CN110037685A - With the portable electrocardiograph for improving convolutional neural networks recognizer - Google Patents
With the portable electrocardiograph for improving convolutional neural networks recognizer Download PDFInfo
- Publication number
- CN110037685A CN110037685A CN201910257532.6A CN201910257532A CN110037685A CN 110037685 A CN110037685 A CN 110037685A CN 201910257532 A CN201910257532 A CN 201910257532A CN 110037685 A CN110037685 A CN 110037685A
- Authority
- CN
- China
- Prior art keywords
- electrocardiosignal
- module
- neural networks
- convolutional neural
- layer
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
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
- A61B5/361—Detecting fibrillation
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
- A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
Abstract
This application involves a kind of with the portable electrocardiograph for improving convolutional neural networks recognizer, is identified using preparatory trained neural network to electrocardiosignal, to identify the corresponding rhythm of the heart type of the electrocardiosignal, determines whether it is atrial fibrillation.Specifically, by using loss function are as follows:Convolutional neural networks, convolutional neural networks can be made to increase punishment to false negative in training using the loss function, to reduce omission factor under the premise of realizing and guaranteeing accuracy rate, improve accuracy.
Description
Technical field
The application belongs to the field of medical instrument technology, has more particularly, to one kind and improves convolutional neural networks recognizer
Portable electrocardiograph.
Background technique
Heart is in each cardiac cycle, and since pacemaker, atrium, ventricle are in succession excited, excitement can occur for cardiac muscle, is swashing
During dynamic, faint bioelectricity can be generated, each cardiac cycle of heart is along with biological Electrical change.This biology
Electrical change can be passed to each position of body surface.Since body parts tissue is different, the distance away from heart is different, electrocardio
Signal is also different in the current potential that the different position of body is shown.For normal heart, the direction of this biology Electrical change,
Frequency, intensity are regular.
Doctor according to the relative time relationship between the form, wave amplitude size and each wave of the electrocardiographic wave recorded,
Again compared with normal ECG, it just can be diagnosed to be heart disease.Such as cardiac arrhythmia, myocardial infarction, proiosystole, high blood
Pressure, ectocardia beating etc..
Electrocardiograph is the equipment for carrying out checkout and diagnosis for the electrocardiogram to heart, and portable electrocardiograph is a kind of small
The ecg equipment of type, electrocardiograph is come out the electrical signal detection of body surface by electrode, laggard by the amplification of amplifier
Electrocardiogram can be obtained in row record.
Convolutional neural networks (Convolutional Neural Networks, CNN) are a kind of comprising convolutional calculation and tool
There is the feedforward neural network (Feedforward Neural Networks) of depth structure.In recent years, convolutional neural networks are utilized
Etc. neural networks obtained more and more applications come the atrial fibrillation type identified in electrocardiogram.Such as the Master's thesis of Xu Xiaoyan
In " the atrial fibrillation recognizer research based on machine learning ", the side that atrial fibrillation identification is carried out using convolutional neural networks has been expressly recited
Method.In the Master's thesis " being detected based on the atrial fibrillation of atrial activity feature and convolutional neural networks " of Han little Cen, also to using convolution
Neural network carries out atrial fibrillation knowledge method for distinguishing and is studied, however is that existing convolutional neural networks are straight in these papers
It scoops out on atrial fibrillation identifies, does not carry out any optimization for the identification of atrial fibrillation, this causes it to identify atrial fibrillation in electrocardiogram
Accuracy rate not can be further improved.
Summary of the invention
The technical problem to be solved by the present invention is for solve in the prior art in electrocardiogram atrial fibrillation identify accuracy rate compared with
Low, the higher deficiency of omission factor, to provide the electrocardiogram that the accuracy rate that atrial fibrillation identifies in a kind of pair of electrocardiogram is high, omission factor is low
Instrument.
The technical solution adopted by the present invention to solve the technical problems is:
A kind of portable electrocardiograph with improvement convolutional neural networks recognizer, comprising:
Main control unit, for being handled electrocardiosignal and exporting processing result；
Electrocardio signal collecting unit carries out operation amplifier for receiving the electrocardiosignal of human body, and to electrocardiosignal, by mould
Quasi- signal is transferred to main control unit after being converted to digital signal, and the electrocardio signal collecting unit includes analog signal conditioner mould
Block, collection of simulant signal module and digital filtering module；
Memory module, for receiving electrocardiosignal that main control unit is sent and the electrocardiosignal of storage being transferred to master
Control unit；
Alarm module, for issuing alarm signal；
Display module, for showing the processing result of electrocardiosignal；
Input module, for being instructed to main control unit input control；
Power module is connected with battery, and to main control unit, memory module, electrocardio signal collecting unit, alarm module and
Display module power supply；
The main control unit can pre-process electrocardiosignal after receiving electrocardiosignal, and will be after pretreatment
Electrocardiosignal be input in advance trained neural network and identified, export the corresponding rhythm of the heart type of the electrocardiosignal,
In, the neural network uses loss function are as follows:Convolutional neural networks,
Wherein, c indicates loss size, and n indicates that number of samples, x are sample, and y is the corresponding label of sample x, and a is by being with sample x
The output size for inputting and being gone out with current neural computing, α, β are regularization coefficient.
Preferably, the portable electrocardiograph with improvement convolutional neural networks recognizer of the invention, the storage
Module is TF card；The wireless communication module is bluetooth module, Wifi module, 3G/4G module.
Preferably, the portable electrocardiograph with improvement convolutional neural networks recognizer of the invention, the master control
Unit is STM32F412RG chip, have in the piece of the STM32F412RG chip FPU module, A/D module, SPI module,
SDIO module, UART module and SRAM space；
The electrocardio signal collecting unit is the ADS1293 chip of Texas Instrument.
Preferably, the portable electrocardiograph with improvement convolutional neural networks recognizer of the invention, further includes ring
Border optical sensor, the ambient light sensor for incuding ambient light intensity, can when ambient light intensity is lower than setting value to
Main control unit sends signal；
The main control unit can close the display module when receiving the signal of ambient light sensor sending.
Preferably, the portable electrocardiograph with improvement convolutional neural networks recognizer of the invention, the nerve
Network includes 1 input layer, multiple convolutional layers and pond layer, 1 full articulamentum and 1 classifier layer, the loss function
Middle α=0.8-0.9, β=1.1-1.2.
Preferably, the portable electrocardiograph with improvement convolutional neural networks recognizer of the invention,
The electrocardiosignal is pre-processed, comprising the following steps:
Whether the sample frequency for judging the electrocardiosignal is predeterminated frequency；
When the sample frequency is not the predeterminated frequency, using interpolation method by the electrocardiosignal resampling is described
The electrocardiosignal of predeterminated frequency.
Preferably, the portable electrocardiograph with improvement convolutional neural networks recognizer of the invention, it is described preparatory
Trained neural network is obtained by following steps:
S31: tranining database is obtained, tranining database is the electrocardiogram (ECG) data for being known as atrial fibrillation or non-atrial fibrillation；
S32: electrocardiogram (ECG) data is pre-processed；
S33: being trained with convolutional neural networks, and convolutional neural networks are by several convolutional layers and several pond layers, one
Full articulamentum and a classifier layer are constituted, and input the result of known atrial fibrillation or non-atrial fibrillation when training in classifier layer.
Preferably, the portable electrocardiograph with improvement convolutional neural networks recognizer of the invention,
In S31 step, using the atrial fibrillation electrocardiosignal and at least 10,000 mixed uniformly other types of at least 10,000 10s
Electrocardiosignal as training data formed tranining database, using 0 and 1 as atrial fibrillation and the electrocardiosignal of non-atrial fibrillation mark
Label；
In S32 step, electrocardiogram (ECG) data is carried out using the fir filter that upper lower limiting frequency is respectively 0.1Hz, 100Hz
Filtering, if electrocardiosignal sample frequency is not 500Hz, use closest interpolation method by electrocardiosignal resampling for
500Hz；
In S33 step, convolutional neural networks have 8 layer networks, and 1-6 layer are layer1-layer6 in 8 layer networks,
It is made of a convolutional layer and a pond layer, convolutional layer includes 5 cores in layer1, and convolution kernel size is 224, layer1
Step-length and core size in middle pond layer are 2；Layer2 convolutional layer includes 5 cores, and convolution kernel size is 112, layer2
Step-length and core size in middle pond layer are 2；Layer3 convolutional layer includes 10 cores, and convolution kernel size is 100, layer3
Step-length and core size in middle pond layer are 2；Layer4 convolutional layer includes 10 cores, and convolution kernel size is 50, layer4
Step-length and core size in middle pond layer are 2；Layer5 convolutional layer includes 10 cores, and convolution kernel size is 48, layer5
Step-length and core size in middle pond layer are 2；Layer6 convolutional layer includes 10 cores, and convolution kernel size is 24, layer4
Step-length and core size in middle pond layer are 2；The output Characteristic Number of layer6 is 30, eventually passes through full articulamentum
Layer7 exports 10 features after calculating.
Preferably, the portable electrocardiograph with improvement convolutional neural networks recognizer of the invention,
Trained neural network includes first nerves network and nervus opticus network in advance, and the first nerves network uses
Loss function are as follows:Convolutional neural networks, wherein c indicate loss size, n
Indicate that number of samples, x are sample, y is the corresponding label of sample x, and a is by being input and with current nerve with sample x
The output size that network query function goes out, α, β are regularization coefficient；α=1-1.2 in the loss function, β=0.8-1, electrocardio letter
Number corresponding rhythm of the heart type is exported after first nerves Network Recognition；
If rhythm of the heart type is non-atrial fibrillation, then it is assumed that the electrocardiosignal is non-atrial fibrillation；
If rhythm of the heart type is atrial fibrillation, which is input to nervus opticus trained in advance again
Network is identified；
Nervus opticus network uses loss function are as follows:Convolutional Neural net
Network, wherein c indicates loss size, and n indicates that number of samples, x are sample, and y is the corresponding label of sample x, and a is by with sample
X is input and the output size with current neural computing out, and α, β are regularization coefficient；α in the loss function
=0.8-1, β=1.0-1.2；The corresponding rhythm of the heart type of the electrocardiosignal through nervus opticus Network Recognition is exported, with second
The rhythm of the heart type of neural network recognization is the rhythm of the heart type of the electrocardiosignal.
Preferably, the portable electrocardiograph with improvement convolutional neural networks recognizer of the invention,
First nerves network is using α=1.2 in loss function, β=1；Nervus opticus network using α in loss function=
0.8, β=1.2.
The beneficial effects of the present invention are:
Of the invention has the portable electrocardiograph for improving convolutional neural networks recognizer, using trained in advance
Neural network identifies electrocardiosignal, to identify the corresponding rhythm of the heart type of the electrocardiosignal, determines whether it is atrial fibrillation.
Specifically, by using loss function are as follows:Convolutional neural networks, use this
Loss function can make convolutional neural networks increase the punishment to false negative in training, to realize the premise for guaranteeing accuracy rate
Under, omission factor is reduced, accuracy is improved.
Detailed description of the invention
The technical solution of the application is further illustrated with reference to the accompanying drawings and examples.
Fig. 1 is the circuit structure diagram of the electrocardiograph of the embodiment of the present application；
Fig. 2 is the schematic diagram of the improvement convolutional neural networks of the embodiment of the present application.
Appended drawing reference in figure are as follows:
1- main control unit；2- ambient light sensor；3- memory module；4- electrocardio signal collecting unit；41- analog signal tune
Manage module；42- collection of simulant signal module；43- digital filtering module；5- alarm module；61- display module；62- button；7-
Wireless communication module；8- power module；9- battery.
Specific embodiment
It should be noted that in the absence of conflict, the features in the embodiments and the embodiments of the present application can phase
Mutually combination.
In the description of the present application, it should be noted that unless otherwise clearly defined and limited, term " installation ", " phase
Even ", " connection " shall be understood in a broad sense, for example, it may be being fixedly connected, may be a detachable connection, or be integrally connected；It can
To be mechanical connection, it is also possible to be electrically connected；It can be directly connected, can also can be indirectly connected through an intermediary
Connection inside two elements.For the ordinary skill in the art, above-mentioned term can be understood by concrete condition
Concrete meaning in this application.
It is described in detail the technical solution of the application below with reference to the accompanying drawings and in conjunction with the embodiments.
Embodiment
The present embodiment provides a kind of portable electrocardiographs, as shown in Figure 1, comprising:
Main control unit 1, for being handled electrocardiosignal and being exported processing result, the main control unit 1 is
STM32F412RG chip, have in the piece of STM32F412RG chip FPU module, A/D module, SPI module, SDIO module,
The SRAM space of UART module and 256k Byte, and it is small in size, it is (5.6mA@20MHz) low in energy consumption；
Electrocardio signal collecting unit 4 carries out operation amplifier for receiving the electrocardiosignal of human body, and to electrocardiosignal, will
Analog signal is transferred to main control unit 1 after being converted to digital signal, and the electrocardio signal collecting unit 4 includes analog signal conditioner
Module 41, collection of simulant signal module 42 and digital filtering module 43, the electrocardio signal collecting unit 4 are Texas Instrument
ADS1293 chip, the combination of ADS1293 built-in chip type multichannel amplifier, crossbar switch, 24Bit ADC mould group, right leg drive and benefit
It repays, dot circuit in Wilson's, digital filtering unit, 5 lead applications can be realized with monomer, which also there is register setting to adopt
Sample frequency realizes the sample frequency of 1kHz under conditions of being not take up the clock sources of control chip；
Memory module 3, for receiving electrocardiosignal that main control unit 1 is sent and being transferred to the electrocardiosignal of storage
Main control unit 1, the memory module 3 are TF storage card etc., and memory module 3 is connected in the SDIO module of main control unit 1；
Alarm module 5, for issuing alarm signal, alarm module 5 can be LED light, speaker, when the processing of electrocardiosignal
When being as a result abnormal, user is reminded；
Display module 61, it is job information, exception information and alarm content, secondary for showing the processing result of electrocardiosignal
Number etc.；
Input module 62 can be button, such as booting, shutdown, selection for instructing to 1 input control of main control unit；
Wireless communication module 7, the processing result for being communicated with intelligent terminal and by main control unit 1 to electrocardiosignal
It is sent to intelligent terminal, bluetooth module, Wifi module, 3G/4G module, NB-LOT communication module etc., intelligent terminal can be electricity
Brain, smart phone or Internet server；
Ambient light sensor 2 can be when ambient light intensity be lower than setting value to master control list for incuding ambient light intensity
Member 1 sends signal, such as the environment light using the similar model such as light treasured/LITEON LTR-303ALS-DR ambient light sensor
Sensor；
Power module 8 is connected with battery 9, and to main control unit 1, memory module 3, electrocardio signal collecting unit 4, alarm
Module 5 and display module 61 are powered, and power module is powered to entire electrocardiograph；
The main control unit 1 can close the display module 61 when receiving the signal of the sending of ambient light sensor 2.
The portable electrocardiograph of the present embodiment is provided with ambient light sensor 2, passes through 2 inductance loop of ambient light sensor
Border luminous intensity, when in use due to portable electrocardiograph, when being blocked by clothes, ambient light sensor 2 incudes ambient light intensity
It is low, signal is sent to main control unit 1 when ambient light intensity is lower than setting value when ambient light sensor 2 incudes, main control unit 1 is connecing
Close the display module 61 when receiving the signal of the sending of ambient light sensor 2 makes just to stop the power consumption of display module 61
The formula electrocardiograph of taking once charges, and that the time can be used is longer.
The UART module of the main control unit 1 turns UART chip by USB and connect with USB interface, to realize the portable heart
Electrograph instrument realizes the communication with host computer, facilitates and carries out check and correction or data-transformation facility.
Improve the training method of convolutional neural networks are as follows:
S1: tranining database is obtained, tranining database is the electrocardiogram (ECG) data for being known as atrial fibrillation or non-atrial fibrillation, and electrocardiogram (ECG) data is long
Degree can be 10s, and preferably 4s, the 4s time can include at least 6 heartbeats, can improve in the case where guaranteeing accuracy rate
Recognition efficiency, electrocardiogram (ECG) data length when training neural network and electrocardiosignal to be identified is isometric in the future；
Using the atrial fibrillation electrocardiosignal and at least 10,000 mixed uniformly other types electrocardiosignals works of at least 10,000 10s
For training data formed tranining database, wherein 0 and 1 respectively as atrial fibrillation and the electrocardiosignal of non-atrial fibrillation label；
S2: electrocardiogram (ECG) data is pre-processed；
Electrocardiogram (ECG) data is pre-processed；Place is filtered to the electrocardiosignal using the filter of default cutoff frequency
Reason.
The sample frequency of the electrocardiosignal can also be pre-processed, comprising the following steps:
Whether the sample frequency for judging the electrocardiosignal is predeterminated frequency；
When the sample frequency is not the predeterminated frequency, using interpolation method by the electrocardiosignal resampling is described
The electrocardiosignal of predeterminated frequency.
Such as: when pretreatment, it is filtered using the fir filter that upper lower limiting frequency is respectively 0.1Hz, 100Hz,
If electrocardiosignal sample frequency is not 500Hz, use closest interpolation method by electrocardiosignal resampling for 500Hz.
S3: the convolutional neural networks (CNN) of 8 layer networks are trained, the structure of convolutional neural networks (CNN) is 8 layers
1-6 layers (layer1-layer6) in network is made of a convolutional layer and a pond layer；Convolutional layer packet in layer1
Containing 5 cores, convolution kernel size is step-length in 224, layer1 in the layer of pond and core size is 2；Layer2 convolutional layer packet
Containing 5 cores, convolution kernel size is step-length in 112, layer2 in the layer of pond and core size is 2；Layer3 convolutional layer packet
Containing 10 cores, convolution kernel size is step-length in 100, layer3 in the layer of pond and core size is 2；Layer4 convolutional layer
Comprising 10 cores, convolution kernel size is step-length in 50, layer4 in the layer of pond and core size is 2；Layer5 convolutional layer
Comprising 10 cores, convolution kernel size is step-length in 48, layer5 in the layer of pond and core size is 2；Layer6 convolutional layer
Comprising 10 cores, convolution kernel size is step-length in 24, layer4 in the layer of pond and core size is 2；Full articulamentum
(layer7) input neuron number is consistent with the output Characteristic Number of layer6, and loss function uses following public affairs in training
Formula:The output Characteristic Number of layer6 is 30, eventually passes through full articulamentum
10 features are exported after calculating.10 features exported using full articulamentum are input to classifier layer (layer8) again as input
In, and input the known result for atrial fibrillation or non-atrial fibrillation and be trained, wherein used training algorithm can be using existing
Any training algorithm having.Training algorithm can be with are as follows: stochastic gradient descent algorithm, Adam algorithm, RMSProp algorithm,
Adagrad algorithm, Adadelta algorithm, Adamax algorithm etc..
In loss function, c indicates loss size, and n indicates that number of samples, x are sample, and y is the corresponding label of sample x, and a is
Pass through the output size for being input with sample x and being gone out with current neural computing.Wherein, α, β are regularization coefficient.
It is preferred that α=0.8-1.2, β=0.8-1.2.Loss function is the function for calculating loss, and loss is the prediction of single sample
The difference of value and true value can make convolutional neural networks increase the punishment to false negative in training using the loss function, from
And under the premise of realizing guarantee accuracy rate, omission factor is reduced, accuracy is improved.
When carrying out rhythm of the heart type identification, pretreated electrocardiosignal to be identified is inputted in neural network, by pre-
10 features first exported after the layer1-7 of trained neural network, 10 features are input to classifier layer (layer8) again
In, layer8 passes through to be carried out calculating output as a result, output result is 0 or 1, wherein 0 represents non-room to electrocardiosignal to be identified
It quivers, 1 represents atrial fibrillation.
Each clathrum of table 1 is as shown in the table:
Effect example
Using 1760 electrocardiogram (ECG) datas for being known as atrial fibrillation and 2232 is non-atrial fibrillation electrocardiosignal as experimental data, is used
Trained network is classified in method mentioned above, the loss function that when classification uses forFor under different α and β value, the results are shown in Table 2 for classification accuracy:
Table 2: the accuracy of the electrocardiogram (ECG) data identification to atrial fibrillation of the CNN under different loss functions
It can be seen that, α=0.8-0.9 in loss function, β=1.1-1.2, α obtain atrial fibrillation standard at 0.8 from upper table
The maximum value of true rate, β obtain the maximum value of atrial fibrillation accuracy rate at 1.2.α obtains the maximum value of non-atrial fibrillation accuracy rate at 1.2,
β obtains the maximum value of non-atrial fibrillation accuracy rate at 1.
Therefore, as a further improvement, the improvement convolutional neural networks of the type of the rhythm of the heart for identification of the application:
First nerves network is obtained by the training method in embodiment, and α=1-1.2 in the loss function, β=
Knowledge is worked as carrying out rhythm of the heart type identification to pretreated electrocardiosignal to be identified in 0.8-1, preferably α=1.2, β=1
Rhythm of the heart type is exported when other rhythm of the heart type is non-atrial fibrillation；
Nervus opticus network is obtained by the training method in embodiment, α=0.8-1, β=1.0- in the loss function
1.2, preferably α=0.8, β=1.2, for believing the electrocardio that the rhythm of the heart type through the first nerves Network Recognition is atrial fibrillation
Rhythm of the heart type identification is carried out number again, and exports rhythm of the heart type.
α=1.2, the first nerves network of β=1 is high to the electrocardiogram recognition accuracy of non-atrial fibrillation, and the electrocardiogram of atrial fibrillation is known
Other accuracy rate is low, therefore demonstrate,proves whether electrocardiogram is non-atrial fibrillation by first nerves network topology.α=0.8, the second mind of β=1.2
High through electrocardiogram recognition accuracy of the network to atrial fibrillation, the electrocardiogram recognition accuracy of non-atrial fibrillation is low, and first nerves network is known
The result not obtained is that the electrocardiogram of atrial fibrillation is identified again by nervus opticus network, can combine two neural networks
Advantage improves the accuracy rate of identification.First nerves network and nervus opticus network identify just to be recognized when electrocardiogram is atrial fibrillation type
It is atrial fibrillation type for electrocardiogram, first nerves network and nervus opticus Network Recognition are that the electrocardiogram that is considered as of non-atrial fibrillation type is
Non- atrial fibrillation type.
It is enlightenment with the above-mentioned desirable embodiment according to the application, through the above description, relevant staff is complete
Full various changes and amendments can be carried out in the range of without departing from this item application technical idea.The technology of this item application
Property range is not limited to the contents of the specification, it is necessary to which the technical scope thereof is determined according to the scope of the claim.
Claims (10)
1. a kind of with the portable electrocardiograph for improving convolutional neural networks recognizer characterized by comprising
Main control unit (1), for being handled electrocardiosignal and exporting processing result；
Electrocardio signal collecting unit (4) carries out operation amplifier for receiving the electrocardiosignal of human body, and to electrocardiosignal, by mould
Quasi- signal is transferred to main control unit (1) after being converted to digital signal, and the electrocardio signal collecting unit (4) includes analog signal tune
Manage module (41), collection of simulant signal module (42) and digital filtering module (43)；
Memory module (3), for receiving electrocardiosignal that main control unit (1) is sent and being transferred to the electrocardiosignal of storage
Main control unit (1)；
Alarm module (5), for issuing alarm signal；
Display module (61), for showing the processing result of electrocardiosignal；
Input module (62), for being instructed to main control unit (1) input control；
Power module (8) is connected with battery (9), and gives main control unit (1), memory module (3), electrocardio signal collecting unit
(4), alarm module (5) and display module (61) power supply；
The main control unit (1) can pre-process electrocardiosignal after receiving electrocardiosignal, and will be pretreated
Electrocardiosignal is input to neural network trained in advance and is identified, exports the corresponding rhythm of the heart type of the electrocardiosignal, wherein
The neural network uses loss function are as follows:Convolutional neural networks, wherein
C expression loss size, n indicate that number of samples, x are sample, and y is the corresponding label of sample x, and a is by being to input simultaneously with sample x
And the output size gone out with current neural computing, α, β are regularization coefficient.
2. according to claim 1 have the portable electrocardiograph for improving convolutional neural networks recognizer, feature
It is, the memory module (3) is TF card；The wireless communication module (7) is bluetooth module, Wifi module, 3G/4G module.
3. according to claim 1 have the portable electrocardiograph for improving convolutional neural networks recognizer, feature
It is, the main control unit (1) is STM32F412RG chip, has FPU module, AD in the piece of the STM32F412RG chip
Module, SPI module, SDIO module, UART module and SRAM space；
The electrocardio signal collecting unit (4) is the ADS1293 chip of Texas Instrument.
4. according to claim 1 have the portable electrocardiograph for improving convolutional neural networks recognizer, feature
It is, further includes ambient light sensor (2), the ambient light sensor (2), can be in environment light for incuding ambient light intensity
Signal is sent to main control unit (1) when intensity is lower than setting value；
The main control unit (1) can close the display module when receiving the signal of ambient light sensor (2) sending
(61)。
5. according to claim 1 have the portable electrocardiograph for improving convolutional neural networks recognizer, feature
It is, the neural network includes 1 input layer, multiple convolutional layers and pond layer, 1 full articulamentum and 1 classifier layer,
α=0.8-0.9, β=1.1-1.2 in the loss function.
6. according to claim 1 have the portable electrocardiograph for improving convolutional neural networks recognizer, feature
It is,
The electrocardiosignal is pre-processed, comprising the following steps:
Whether the sample frequency for judging the electrocardiosignal is predeterminated frequency；
When the sample frequency is not the predeterminated frequency, interpolation method is used to preset the electrocardiosignal resampling to be described
The electrocardiosignal of frequency.
7. according to claim 1 have the portable electrocardiograph for improving convolutional neural networks recognizer, feature
It is, the neural network trained in advance is obtained by following steps:
S31: tranining database is obtained, tranining database is the electrocardiogram (ECG) data for being known as atrial fibrillation or non-atrial fibrillation；
S32: electrocardiogram (ECG) data is pre-processed；
S33: being trained with convolutional neural networks, and convolutional neural networks are connected entirely by several convolutional layers and several pond layers, one
It connects layer and a classifier layer is constituted, input the result of known atrial fibrillation or non-atrial fibrillation when training in classifier layer.
8. according to claim 7 have the portable electrocardiograph for improving convolutional neural networks recognizer, feature
It is,
In S31 step, using the atrial fibrillation electrocardiosignal and at least 10,000 mixed uniformly other types electrocardios of at least 10,000 10s
Signal as training data formed tranining database, using 0 and 1 as atrial fibrillation and the electrocardiosignal of non-atrial fibrillation label；
In S32 step, electrocardiogram (ECG) data is filtered using the fir filter that upper lower limiting frequency is respectively 0.1Hz, 100Hz,
If electrocardiosignal sample frequency is not 500Hz, use closest interpolation method by electrocardiosignal resampling for 500Hz；
In S33 step, convolutional neural networks have 8 layer networks, and 1-6 layer are layer1-layer6 in 8 layer networks, by one
A convolutional layer and a pond layer form, and convolutional layer includes 5 cores in layer1, and convolution kernel size is pond in 224, layer1
Step-length and core size in change layer are 2；Layer2 convolutional layer includes 5 cores, and convolution kernel size is pond in 112, layer2
Step-length and core size in change layer are 2；Layer3 convolutional layer includes 10 cores, and convolution kernel size is pond in 100, layer3
Step-length and core size in change layer are 2；Layer4 convolutional layer includes 10 cores, and convolution kernel size is pond in 50, layer4
Step-length and core size in change layer are 2；Layer5 convolutional layer includes 10 cores, and convolution kernel size is pond in 48, layer5
Step-length and core size in change layer are 2；Layer6 convolutional layer includes 10 cores, and convolution kernel size is pond in 24, layer4
Step-length and core size in change layer are 2；The output Characteristic Number of layer6 is 30, eventually passes through full articulamentum layer7 meter
10 features are exported after calculation.
9. according to claim 1-8 have the cardioscribe for improving convolutional neural networks recognizer
Instrument, which is characterized in that
Trained neural network includes first nerves network and nervus opticus network in advance, and the first nerves network is using loss
Function are as follows:Convolutional neural networks, wherein c indicate loss size, n indicate
Number of samples, x are sample, and y is the corresponding label of sample x, and a is by being input and with current neural network with sample x
Calculated output size, α, β are regularization coefficient；α=1-1.2 in the loss function, β=0.8-1, electrocardiosignal warp
Corresponding rhythm of the heart type is exported after first nerves Network Recognition；
If rhythm of the heart type is non-atrial fibrillation, then it is assumed that the electrocardiosignal is non-atrial fibrillation；
If rhythm of the heart type is atrial fibrillation, which is input to nervus opticus network trained in advance again
It is identified；
Nervus opticus network uses loss function are as follows:Convolutional neural networks,
Wherein, c indicates loss size, and n indicates that number of samples, x are sample, and y is the corresponding label of sample x, and a is by being with sample x
The output size for inputting and being gone out with current neural computing, α, β are regularization coefficient；α in the loss function=
0.8-1, β=1.0-1.2；The corresponding rhythm of the heart type of the electrocardiosignal through nervus opticus Network Recognition is exported, with the second mind
Rhythm of the heart type through Network Recognition is the rhythm of the heart type of the electrocardiosignal.
10. according to claim 9 have the portable electrocardiograph for improving convolutional neural networks recognizer, feature
It is,
First nerves network is using α=1.2 in loss function, β=1；Nervus opticus network is using α=0.8, β in loss function
=1.2.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910257532.6A CN110037685A (en) | 2019-04-01 | 2019-04-01 | With the portable electrocardiograph for improving convolutional neural networks recognizer |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910257532.6A CN110037685A (en) | 2019-04-01 | 2019-04-01 | With the portable electrocardiograph for improving convolutional neural networks recognizer |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110037685A true CN110037685A (en) | 2019-07-23 |
Family
ID=67275780
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910257532.6A Pending CN110037685A (en) | 2019-04-01 | 2019-04-01 | With the portable electrocardiograph for improving convolutional neural networks recognizer |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110037685A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112617850A (en) * | 2021-01-04 | 2021-04-09 | 苏州大学 | Premature beat and heart beat detection method for electrocardiosignals |
| CN113180690A (en) * | 2021-04-21 | 2021-07-30 | 上海数创医疗科技有限公司 | Electrocardio diagnosis model and electrocardio detection device |
| CN113749668A (en) * | 2021-08-23 | 2021-12-07 | 华中科技大学 | Wearable electrocardiogram real-time diagnosis system based on deep neural network |
| CN117257324A (en) * | 2023-11-22 | 2023-12-22 | 齐鲁工业大学(山东省科学院) | Atrial fibrillation detection method based on convolutional neural network and ECG (electro-magnetic resonance) signals |
| CN117648570A (en) * | 2024-01-30 | 2024-03-05 | 吉林大学 | Electrocardiogram monitoring and data analysis system based on artificial intelligence |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101887700A (en) * | 2009-05-15 | 2010-11-17 | 宏碁股份有限公司 | Computer system, display device and backlight control method |
| US20170032221A1 (en) * | 2015-07-29 | 2017-02-02 | Htc Corporation | Method, electronic apparatus, and computer readable medium of constructing classifier for disease detection |
| CN108113666A (en) * | 2017-12-19 | 2018-06-05 | 中国科学院深圳先进技术研究院 | Recognition methods, device and the equipment of atrial fibrillation signal |
| CN108403110A (en) * | 2018-04-19 | 2018-08-17 | 成都天奥电子股份有限公司 | A kind of real-time atrial fibrillation automatic detection system and its detection method based on wearable device |
-
2019
- 2019-04-01 CN CN201910257532.6A patent/CN110037685A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101887700A (en) * | 2009-05-15 | 2010-11-17 | 宏碁股份有限公司 | Computer system, display device and backlight control method |
| US20170032221A1 (en) * | 2015-07-29 | 2017-02-02 | Htc Corporation | Method, electronic apparatus, and computer readable medium of constructing classifier for disease detection |
| CN108113666A (en) * | 2017-12-19 | 2018-06-05 | 中国科学院深圳先进技术研究院 | Recognition methods, device and the equipment of atrial fibrillation signal |
| CN108403110A (en) * | 2018-04-19 | 2018-08-17 | 成都天奥电子股份有限公司 | A kind of real-time atrial fibrillation automatic detection system and its detection method based on wearable device |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112617850A (en) * | 2021-01-04 | 2021-04-09 | 苏州大学 | Premature beat and heart beat detection method for electrocardiosignals |
| CN112617850B (en) * | 2021-01-04 | 2022-08-30 | 苏州大学 | Premature beat and heart beat detection system for electrocardiosignals |
| CN113180690A (en) * | 2021-04-21 | 2021-07-30 | 上海数创医疗科技有限公司 | Electrocardio diagnosis model and electrocardio detection device |
| CN113749668A (en) * | 2021-08-23 | 2021-12-07 | 华中科技大学 | Wearable electrocardiogram real-time diagnosis system based on deep neural network |
| CN113749668B (en) * | 2021-08-23 | 2022-08-09 | 华中科技大学 | Wearable electrocardiogram real-time diagnosis system based on deep neural network |
| CN117257324A (en) * | 2023-11-22 | 2023-12-22 | 齐鲁工业大学(山东省科学院) | Atrial fibrillation detection method based on convolutional neural network and ECG (electro-magnetic resonance) signals |
| CN117257324B (en) * | 2023-11-22 | 2024-01-30 | 齐鲁工业大学(山东省科学院) | Atrial fibrillation detection method based on convolutional neural network and ECG (electro-magnetic resonance) signals |
| CN117648570A (en) * | 2024-01-30 | 2024-03-05 | 吉林大学 | Electrocardiogram monitoring and data analysis system based on artificial intelligence |
| CN117648570B (en) * | 2024-01-30 | 2024-04-12 | 吉林大学 | Electrocardiogram monitoring and data analysis system based on artificial intelligence |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110037685A (en) | With the portable electrocardiograph for improving convolutional neural networks recognizer | |
| Liu et al. | Real-time multilead convolutional neural network for myocardial infarction detection | |
| Xia et al. | A novel wearable electrocardiogram classification system using convolutional neural networks and active learning | |
| CN103038772B (en) | Predict the system and device of the viability of patient | |
| US20170112401A1 (en) | Automatic method to delineate or categorize an electrocardiogram | |
| Li et al. | A low-complexity data-adaptive approach for premature ventricular contraction recognition | |
| CN106725428A (en) | A kind of electrocardiosignal sorting technique and device | |
| CN108836302A (en) | Electrocardiogram intelligent analysis method and system based on deep neural network | |
| Li et al. | Automated heartbeat classification exploiting convolutional neural network with channel-wise attention | |
| CN111557660A (en) | Arrhythmia identification method under sub-population deep learning framework | |
| Li et al. | Arrhythmia classification using biased dropout and morphology-rhythm feature with incremental broad learning | |
| Liu et al. | ECG-based heart arrhythmia diagnosis through attentional convolutional neural networks | |
| CN105943022B (en) | A kind of cardioelectric monitor system that there are three leads to reconstruct 12 lead function | |
| CN205041399U (en) | Intelligent electrocardiogram monitoring system | |
| Jin et al. | Normal versus abnormal ECG classification by the aid of deep learning | |
| Falaschetti et al. | ECG-based arrhythmia classification using recurrent neural networks in embedded systems | |
| CN106037714A (en) | Heart beat monitoring device | |
| CN204581273U (en) | A kind of hand held ECG Signal Sampling System for identity validation | |
| Hu et al. | Unsupervised feature learning for heart sounds classification using autoencoder | |
| WO2013186634A2 (en) | Predicting acute cardiopulmonary events and survivability of a patient | |
| CN109770851A (en) | A kind of health of heart condition monitoring system and method based on Bluetooth wireless communication | |
| Hu et al. | Deep multi-instance networks for bundle branch block detection from multi-lead ECG | |
| Wu et al. | Research on personal identity verification based on convolutional neural network | |
| Chudacek et al. | Comparison of seven approaches for holter ECG clustering and classification | |
| CN102764118A (en) | Internet of things-based electrocardiogram monitoring system |
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: 20190723 |
|
| RJ01 | Rejection of invention patent application after publication |