention devises the connection type for proposing multi-scale feature fusion to tie
Close the feature that different characteristic extracts block.Referring to shown in Fig. 2, here there are three characteristic pattern, the characteristic pattern size of Block 3 is
The feature arest neighbors method of Block 3 is up-sampled 2 times first and obtained by the 1/2 of Block2 to merge this two-part feature
The big feature with Block 2 etc., then the stacking (Concatenation) on channel space is carried out with the feature of Block 2, later
It is polymerize with 1 × 1 convolution on the level of channel, increases the feature mobility of interchannel.
Equally, Block1 and the fused feature of Block2 and Block3 (i.e. the feature of heap poststack) carry out spy again
Sign fusion, concrete mode are that the feature for merging Block2 and Block3 is upsampled to the feature big with Block1 etc., then with
The feature of Block1 is stacked on a passage, is polymerize later with 1 × 1 convolution on the level of channel, the space of feature-rich
Information
205: transfer learning:
Electrocardiosignal is by sensor acquisition come, different location, different conditions, varying environment, distinct device acquisition
To ecg signal data distribution on have certain deviation, and interpatient physiological function is different, there are patient-specific,
It is difficult to ensure that trained and test data with distribution, therefore, needs to consider in network design process data distribution difference band
The domain offset problem come.
The embodiment of the present invention designs mirror symmetry network, in a manner of identical characteristic processing (feature extraction, Fusion Features)
Extract the ECG feature of source domain and aiming field.To excavate the common information between different characteristic of field, the difference of feature between reduction domain,
Adaptation layer (Adaptionlayer) is separately added into after feature extraction, which is made of fully-connected network.
The Largest Mean difference loss (MMD loss) of design of the embodiment of the present invention is measurement criterion.Wherein, defined feature
It is transformed to φ (), the data of source domain are xs∈XS, target numeric field data is xt∈XT, then MMD loses is defined as:
The network not only needs to minimize the difference between domain, also to excavate the feature of discrimination between class, this is conducive to
The training of classifier.To meet the two standards, the embodiment of the present invention devises associated losses function:
L=LC(XL,y)+λMMD2(XS,XT)
Wherein, LC(XL, y) and it is Classification Loss, XLFor prediction result, y be true value label.MMD(XS,XT) it is source domain data
XSWith target numeric field data XTDistance, hyper parameter λ balance both loss intensity.
Depth migration network structure is as shown in Figure 3, wherein source domain is with the data of mark, and aiming field is without mark
Data, the characteristic extraction part of the two share network weight, and all data participate in calculating Largest Mean difference and lose, and only band
The sample of mark carries out the calculating of Classification Loss.
206: more classification: last in network realizes a variety of arrhythmia cordis identifications with softmax layers.
Weight is initialized in training process first, the error of predicted value and true value is calculated by propagated forward, and by random
Gradient descent method minimizes loss function and carries out backpropagation updates each layer parameter of network again.This process is repeated until error
No longer smaller, network tends to restrain.
In conclusion the embodiment of the present invention proposes the connection type of Multiscale Fusion, small scale features are up-sampled and are inserted
It is worth the big scale such as adjacent characteristic pattern and carries out Fusion Features, improves the standard of a variety of arrhythmia cordis identifications of not same area ECG data
True rate.
Embodiment 3
Feasibility verifying is carried out to the scheme in Examples 1 and 2 below with reference to table 1, described below:
MIT-BIH (- Beth-Israel hospital, Massachusetts Institute Technology) arrhythmia cordis database includes 48 ECG notes
Record, every record is had recorded by two kinds of different type leads (being denoted as A, B lead) is about 30 minutes data, and sample rate is
360Hz;Wherein the A lead of 45 records uses MLII lead, remaining uses V5 lead;The B lead of 40 records is led using V1
Connection, remaining uses II, V2, V4 and V5 lead.
The ECG data collection that this experiment the uses data that MLII lead records in MIT database, it is total according to AAMI standard
There are 5 kinds of hearts to clap type: ectopic beat (supraventricular ectopic on normal beats (normal beats), room
Beats), ventricular ectopic beat (ventricular ectopic beats), fusion heartbeat (fusion beats), Wu Fafen
((unclassifiable beats) classifies to this 5 kinds of heartbeat types for class heartbeat.
Table 1 gives the performance classified using different extraction characterization methods, and experimental result is as follows:
Table 1: the classification performance under different characteristic extracting method
It can be seen from Table 1 that the depth migration network that the embodiment of the present invention is proposed is in rhythm abnormality classification task
It is good using 1D-CNN and LSTM extraction feature, the network class effect of fusion feature to be substantially better than, and meets in practical application
It needs.
Bibliography:
[1] where field electrocardiogram detailed annotation with diagnosis publishing house, [M] Zhejiang University, 2010
[2]Gers F A,Schmidhuber J,Cummins F.Learning to forget:continual
prediction with LSTM[J].Neural Computation,2014,12(10):2451-2471.
It will be appreciated by those skilled in the art that attached drawing is the schematic diagram of a preferred embodiment, the embodiments of the present invention
Serial number is for illustration only, does not represent the advantages or disadvantages of the embodiments.
The foregoing is merely presently preferred embodiments of the present invention, is not intended to limit the invention, it is all in spirit of the invention and
Within principle, any modification, equivalent replacement, improvement and so on be should all be included in the protection scope of the present invention.
Claims (4)
1. a kind of rhythm abnormality classification method based on depth migration study, which is characterized in that the described method includes:
The subtended network configuration that treated ECG signal is passed through to 1D-CNN and LSTM, the space characteristics of signal and time are special
Sign, which combines, extracts ECG signal feature;
By the connection type of Multiscale Fusion, small scale features up-sampling is interpolated into the ruler with adjacent characteristic pattern equal sizes
Fusion Features are carried out on degree;
The feature of source domain and target numeric field data is subjected to feature difference adjustment by adaptation layer, source domain data are defeated from adaptation layer
Feature out by full articulamentum and Softmax classifier calculated Classification Loss, calculate the source domain that export by adaptation layer with
MMD loss between target domain characterization;Last combining classification loss and the common adjustment network parameter of MMD loss.
2. a kind of rhythm abnormality classification method based on depth migration study according to claim 1, which is characterized in that institute
The connection type by Multiscale Fusion is stated, small scale features up-sampling is interpolated into the scale with adjacent characteristic pattern equal sizes
Upper carry out Fusion Features specifically:
The feature arest neighbors method of zonule block is up-sampled 2 times and obtains the feature big with big region unit etc., then with big region
The feature of block carries out the stacking on channel space;
It is polymerize with 1 × 1 convolution on the level of channel, increases the feature mobility of interchannel;Same another zonule block and heap
The feature of poststack carries out multi-scale feature fusion, the spatial information of feature-rich again.
3. a kind of rhythm abnormality classification method based on depth migration study according to claim 1, which is characterized in that institute
The depth migration network for stating method proposition is mirror-image structure, utilizes the different interpatient electrocardios of identical feature extraction structure extraction
Signal characteristic.
4. a kind of rhythm abnormality classification method based on depth migration study according to claim 1, which is characterized in that institute
State method further include:
Adaptation layer is added in a network, adjusts source domain electrocardiogram (ECG) data and aiming field electrocardiogram (ECG) data by adaptation layer when training network
Between difference.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811353814.8A CN109645980A (en) | 2018-11-14 | 2018-11-14 | A kind of rhythm abnormality classification method based on depth migration study |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811353814.8A CN109645980A (en) | 2018-11-14 | 2018-11-14 | A kind of rhythm abnormality classification method based on depth migration study |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109645980A true CN109645980A (en) | 2019-04-19 |
Family
ID=66111246
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811353814.8A Pending CN109645980A (en) | 2018-11-14 | 2018-11-14 | A kind of rhythm abnormality classification method based on depth migration study |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109645980A (en) |
Cited By (28)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110090016A (en) * | 2019-04-28 | 2019-08-06 | 心医国际数字医疗***(大连)有限公司 | The method and system of positioning R wave position, the R wave automatic testing method using LSTM neural network |
CN110135510A (en) * | 2019-05-22 | 2019-08-16 | 电子科技大学中山学院 | Dynamic domain self-adaptive method, equipment and computer readable storage medium |
CN110169768A (en) * | 2019-07-08 | 2019-08-27 | 河北大学 | A kind of automatic noise-reduction method of electrocardiosignal |
CN110263684A (en) * | 2019-06-06 | 2019-09-20 | 山东省计算中心(国家超级计算济南中心) | Electrocardiogram classification method based on lightweight neural network |
CN110490095A (en) * | 2019-07-31 | 2019-11-22 | 中国人民解放军战略支援部队信息工程大学 | A kind of multi-modal Fusion Features Modulation Identification method and system neural network based |
CN110555388A (en) * | 2019-08-06 | 2019-12-10 | 浙江大学 | CNN and LSTM-based method for constructing intracardiac abnormal excitation point positioning model |
CN110569928A (en) * | 2019-09-23 | 2019-12-13 | 深圳大学 | Micro Doppler radar human body action classification method of convolutional neural network |
CN110598677A (en) * | 2019-10-08 | 2019-12-20 | 电子科技大学 | Space-time multi-channel deep learning system for automatic modulation recognition |
CN110623658A (en) * | 2019-09-24 | 2019-12-31 | 京东方科技集团股份有限公司 | Signal processing method, signal processing apparatus, medical device, and storage medium |
CN110638430A (en) * | 2019-10-23 | 2020-01-03 | 苏州大学 | Multi-task cascade neural network ECG signal arrhythmia disease classification model and method |
CN110693486A (en) * | 2019-09-27 | 2020-01-17 | 武汉中旗生物医疗电子有限公司 | Electrocardiogram abnormity labeling method and device |
CN110897630A (en) * | 2019-06-27 | 2020-03-24 | 上海数创医疗科技有限公司 | Ventricular tachycardia and heart rhythm identification method and device based on transfer learning |
CN111401166A (en) * | 2020-03-06 | 2020-07-10 | 中国科学技术大学 | Robust gesture recognition method based on electromyographic information decoding |
CN111460953A (en) * | 2020-03-26 | 2020-07-28 | 山东省人工智能研究院 | Electrocardiosignal classification method based on self-adaptive learning of countermeasure domain |
CN111493850A (en) * | 2020-04-13 | 2020-08-07 | 中国科学院深圳先进技术研究院 | Blood pressure measuring method and device based on deep neural network |
CN111904411A (en) * | 2020-08-25 | 2020-11-10 | 浙江工业大学 | Multi-lead heartbeat signal classification method and device based on multi-scale feature extraction |
CN112315482A (en) * | 2020-11-25 | 2021-02-05 | 新智数字科技有限公司 | Electrocardio information determination method and device, readable storage medium and electronic equipment |
CN112365978A (en) * | 2020-11-10 | 2021-02-12 | 北京航空航天大学 | Method and device for establishing early risk assessment model of tachycardia event |
CN112529035A (en) * | 2020-10-30 | 2021-03-19 | 西南电子技术研究所(中国电子科技集团公司第十研究所) | Intelligent identification method for identifying individual types of different radio stations |
CN112932431A (en) * | 2021-01-26 | 2021-06-11 | 山西三友和智慧信息技术股份有限公司 | Heart rate identification method based on 1DCNN + Inception Net + GRU fusion network |
CN112971800A (en) * | 2021-01-27 | 2021-06-18 | 东南大学 | Deep learning-based electrocardio premature beat signal processing method, module and device |
CN113116361A (en) * | 2021-03-09 | 2021-07-16 | 山东大学 | Sleep staging method based on single-lead electroencephalogram |
CN113158364A (en) * | 2021-04-02 | 2021-07-23 | 中国农业大学 | Circulating pump bearing fault detection method and system |
CN113554013A (en) * | 2021-09-22 | 2021-10-26 | 华南理工大学 | Cross-scene recognition model training method, cross-scene road recognition method and device |
CN113990499A (en) * | 2021-10-22 | 2022-01-28 | 安徽心之声医疗科技有限公司 | Heart age assessment method and device, electronic equipment and storage medium |
CN114022049A (en) * | 2021-12-10 | 2022-02-08 | 萍乡市圣迈互联网科技有限公司 | Intelligent service information risk processing method and system based on cloud computing |
CN114521900A (en) * | 2022-02-21 | 2022-05-24 | 贵州省人民医院 | Arrhythmia classification and identification method based on transfer learning |
CN114699080A (en) * | 2022-04-28 | 2022-07-05 | 电子科技大学 | Driver mental stress degree identification method based on fusion characteristics |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102988042A (en) * | 2012-08-30 | 2013-03-27 | 重庆电子工程职业学院 | Method for identifying sino atrial node electrogram based on integration of wavelet transform and support vector machine |
CN104367317A (en) * | 2014-10-15 | 2015-02-25 | 北京理工大学 | Electrocardiogram electrocardiosignal classification method with multi-scale characteristics combined |
US20170032221A1 (en) * | 2015-07-29 | 2017-02-02 | Htc Corporation | Method, electronic apparatus, and computer readable medium of constructing classifier for disease detection |
CN106901723A (en) * | 2017-04-20 | 2017-06-30 | 济南浪潮高新科技投资发展有限公司 | A kind of electrocardiographic abnormality automatic diagnosis method |
CN107811626A (en) * | 2017-09-10 | 2018-03-20 | 天津大学 | A kind of arrhythmia classification method based on one-dimensional convolutional neural networks and S-transformation |
CN108710919A (en) * | 2018-05-25 | 2018-10-26 | 东南大学 | A kind of crack automation delineation method based on multi-scale feature fusion deep learning |
-
2018
- 2018-11-14 CN CN201811353814.8A patent/CN109645980A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102988042A (en) * | 2012-08-30 | 2013-03-27 | 重庆电子工程职业学院 | Method for identifying sino atrial node electrogram based on integration of wavelet transform and support vector machine |
CN104367317A (en) * | 2014-10-15 | 2015-02-25 | 北京理工大学 | Electrocardiogram electrocardiosignal classification method with multi-scale characteristics combined |
US20170032221A1 (en) * | 2015-07-29 | 2017-02-02 | Htc Corporation | Method, electronic apparatus, and computer readable medium of constructing classifier for disease detection |
CN106901723A (en) * | 2017-04-20 | 2017-06-30 | 济南浪潮高新科技投资发展有限公司 | A kind of electrocardiographic abnormality automatic diagnosis method |
CN107811626A (en) * | 2017-09-10 | 2018-03-20 | 天津大学 | A kind of arrhythmia classification method based on one-dimensional convolutional neural networks and S-transformation |
CN108710919A (en) * | 2018-05-25 | 2018-10-26 | 东南大学 | A kind of crack automation delineation method based on multi-scale feature fusion deep learning |
Non-Patent Citations (5)
Title |
---|
TSUNG-YI LIN,等: "Feature Pyramid Networks for Object Detection", 《ARXIV》 * |
张忠怀: "《自动化应用技术》", 31 July 2001 * |
张景祥,等: "融合异构特征的子空间迁移学习算法", 《自动化学报》 * |
邢晨: "基于深度学习的高光谱遥感图像分类", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
马宇婷: "基于深度学习的领域适应问题研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Cited By (38)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110090016A (en) * | 2019-04-28 | 2019-08-06 | 心医国际数字医疗***(大连)有限公司 | The method and system of positioning R wave position, the R wave automatic testing method using LSTM neural network |
CN110135510A (en) * | 2019-05-22 | 2019-08-16 | 电子科技大学中山学院 | Dynamic domain self-adaptive method, equipment and computer readable storage medium |
CN110263684A (en) * | 2019-06-06 | 2019-09-20 | 山东省计算中心(国家超级计算济南中心) | Electrocardiogram classification method based on lightweight neural network |
CN110263684B (en) * | 2019-06-06 | 2021-02-09 | 山东省计算中心(国家超级计算济南中心) | Electrocardiogram classification method based on lightweight neural network |
CN110897630A (en) * | 2019-06-27 | 2020-03-24 | 上海数创医疗科技有限公司 | Ventricular tachycardia and heart rhythm identification method and device based on transfer learning |
CN110169768A (en) * | 2019-07-08 | 2019-08-27 | 河北大学 | A kind of automatic noise-reduction method of electrocardiosignal |
CN110490095A (en) * | 2019-07-31 | 2019-11-22 | 中国人民解放军战略支援部队信息工程大学 | A kind of multi-modal Fusion Features Modulation Identification method and system neural network based |
CN110555388A (en) * | 2019-08-06 | 2019-12-10 | 浙江大学 | CNN and LSTM-based method for constructing intracardiac abnormal excitation point positioning model |
WO2021022763A1 (en) * | 2019-08-06 | 2021-02-11 | 浙江大学 | Method for constructing intracardiac abnormal activation point location model based on cnn and lstm |
CN110569928A (en) * | 2019-09-23 | 2019-12-13 | 深圳大学 | Micro Doppler radar human body action classification method of convolutional neural network |
CN110623658A (en) * | 2019-09-24 | 2019-12-31 | 京东方科技集团股份有限公司 | Signal processing method, signal processing apparatus, medical device, and storage medium |
CN110623658B (en) * | 2019-09-24 | 2022-07-26 | 京东方科技集团股份有限公司 | Signal processing method, signal processing apparatus, medical device, and storage medium |
CN110693486A (en) * | 2019-09-27 | 2020-01-17 | 武汉中旗生物医疗电子有限公司 | Electrocardiogram abnormity labeling method and device |
CN110598677A (en) * | 2019-10-08 | 2019-12-20 | 电子科技大学 | Space-time multi-channel deep learning system for automatic modulation recognition |
CN110638430A (en) * | 2019-10-23 | 2020-01-03 | 苏州大学 | Multi-task cascade neural network ECG signal arrhythmia disease classification model and method |
CN111401166A (en) * | 2020-03-06 | 2020-07-10 | 中国科学技术大学 | Robust gesture recognition method based on electromyographic information decoding |
CN111460953B (en) * | 2020-03-26 | 2021-05-18 | 山东省人工智能研究院 | Electrocardiosignal classification method based on self-adaptive learning of countermeasure domain |
CN111460953A (en) * | 2020-03-26 | 2020-07-28 | 山东省人工智能研究院 | Electrocardiosignal classification method based on self-adaptive learning of countermeasure domain |
CN111493850A (en) * | 2020-04-13 | 2020-08-07 | 中国科学院深圳先进技术研究院 | Blood pressure measuring method and device based on deep neural network |
CN111904411A (en) * | 2020-08-25 | 2020-11-10 | 浙江工业大学 | Multi-lead heartbeat signal classification method and device based on multi-scale feature extraction |
CN112529035A (en) * | 2020-10-30 | 2021-03-19 | 西南电子技术研究所(中国电子科技集团公司第十研究所) | Intelligent identification method for identifying individual types of different radio stations |
CN112365978A (en) * | 2020-11-10 | 2021-02-12 | 北京航空航天大学 | Method and device for establishing early risk assessment model of tachycardia event |
CN112315482A (en) * | 2020-11-25 | 2021-02-05 | 新智数字科技有限公司 | Electrocardio information determination method and device, readable storage medium and electronic equipment |
CN112932431A (en) * | 2021-01-26 | 2021-06-11 | 山西三友和智慧信息技术股份有限公司 | Heart rate identification method based on 1DCNN + Inception Net + GRU fusion network |
CN112932431B (en) * | 2021-01-26 | 2022-09-27 | 山西三友和智慧信息技术股份有限公司 | Heart rate identification method based on 1DCNN + Inception Net + GRU fusion network |
CN112971800A (en) * | 2021-01-27 | 2021-06-18 | 东南大学 | Deep learning-based electrocardio premature beat signal processing method, module and device |
CN112971800B (en) * | 2021-01-27 | 2023-01-03 | 东南大学 | Deep learning-based electrocardio premature beat signal processing method, module and device |
CN113116361A (en) * | 2021-03-09 | 2021-07-16 | 山东大学 | Sleep staging method based on single-lead electroencephalogram |
CN113158364A (en) * | 2021-04-02 | 2021-07-23 | 中国农业大学 | Circulating pump bearing fault detection method and system |
CN113158364B (en) * | 2021-04-02 | 2024-03-22 | 中国农业大学 | Method and system for detecting bearing faults of circulating pump |
CN113554013A (en) * | 2021-09-22 | 2021-10-26 | 华南理工大学 | Cross-scene recognition model training method, cross-scene road recognition method and device |
CN113990499A (en) * | 2021-10-22 | 2022-01-28 | 安徽心之声医疗科技有限公司 | Heart age assessment method and device, electronic equipment and storage medium |
CN114022049B (en) * | 2021-12-10 | 2022-07-22 | 佛山市蜂王人力资源有限公司 | Intelligent service information risk processing method and system based on cloud computing |
CN114022049A (en) * | 2021-12-10 | 2022-02-08 | 萍乡市圣迈互联网科技有限公司 | Intelligent service information risk processing method and system based on cloud computing |
CN114521900A (en) * | 2022-02-21 | 2022-05-24 | 贵州省人民医院 | Arrhythmia classification and identification method based on transfer learning |
CN114521900B (en) * | 2022-02-21 | 2022-11-15 | 贵州省人民医院 | Arrhythmia classification and identification method based on transfer learning |
CN114699080A (en) * | 2022-04-28 | 2022-07-05 | 电子科技大学 | Driver mental stress degree identification method based on fusion characteristics |
CN114699080B (en) * | 2022-04-28 | 2023-04-25 | 电子科技大学 | Driver mental stress degree identification method based on fusion characteristics |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109645980A (en) | A kind of rhythm abnormality classification method based on depth migration study | |
Lu et al. | Feature fusion for imbalanced ECG data analysis | |
Oh et al. | Automated diagnosis of arrhythmia using combination of CNN and LSTM techniques with variable length heart beats | |
Golrizkhatami et al. | ECG classification using three-level fusion of different feature descriptors | |
Martis et al. | Computer aided diagnosis of atrial arrhythmia using dimensionality reduction methods on transform domain representation | |
CN109620211A (en) | A kind of intelligent abnormal electrocardiogram aided diagnosis method based on deep learning | |
CN107811626A (en) | A kind of arrhythmia classification method based on one-dimensional convolutional neural networks and S-transformation | |
Cao et al. | Atrial fibrillation detection using an improved multi-scale decomposition enhanced residual convolutional neural network | |
He et al. | Recognition of ECG patterns using artificial neural network | |
Orhan | Real-time CHF detection from ECG signals using a novel discretization method | |
CN109602417A (en) | Sleep stage method and system based on random forest | |
CN110313894A (en) | Arrhythmia cordis sorting algorithm based on convolutional neural networks | |
CN108647584B (en) | Arrhythmia identification and classification method based on sparse representation and neural network | |
CN111202512A (en) | Electrocardiogram classification method and device based on wavelet transformation and DCNN | |
CN114532993B (en) | Automatic detection method for electroencephalogram high-frequency oscillation signals of epileptic patients | |
Wang et al. | Multi-class arrhythmia detection based on neural network with multi-stage features fusion | |
Ma et al. | An effective data enhancement method for classification of ECG arrhythmia | |
WO2022135449A1 (en) | Interictal epileptiform discharge activity detection apparatus and method for epileptic patient | |
Ka | ECG beats classification using waveform similarity and RR interval | |
Peláez et al. | Ischemia classification via ECG using MLP neural networks | |
CN113116300A (en) | Physiological signal classification method based on model fusion | |
Prakash et al. | A system for automatic cardiac arrhythmia recognition using electrocardiogram signal | |
Kumari et al. | Performance evaluation of neural networks and adaptive neuro fuzzy inference system for classification of cardiac arrhythmia | |
Chandrasekar et al. | Detection of arrhythmia from electrocardiogram signals using a novel gaussian assisted signal smoothing and pattern recognition | |
Zhang et al. | Automatic diagnosis of arrhythmia with electrocardiogram using multiple instance learning: From rhythm annotation to heartbeat prediction |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190419 |