CN116712083A - QRS complex wave detection method based on U-Net network - Google Patents
QRS complex wave detection method based on U-Net network Download PDFInfo
- Publication number
- CN116712083A CN116712083A CN202310417011.9A CN202310417011A CN116712083A CN 116712083 A CN116712083 A CN 116712083A CN 202310417011 A CN202310417011 A CN 202310417011A CN 116712083 A CN116712083 A CN 116712083A
- Authority
- CN
- China
- Prior art keywords
- qrs complex
- layer
- image
- convolution
- network
- 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
- 238000000718 qrs complex Methods 0.000 title claims abstract description 177
- 238000001514 detection method Methods 0.000 title claims abstract description 134
- 238000012549 training Methods 0.000 claims abstract description 56
- 238000000034 method Methods 0.000 claims description 20
- 230000004913 activation Effects 0.000 claims description 16
- 238000010606 normalization Methods 0.000 claims description 13
- 238000012805 post-processing Methods 0.000 claims description 11
- 238000004590 computer program Methods 0.000 claims description 10
- 238000004891 communication Methods 0.000 claims description 6
- 238000004422 calculation algorithm Methods 0.000 claims description 5
- 238000005070 sampling Methods 0.000 claims description 5
- 238000009499 grossing Methods 0.000 claims description 2
- 230000006870 function Effects 0.000 description 26
- 238000010586 diagram Methods 0.000 description 23
- 230000008569 process Effects 0.000 description 6
- 238000012545 processing Methods 0.000 description 5
- 238000007781 pre-processing Methods 0.000 description 4
- 206010003119 arrhythmia Diseases 0.000 description 3
- 230000006793 arrhythmia Effects 0.000 description 3
- 230000015556 catabolic process Effects 0.000 description 3
- 238000006731 degradation reaction Methods 0.000 description 3
- 230000008034 disappearance Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 239000000203 mixture Substances 0.000 description 3
- 238000003860 storage Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 2
- 208000000418 Premature Cardiac Complexes Diseases 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 230000005284 excitation Effects 0.000 description 2
- 230000036039 immunity Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 230000002861 ventricular Effects 0.000 description 2
- 102100032202 Cornulin Human genes 0.000 description 1
- 206010015856 Extrasystoles Diseases 0.000 description 1
- 101000920981 Homo sapiens Cornulin Proteins 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000001746 atrial effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 238000006073 displacement reaction Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 238000002565 electrocardiography Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000003183 myoelectrical effect Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 230000034225 regulation of ventricular cardiomyocyte membrane depolarization Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000011426 transformation method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
-
- 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
- A61B5/366—Detecting abnormal QRS complex, e.g. widening
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Health & Medical Sciences (AREA)
- Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Cardiology (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Veterinary Medicine (AREA)
- Heart & Thoracic Surgery (AREA)
- Medical Informatics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Public Health (AREA)
- Pathology (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Physiology (AREA)
- Psychiatry (AREA)
- Signal Processing (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The application relates to a QRS complex detection method based on a U-Net network, which comprises the steps of generating a noise electrocardiosignal by adding clean electrocardiosignals and noise to obtain pre-training data; taking the noise electrocardiosignals and the corresponding clean electrocardiosignals as the input and the output of a pre-constructed QRS complex detection network, and pre-training the QRS complex detection network; the QRS complex detection network is constructed based on a U-Net network; the mean square error is used as a loss function, and after training for preset times, network parameters are output; collecting an original electrocardiosignal and a corresponding QRS complex label, and initializing a QRS complex detection network by using network parameters; inputting the original electrocardiosignal into an initialized QRS complex detection network, continuously reducing the network output result and the Dice loss function of the corresponding QRS complex label, and obtaining a trained QRS complex detection network after training for preset times; and acquiring an electrocardiosignal image to be detected, inputting the electrocardiosignal image into a trained QRS complex detection network, and acquiring a QRS complex label image.
Description
Technical Field
The application relates to the technical field of electrocardiogram detection, in particular to a QRS complex detection method, device and equipment based on a U-Net network.
Background
Typical electrocardiogram heartbeats are mainly composed of P-wave, QRS complex and T-wave; wherein the QRS waveform indicates ventricular depolarization, which provides rich information about ventricular excitation and conduction. Since QRS waveforms are the most prominent features in Electrocardiogram (ECG) signals, they are often used as a basis for determining heart rate, and are also of great importance for the diagnosis of ectopic beats and for the analysis of HRV.
The traditional QRS complex detection method mainly comprises a preprocessing stage and a decision stage, wherein the preprocessing is aimed at enhancing the QRS complex component and attenuating other waves and noise, a detection rule is designed to identify the QRS complex group in the decision stage, and a post-processing scheme is additionally designed for part of research to optimize the detection result. The traditional QRS wave detection method mainly comprises wavelet transformation, empirical mode decomposition, hilbert transformation and machine learning methods. In summary, traditional detection methods rely largely on hand-made manual features and parameters, which makes these methods lacking generalization ability to unknown data, difficult to generalize to data that is out of distribution, especially on low quality electrocardiography between patients and arrhythmias.
In recent years, there have also been studies on applying a deep learning method to QRS complex detection. Runnen He first applied a preprocessor with mean filtering and discrete wavelet transform to remove different types of noise, etc.; then, a new algorithm for automatically detecting QRS complex wave based on U-Net and two-way long-short-term memory is designed; the accuracy using the data sets from the MIT-BIH arrhythmia database and CPSC2019 were 98.29% and 78.73%, respectively. Cai et al propose two deep learning models based on multiple expansion convolution blocks; one of the models (CNN) is mainly composed of convolutions and a Squeeze and Excitation Network (SENET), the other model (CRNN) contains a mixed convolutions and recurrent neural network; the set of these two models has obtained the first name in the chinese physiological signal challenge (2019). Mohamed et al propose a low-complexity stacked self-encoder deep neural network model, the model does not need preprocessing, has a simple structure, only comprises two hidden layers, has high algorithm running speed, and is more suitable for the realization of an embedded system; the experimental result adopts a plurality of reference data sets MIT/BIH, INCART DB and the like, and proves the generalization and high performance of the algorithm.
However, the existing QRS complex detection method has poor QRS complex detection effect on abnormal electrocardiograms or electrocardiograms greatly influenced by noise, and complicated denoising operation is required to be performed on electrocardiosignals before detection.
Disclosure of Invention
Therefore, the application aims to solve the technical problems that the detection effect is poor and complex denoising operation is needed in the prior art when QRS wave detection is performed.
In order to solve the technical problems, the application provides a QRS complex detection method based on a U-Net network, which comprises the following steps:
generating a noise electrocardiosignal by adding the clean electrocardiosignal and noise to obtain pre-training data;
taking the noise electrocardiosignals and the corresponding clean electrocardiosignals as the input and the output of a pre-constructed QRS complex detection network, and pre-training the QRS complex detection network; the QRS complex detection network is constructed based on a U-Net network;
after training for preset times, outputting network parameters of the QRS complex detection network by using the mean square error as a loss function;
collecting an original electrocardiosignal and a corresponding QRS complex label, and initializing a QRS complex detection network by using the network parameters;
inputting the original electrocardiosignal into an initialized QRS complex detection network, continuously reducing the network output result and the Dice loss function of the corresponding QRS complex label, and obtaining a trained QRS complex detection network after training for preset times;
and acquiring an electrocardiosignal image to be detected, inputting the electrocardiosignal image to a trained QRS complex detection network, and acquiring a QRS complex label image of the electrocardiosignal image to be detected.
In one embodiment of the present application, the acquiring the electrocardiograph signal image to be detected, inputting the electrocardiograph signal image to a trained QRS complex detection network, and acquiring a QRS complex label image of the electrocardiograph signal image to be detected includes:
acquiring an electrocardiosignal image to be detected, inputting the electrocardiosignal image to a multi-scale convolution sum feature multiplexing module of a pre-trained QRS complex detection network, and carrying out three parallel convolution branches with different convolution kernel sizes;
an electrocardiograph signal image to be detected sequentially passes through a1 multiplied by 1 convolution unit, a first convolution unit and a second convolution unit which are connected in series along the positive propagation direction in each path of convolution branch; the output images of the three convolution units are spliced to be used as the output images of the convolution branches; splicing the output images of the three paths of parallel convolution branches to serve as an output characteristic image of the multi-scale convolution and characteristic taking module;
after downsampling the output characteristic image, inputting the code path of the QRS complex detection network for downsampling for a plurality of times, and outputting a code image;
after up-sampling the coded image, inputting the coded image into a decoding path of a QRS complex detection network, and outputting a decoded image after up-sampling the coded image for the same times;
inputting the decoded image into a1 multiplied by 1 convolution, and then obtaining a QRS complex wave label graph of the electrocardiosignal graph to be detected through an activation function layer.
In one embodiment of the present application, the first convolution unit and the second convolution unit each include a feature disturbance layer Disout, a feature discard layer Dropout, a 1D convolution layer of a preset convolution kernel size, a batch normalization layer BN, an activation function layer ReLU, and an effective channel attention layer EAC, which are sequentially connected in series along a positive propagation direction.
In one embodiment of the application, the downsampling is accomplished using a large step depth convolution module; the large step depth convolution module includes: and outputting the input characteristic image after passing through a characteristic disturbance layer Disout, a characteristic discarding layer Dropout, a 1D depth convolution layer with a preset convolution kernel size, a batch normalization layer BN, an activation function layer ReLU and an effective channel attention layer EAC which are sequentially connected in series along the forward propagation direction, and taking the output characteristic image as the input characteristic image of the next coding layer.
In one embodiment of the present application, the coding path includes four coding layers sequentially connected in series along the forward propagation direction; each coding layer convolves the input characteristic image by using a GConv-Block module and then outputs the coding characteristic image of the coding layer;
the GConv-Block module comprises a first Group convolution unit, a second Group convolution unit and a third Group convolution unit, wherein the first Group convolution unit, the second Group convolution unit and the third Group convolution unit are sequentially connected in series along the positive propagation direction;
performing pixel point addition on the output image of the first grouping convolution unit and the output image of the second grouping convolution unit, and taking the pixel point addition as an input of a third convolution unit;
and carrying out pixel point addition on the output image of the 1 multiplied by 1 convolution and the output image of the third convolution unit to obtain the output of the GConv-Block module.
In one embodiment of the present application, the first packet convolution unit and the second packet convolution unit each include a feature disturbance layer distut, a feature discard layer Dropout, a 1D packet convolution layer of a preset convolution kernel size, a batch normalization layer BN, an activation function layer ReLU, and an effective channel attention layer EAC, which are sequentially connected in series along a positive propagation direction;
the third convolution unit comprises a characteristic disturbance layer Disout, a characteristic discarding layer Dropout, a 1D convolution layer with a preset convolution kernel size, a batch normalization layer BN, an activation function layer ReLU and an effective channel attention layer EAC which are sequentially connected in series along the positive propagation direction.
In one embodiment of the present application, the decoding path includes four decoding layers sequentially connected in series along the forward propagation direction;
each decoding layer is in jump connection with a corresponding coding layer, an output image of the corresponding coding layer and an output image of an upper decoding layer of the upper layer are obtained, and the output images are spliced in series to be used as input characteristic images of the decoding layers of the layer;
the upsampling uses nearest neighbor interpolation.
In one embodiment of the present application, after the QRS complex label map of the electrocardiograph signal map to be detected is obtained, post-processing is performed to obtain the position of the QRS complex, where the post-processing includes:
and smoothing the QRS complex label graph by using a moving average algorithm to obtain smoothed peak point coordinates which are the QRS complex position of the electrocardiosignal graph to be detected.
The embodiment of the application also provides a QRS complex detection device based on U-Net, which comprises the following components:
the pre-training module is used for generating a noise electrocardiosignal by adding the clean electrocardiosignal and the noise to obtain pre-training data; taking the noise electrocardiosignals and the corresponding clean electrocardiosignals as the input and the output of a QRS complex detection network, and pre-training the QRS complex detection network; after training for preset times, outputting network parameters of the QRS complex detection network by using the mean square error as a loss function;
the formal training module is used for collecting an original electrocardiosignal and a corresponding QRS complex label and initializing a QRS complex detection network by utilizing the network parameters; inputting the original electrocardiosignal into an initialized QRS complex detection network, continuously reducing the network output result and the Dice loss function of the corresponding QRS complex label, and obtaining a trained QRS complex detection network after training for preset times;
the detection module is used for acquiring an electrocardiosignal image to be detected, inputting the electrocardiosignal image to a trained QRS complex detection network and acquiring a QRS complex label image of the electrocardiosignal image to be detected.
The embodiment of the application also provides a QRS complex detection device based on U-Net, which comprises:
the electrocardiosignal acquisition device is used for acquiring an electrocardiosignal image to be detected;
the upper computer is in communication connection with the electrocardiosignal acquisition device and is used for realizing the steps of the QRS complex wave detection method based on U-Net when executing a computer program;
and the display device is in communication connection with the upper computer and is used for displaying a QRS complex wave label graph of the electrocardiosignal graph to be detected.
Compared with the prior art, the technical scheme of the application has the following advantages:
when training the QRS complex detection network, the QRS complex detection method based on the U-Net network firstly uses noise electrocardiosignals to pretrain, acquires pretrained network parameters, and initializes the QRS complex detection network; and acquiring an original electrocardiosignal, and performing formal training on the initialized QRS complex detection network. According to the application, the QRS complex detection network is initialized by utilizing the network parameters obtained by pre-training, so that random initialization parameters are avoided in direct training, pre-denoising is realized, the anti-noise capability is good, complex denoising operation on the electrocardiosignal is not required before detection, and the robustness of the QRS complex detection network is effectively improved;
the application utilizes the multi-scale convolution and feature multiplexing module MCRF to enable the input image to pass through three parallel convolution branches with different convolution kernel sizes, extract the effective information of different scales of the input feature image as much as possible, and repeatedly use the shallow features through jump connection, thereby effectively relieving the problem of gradient disappearance;
according to the application, on the encoding path and the decoding path, the grouping convolution unit is utilized to reduce the parameter quantity and simultaneously compress the expression of the redundancy characteristic; the conventional convolution unit is utilized to effectively promote the connection between the feature graphs and further extract the image features; and jump connection is arranged at symmetrical positions, deep features and shallow features are spliced, and gradient degradation is relieved.
Drawings
In order that the application may be more readily understood, a more particular description of the application will be rendered by reference to specific embodiments thereof that are illustrated in the appended drawings, in which
Fig. 1 is a flowchart of a QRS complex detection method based on a U-Net network provided by the present application;
fig. 2 is a schematic diagram of a binary label according to an embodiment of the present application, where (a) of fig. 2 is a schematic diagram of an original signal, and (b) of fig. 2 is a QRS complex label diagram;
fig. 3 is a block diagram of a QRS complex detection network U according to an embodiment of the present application QRS -network architecture diagram of Net;
FIG. 4 is a schematic diagram of the structural composition of a multi-scale convolution and feature multiplexing module MCRF according to an embodiment of the present application;
FIG. 5 is a schematic diagram of the structural composition of a GConv-Block module according to an embodiment of the present application;
FIG. 6 is a schematic diagram of the structural composition of a large step depth convolution module provided by an embodiment of the present application;
fig. 7 is a schematic diagram of a post-processing procedure provided by the embodiment of the present application, where (a) of fig. 7 shows a QRS complex label graph output by the QRS complex detection network, (b) of fig. 7 shows a post-processing result graph, and (c) of fig. 7 shows an original electrocardiograph signal graph.
Fig. 8 is a flowchart of a specific method of QRS complex detection method based on a U-Net network according to an embodiment of the present application.
Detailed Description
The present application will be further described with reference to the accompanying drawings and specific examples, which are not intended to be limiting, so that those skilled in the art will better understand the application and practice it.
Example 1:
referring to fig. 1, the QRS complex detection method based on the U-Net network of the present application specifically includes the steps of:
s101, adding clean electrocardiosignals and noise to generate noise electrocardiosignals, and obtaining pre-training data;
s102, taking the noise electrocardiosignals and the corresponding clean electrocardiosignals as a pre-constructed QRS complex detection network U QRS -an input and an output of Net, pre-training the QRS complex detection network; the QRS complex detection network is constructed based on a U-Net network;
s103, utilizing the mean square error as a loss function, and outputting network parameters of the QRS complex detection network after training for preset times;
s104, acquiring an original electrocardiosignal and a corresponding QRS complex label, and initializing a QRS complex detection network by using the network parameters;
s105, inputting the original electrocardiosignal into an initialized QRS complex detection network, continuously reducing the network output result and the Dice loss function of the corresponding QRS complex label, and obtaining a trained QRS complex detection network after training for preset times;
s106, acquiring an electrocardiosignal image to be detected, inputting the electrocardiosignal image to a trained QRS complex detection network, and acquiring a QRS complex label image of the electrocardiosignal image to be detected.
In particular, in the task of QRS wave detection, how to achieve accurate detection in noisy environments is also one of the key issues, so the present application uses ECG signal recovery tasks to develop pre-training of the model. The ECG data after noise pollution and the clean ECG data are respectively used as the input and the output of the model, so that the model directly obtains more universal characteristics of the ECG signals in the process of learning how to recover the complete ECG signals, and the quality of learning characterization is improved, thereby improving the effect of QRS complex detection.
Specifically, in step S101, the acquiring of the pre-training data includes: firstly, picking and intercepting single-lead clean electrocardiosignals with the length of 10s from a CPSC2018 (the China Physiological Signal Challenge 2018) data set, wherein the single-lead clean electrocardiosignals comprise 9 different types of diseases such as normal electrocardiogram, ventricular premature beat, new atrial premature beat, ST depression and the like; the noise signal is then randomly truncated from the NSTDB (the MIT-BIH Noise Stress Test Database) dataset, which contains three different noise signals: myoelectric noise (MA), electrode displacement motion (EM), and baseline drift Baseline Wander (BW), each noise source consisting of two channels; each noise randomly intercepts 10s segments and resamples to 500Hz; finally, adding the noise and the clean electrocardiosignals to generate a noise electrocardiosignal, wherein the generation mode is expressed as follows:
Noise-convolved ECG＝a×MA+b×EM+c×BW+Ground-truth ECG；
wherein a, b, c e (0, 1) and a+b+c=1;
specifically, in step S104, acquiring the original electrocardiographic signal and the corresponding QRS complex label includes:
three databases of CPSC2019 (the China Physiological Signal Challenge 2019), MITDB (MIT-BIH arrhythmia database) and INCART DB (Incardiology saint petersburg database) are selected as training sets and test sets for QRS complex detection, 10s single-lead electrocardiosignals are randomly intercepted, and all the electrocardiosignals are sampled to 500Hz without any preprocessing.
The application designs a binary label, and converts the task of QRS complex positioning into a target detection task; referring to fig. 2, a binary label schematic diagram provided by the embodiment of the present application is shown, fig. 2 (a) is a schematic diagram of an original signal, fig. 2 (b) is a QRS complex label diagram, labels are based on QRS position points given by authorities, 37 sample points (75 ms) are respectively amplified on the left and right sides, and the width of a1 label corresponding to each QRS position is 37+37+1=75 sample points (150 ms).
Specifically, in step S106, the present application reforms based on the structure of the Unet, and the reformed model is named U QRS -net; referring to FIG. 3, in the present embodiment, U QRS The net has four downsampling and corresponding upsampling and uses a jump connection to perform feature stitching on the corresponding encoding and decoding modules. The input and output sizes of the model remain the same, and the lowest resolution of the feature map in the Unet is 1/40 of the highest resolution (original signal). And as the depth of the network increases, the convolution kernel size gradually decreases, and specific parameters are shown in table 1:
table 1: u (U) QRS Network parameters of the net
| Layer | Output | Kernel Size | Padding | Stride |
| MCFR | (27×1×5000) | 1×15,1×35and 1×65 | same | 1 |
| Downsampling | (27×1×1000) | 1×35 | same | 5 |
| GConv-block | (24×1×1000) | 1×25 | same | 1 |
| Downsampling | (24×1×500) | 1×25 | same | 2 |
| GConv-block | (36×1×500) | 1×15 | same | 1 |
| Downsampling | (36×1×250) | 1×15 | same | 2 |
| GConv-block | (48×1×250) | 1×9 | same | 1 |
| Downsampling | (48×1×125) | 1×9 | same | 2 |
| GConv-block | (48×1×125) | 1×5 | same | 1 |
| Upsampling | (48×1×250) | 1×2 | - | - |
| GConv-block | (36×1×250) | 1×9 | same | 1 |
| Upsampling | (36×1×500) | 1×2 | - | - |
| GConv-block | (24×1×500) | 1×15 | same | 1 |
| Upsampling | (24×1×1000) | 1×2 | - | - |
| GConv-block | (12×1×1000) | 1×25 | same | 1 |
| Upsampling | (12×1×5000) | 1×5 | - | - |
| GConv-block | (12×1×5000) | 1×35 | same | 1 |
| Convolution | (1×1×5000) | 1×1 | same | 1 |
Example 2:
the QRS complex of different concentric electric signals has larger change, the interval of the QRS complex of some signals is short, and the amplitude change (steep edge) is large; the key to improving the model performance is to pay attention to the diversity and complexity of the QRS complex in morphology. Therefore, the application designs a Multi-scale convolution and characteristic multiplexing module (Multi-scale convolution and feature reuse module, MCFR) in a targeted manner; the MCFR module has three branches, and three convolution kernels (1×15,1×35and1×65) with different sizes are used respectively, so that the size of the feature map is kept consistent through padding.
Based on the above embodiment, specifically, after inputting an electrocardiographic signal diagram to be detected into a pre-trained QRS complex detection network, firstly, processing the electrocardiographic signal diagram by using a multi-scale convolution and feature fusion module MCFR; referring to fig. 4, an electrocardiograph signal diagram to be detected passes through three parallel convolution branches with different convolution kernel sizes; an electrocardiograph signal image to be detected sequentially passes through a1 multiplied by 1 convolution unit, a first convolution unit and a second convolution unit which are connected in series along the positive propagation direction in each path of convolution branch; the output images of the three convolution units are spliced to be used as the output images of the convolution branches; splicing the output images of the three paths of parallel convolution branches to serve as an output characteristic image of the multi-scale convolution and characteristic taking module; after downsampling the output characteristic image, inputting the code path of the QRS complex detection network for downsampling for a plurality of times, and outputting a code image; the coded image is up-sampled and then input into a QRS complex detection network U QRS -a Net decoding path for outputting a decoded image after the same number of upsampling; inputting the decoded image into 1 multiplied by 1 convolution, and then obtaining a QRS complex label graph of an electrocardiosignal graph to be detected through an activation function layer。
The first convolution unit and the second convolution unit comprise a characteristic disturbance layer Disout, a characteristic discarding layer Dropout, a 1D convolution layer with a preset convolution kernel size, a batch normalization layer BN, an activation function layer ReLU and an effective channel attention layer EAC which are sequentially connected in series along the positive propagation direction.
By utilizing the multi-scale convolution and feature multiplexing module MCRF provided by the application, the input image passes through three parallel convolution branches with different convolution kernel sizes, so that effective information of different scales of the input feature image is extracted as much as possible, and shallow features are reused through jump connection, thereby effectively relieving the problem of gradient disappearance.
Example 3:
based on the above embodiment, the present embodiment uses a specially (uniformly) designed convolution module GConv-Block in the encoding path and decoding path to replace the original continuous conventional convolution in U-net.
Specifically, the QRS complex detection network U provided by the embodiment of the present application QRS -Net, the coding path comprising four coding layers in series in sequence along the forward propagation direction; each coding layer convolves the input characteristic image by using a GConv-Block module and then outputs the coding characteristic image of the coding layer; the decoding path comprises four decoding layers which are sequentially connected in series along the forward propagation direction; each decoding layer is in jump connection with the corresponding coding layer, and an output image of the corresponding coding layer and an output image of an upper decoding layer of the upper layer are obtained and spliced in series to be used as an input characteristic image of the decoding layer of the layer.
Referring to fig. 5, the GConv-Block module includes a1×1 convolution, a first packet convolution unit with group=6, a second packet convolution unit with group=3, and a third convolution unit, which are sequentially connected in series in the forward propagation direction. The input feature map is first subjected to a1 x 1 convolution transform of the channel number, and the channel number is also kept consistent in the subsequent convolution layers. Performing pixel point addition on the output image of the first grouping convolution unit and the output image of the second grouping convolution unit, and taking the pixel point addition as an input of a third convolution unit; and carrying out pixel point addition on the output image of the 1 multiplied by 1 convolution and the output image of the third convolution unit to obtain the output of the GConv-Block module.
The first grouping convolution unit and the second grouping convolution unit comprise a characteristic disturbance layer Disout, a characteristic discarding layer Dropout, a 1D grouping convolution layer with a preset convolution kernel size, a batch normalization layer BN, an activation function layer ReLU and an effective channel attention layer EAC which are sequentially connected in series along the positive propagation direction; the third convolution unit comprises a characteristic disturbance layer Disout, a characteristic discarding layer Dropout, a 1D convolution layer with a preset convolution kernel size, a batch normalization layer BN, an activation function layer ReLU and an effective channel attention layer EAC which are sequentially connected in series along the positive propagation direction.
In the embodiment of the present application, p=0.1 in Dropout, dist_prob=0.1 in dist, block_size=current convolution layer size, and alpha=30 are set.
The GConv-Block module provided by the embodiment of the application utilizes the grouping convolution to reduce the parameter quantity and simultaneously compress the expression of redundant features, and the third convolution unit is a conventional convolution for improving the connection between effective feature graphs and further extracting the features. In addition, the application is provided with two jump connections at symmetrical positions, and adds deep layer features and shallow layer features, thereby relieving gradient degradation generated in the training process. According to the embodiment of the application, the Disout layer and the Dropout layer are arranged in front of each convolution layer, so that overfitting in the training process can be more effectively inhibited; both the Dropout layer and the distut layer are effective methods for improving model robustness. In the forward propagation process, the Dropout layer realizes the discarding of the feature map with a certain probability, and the Disout layer realizes the disturbance of the feature map with a certain probability. Discarding the feature map is beneficial to improving the universality of the features of the model in the training process, so as to achieve the purpose of regularization; the disturbance of the characteristic diagram is similar to Gaussian noise randomly superimposed in the electrocardiosignal to a certain extent, and the noise immunity of the model is improved.
Example 4:
based on the above embodiment, in this embodiment, the encoding path implements downsampling with a large-step depth convolution module; the deep convolution can be regarded as a special packet convolution with a number of packets equal to the number of channels, so that no characteristic interaction occurs between the channels in the deep convolution. According to the embodiment of the application, the large-step depth convolution module is used for replacing downsampling in the original U-net, so that the loss of high-frequency information is reduced, and the resolving capability of the model is improved. In addition, in the up-sampling of the decoding end, the nearest neighbor interpolation with small calculation amount is continuously used.
Referring to fig. 6, the large-step depth convolution module includes a feature disturbance layer Disout, a feature discarding layer Dropout, a 1D depth convolution layer with a preset convolution kernel size, a batch normalization layer BN, an activation function layer ReLU and an effective channel attention layer EAC, which are sequentially connected in series along the forward propagation direction, and outputs the input feature image as an input feature image of a next coding layer.
Specifically, in the embodiment of the present application, the conventional convolution layer, the packet convolution layer and the depth convolution layer provided by the present application are both followed by the normalization layer and the Relu activation function, and both use ECA (Efficient Channel Attention) to implement channel enhancement.
Specifically, in order to reduce noise in model output, the predicted position of the QRS complex is more accurate, based on the above embodiment, after acquiring the QRS complex label map of the electrocardiograph signal map to be detected, post-processing is performed on the QRS complex label map, and the position of the QRS complex is acquired; the method specifically comprises the following steps: noise and smooth prediction output are suppressed using a three-level moving average. Peak detection is performed by scipy. In the find_peaks of python, the position of the QRS complex can be obtained. Referring to FIG. 7, a schematic diagram of the post-processing procedure is shown, and FIG. 7 (a) shows a QRS complex detection network U QRS -QRS complex label map of Net output, fig. 7 (b) representing post-processing result map, fig. 7 (c) representing original electrocardiographic signal map; a parameter is set in the moving average line, window_width=30 (corresponding to 60 ms). Parameters were set in the find peak function, height=0.4, distance=100 (corresponding to 200 ms).
Specifically, based on the above embodiment, referring to fig. 8, in this embodiment, a QRS complex detection method based on a U-Net network includes:
s201: preparation of Pre-training data, use U QRS The net network removes the last Sigmoid layer, and takes the noisy electrocardiosignals and the clean electrocardiosignals as network input and output training networks respectively. Training was performed 50 times using the mean square error as a loss function. The training method used Adam with ir=0.01, beta1=0.9, beta2=0.999, epsilon=le-08, clipvalue=0.5.
S202: the method comprises the steps of preparing formal training data, initializing by using the weight of a pre-trained network, and taking 10s original electrocardiosignals as input and the corresponding QRS complex position as output. Using Dice as a loss function, training was performed 250 times. The training method used Adam with ir=0.01, beta1=0.9, beta2=0.999, epsilon=le-08, clipvalue=0.5.
S203: and testing the electrocardiosignal graph to be detected by using the network weight parameters after formal training, outputting a result by a network, and obtaining the final QRS complex position after post-processing.
Based on the above embodiment, the embodiment of the present application further provides a QRS complex detection device based on a U-Net network, where the device specifically includes:
the pre-training module 100 adds the clean electrocardiosignals and noise to generate noise electrocardiosignals, so as to obtain pre-training data; taking the noise electrocardiosignals and the corresponding clean electrocardiosignals as the input and the output of a pre-constructed QRS complex detection network, and pre-training the QRS complex detection network; the QRS complex detection network is constructed based on a U-Net network; after training for preset times, outputting network parameters of the QRS complex detection network by using the mean square error as a loss function;
the formal training module 200 collects the original electrocardiosignal and the corresponding QRS complex label, and initializes the QRS complex detection network U by using the network parameters QRS -Net; inputting original electrocardiosignal into initialized QRS complex wave detection network U QRS Continuously reducing the Dice loss function of the network output result and the corresponding QRS complex label in the Net, and obtaining a trained QRS complex detection network U after training for preset times QRS -Net；
Detection module 300, acquires a signal to be detectedDetecting electrocardiosignal graph, inputting trained QRS complex detection network U QRS -acquiring a QRS complex label map of an electrocardiograph signal map to be detected in Net.
The QRS complex detection device based on the U-Net network of the present embodiment is configured to implement the foregoing QRS complex detection method based on the U-Net network, so that the specific embodiment in the QRS complex detection device based on the U-Net network can be seen as the example part of the foregoing QRS complex detection method based on the U-Net network, for example, the pre-training module 100 is configured to implement steps S101, S102 and S103 in the foregoing QRS complex detection method based on the U-Net network; the formal training module 200 is configured to implement steps S104 and S105 in the QRS complex detection method based on the U-Net network; the detection module 300 is configured to implement step S106 in the above-mentioned QRS complex detection method based on the U-Net network, so the specific implementation manner thereof may refer to the description of the corresponding embodiments of each part, which is not repeated herein.
Based on the above embodiment, the embodiment of the present application further provides a QRS complex detection device based on a U-Net network, where the device specifically includes:
the electrocardiosignal acquisition device is used for acquiring an electrocardiosignal image to be detected;
the upper computer is in communication connection with the electrocardiosignal acquisition device and is used for realizing the steps of the QRS complex wave detection method based on U-Net when executing a computer program;
and the display device is in communication connection with the upper computer and is used for displaying a QRS complex wave label graph of the electrocardiosignal graph to be detected.
The QRS complex detection method based on the U-Net network disclosed by the application is used for detecting the network U of the QRS complex QRS When the Net is trained, firstly, the noise electrocardiosignals are utilized for pretraining, network parameters after pretraining are obtained, and the QRS complex detection network U is initialized QRS -Net; acquiring original electrocardiosignal, and detecting network U for initialized QRS complex wave QRS -Net performing formal training; network parameters acquired by pre-training are utilized to initialize QRS complex detection network U QRS Net avoiding the use of random in direct trainingThe initialization parameters of the application realize pre-denoising, have good noise immunity, do not need complex denoising operation on the electrocardiosignal before detection, and effectively improve the QRS complex detection network U of the application QRS Robustness of Net; the application utilizes the multi-scale convolution and feature multiplexing module MCRF to enable the input image to pass through three parallel convolution branches with different convolution kernel sizes, extract the effective information of different scales of the input feature image as much as possible, and repeatedly use the shallow features through jump connection, thereby effectively relieving the problem of gradient disappearance; according to the application, on the encoding path and the decoding path, the grouping convolution unit is utilized to reduce the parameter quantity and simultaneously compress the expression of the redundancy characteristic; the conventional convolution unit is utilized to effectively promote the connection between the feature graphs and further extract the image features; and jump connection is arranged at symmetrical positions, deep features and shallow features are spliced, and gradient degradation is relieved.
It will be appreciated by those skilled in the art that embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
It is apparent that the above examples are given by way of illustration only and are not limiting of the embodiments. Other variations and modifications of the present application will be apparent to those of ordinary skill in the art in light of the foregoing description. It is not necessary here nor is it exhaustive of all embodiments. While still being apparent from variations or modifications that may be made by those skilled in the art are within the scope of the application.
Claims (10)
1. The QRS complex detection method based on the U-Net network is characterized by comprising the following steps of:
generating a noise electrocardiosignal by adding the clean electrocardiosignal and noise to obtain pre-training data;
taking the noise electrocardiosignals and the corresponding clean electrocardiosignals as the input and the output of a pre-constructed QRS complex detection network, and pre-training the QRS complex detection network; the QRS complex detection network is constructed based on a U-Net network;
after training for preset times, outputting network parameters of the QRS complex detection network by using the mean square error as a loss function;
collecting an original electrocardiosignal and a corresponding QRS complex label, and initializing a QRS complex detection network by using the network parameters;
inputting the original electrocardiosignal into an initialized QRS complex detection network, continuously reducing the network output result and the Dice loss function of the corresponding QRS complex label, and obtaining a trained QRS complex detection network after training for preset times;
and acquiring an electrocardiosignal image to be detected, inputting the electrocardiosignal image to a trained QRS complex detection network, and acquiring a QRS complex label image of the electrocardiosignal image to be detected.
2. The method for detecting QRS complex based on U-Net network according to claim 1, wherein the acquiring the to-be-detected electrocardiograph signal map, inputting the to-be-detected electrocardiograph signal map into the trained QRS complex detection network, and acquiring the QRS complex label map of the to-be-detected electrocardiograph signal map, includes:
acquiring an electrocardiosignal image to be detected, inputting the electrocardiosignal image to a multi-scale convolution sum feature multiplexing module of a pre-trained QRS complex detection network, and carrying out three parallel convolution branches with different convolution kernel sizes;
an electrocardiograph signal image to be detected sequentially passes through a1 multiplied by 1 convolution unit, a first convolution unit and a second convolution unit which are connected in series along the positive propagation direction in each path of convolution branch; the output images of the three convolution units are spliced to be used as the output images of the convolution branches; splicing the output images of the three paths of parallel convolution branches to serve as an output characteristic image of the multi-scale convolution and characteristic taking module;
after downsampling the output characteristic image, inputting the code path of the QRS complex detection network for downsampling for a plurality of times, and outputting a code image;
after up-sampling the coded image, inputting the coded image into a decoding path of a QRS complex detection network, and outputting a decoded image after up-sampling the coded image for the same times;
inputting the decoded image into a1 multiplied by 1 convolution, and then obtaining a QRS complex wave label graph of the electrocardiosignal graph to be detected through an activation function layer.
3. The QRS complex detection method based on the U-Net network according to claim 2, wherein the first convolution unit and the second convolution unit each include a characteristic disturbance layer Disout, a characteristic discarding layer Dropout, a 1D convolution layer of a preset convolution kernel size, a batch normalization layer BN, an activation function layer ReLU and an effective channel attention layer EAC, which are sequentially connected in series along a forward propagation direction.
4. The U-Net network based QRS complex detection method of claim 2, wherein the downsampling is accomplished with a large step depth convolution module; the large step depth convolution module includes: and outputting the input characteristic image after passing through a characteristic disturbance layer Disout, a characteristic discarding layer Dropout, a 1D depth convolution layer with a preset convolution kernel size, a batch normalization layer BN, an activation function layer ReLU and an effective channel attention layer EAC which are sequentially connected in series along the forward propagation direction, and taking the output characteristic image as the input characteristic image of the next coding layer.
5. The QRS complex detection method based on a U-Net network according to claim 2, wherein the coding path includes four coding layers sequentially connected in series along the forward propagation direction; each coding layer convolves the input characteristic image by using a GConv-Block module and then outputs the coding characteristic image of the coding layer;
the GConv-Block module comprises a first Group convolution unit, a second Group convolution unit and a third Group convolution unit, wherein the first Group convolution unit, the second Group convolution unit and the third Group convolution unit are sequentially connected in series along the positive propagation direction;
performing pixel point addition on the output image of the first grouping convolution unit and the output image of the second grouping convolution unit, and taking the pixel point addition as an input of a third convolution unit;
and carrying out pixel point addition on the output image of the 1 multiplied by 1 convolution and the output image of the third convolution unit to obtain the output of the GConv-Block module.
6. The QRS complex detection method based on the U-Net network according to claim 5, wherein the first packet convolution unit and the second packet convolution unit each include a characteristic disturbance layer Disout, a characteristic discard layer Dropout, a 1D packet convolution layer of a preset convolution kernel size, a batch normalization layer BN, an activation function layer ReLU and an effective channel attention layer EAC, which are sequentially connected in series along a positive propagation direction;
the third convolution unit comprises a characteristic disturbance layer Disout, a characteristic discarding layer Dropout, a 1D convolution layer with a preset convolution kernel size, a batch normalization layer BN, an activation function layer ReLU and an effective channel attention layer EAC which are sequentially connected in series along the positive propagation direction.
7. The QRS complex detection method based on a U-Net network according to claim 5, wherein the decoding path includes four decoding layers sequentially connected in series along the forward propagation direction;
each decoding layer is in jump connection with a corresponding coding layer, an output image of the corresponding coding layer and an output image of an upper decoding layer of the upper layer are obtained, and the output images are spliced in series to be used as input characteristic images of the decoding layers of the layer;
the upsampling uses nearest neighbor interpolation.
8. The method for detecting QRS complex based on U-Net network according to claim 1, wherein after obtaining the QRS complex label map of the electrocardiograph signal map to be detected, post-processing is performed to obtain the position of the QRS complex, and the post-processing includes:
and smoothing the QRS complex label graph by using a moving average algorithm to obtain smoothed peak point coordinates which are the QRS complex position of the electrocardiosignal graph to be detected.
9. A U-Net based QRS complex detection device, comprising:
the pre-training module is used for generating a noise electrocardiosignal by adding the clean electrocardiosignal and the noise to obtain pre-training data; taking the noise electrocardiosignals and the corresponding clean electrocardiosignals as the input and the output of a pre-constructed QRS complex detection network, and pre-training the QRS complex detection network; the QRS complex detection network is constructed based on a U-Net network; after training for preset times, outputting network parameters of the QRS complex detection network by using the mean square error as a loss function;
the formal training module is used for collecting an original electrocardiosignal and a corresponding QRS complex label and initializing a QRS complex detection network by utilizing the network parameters; inputting the original electrocardiosignal into an initialized QRS complex detection network, continuously reducing the network output result and the Dice loss function of the corresponding QRS complex label, and obtaining a trained QRS complex detection network after training for preset times;
the detection module is used for acquiring an electrocardiosignal image to be detected, inputting the electrocardiosignal image to a trained QRS complex detection network and acquiring a QRS complex label image of the electrocardiosignal image to be detected.
10. A U-Net based QRS complex detection device, comprising:
the electrocardiosignal acquisition device is used for acquiring an electrocardiosignal image to be detected;
the upper computer is in communication connection with the electrocardiosignal acquisition device and is used for realizing the steps of the QRS complex detection method based on U-Net according to any one of claims 1 to 8 when executing a computer program;
and the display device is in communication connection with the upper computer and is used for displaying a QRS complex wave label graph of the electrocardiosignal graph to be detected.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310417011.9A CN116712083A (en) | 2023-04-19 | 2023-04-19 | QRS complex wave detection method based on U-Net network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310417011.9A CN116712083A (en) | 2023-04-19 | 2023-04-19 | QRS complex wave detection method based on U-Net network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116712083A true CN116712083A (en) | 2023-09-08 |
Family
ID=87868545
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310417011.9A Pending CN116712083A (en) | 2023-04-19 | 2023-04-19 | QRS complex wave detection method based on U-Net network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116712083A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117357129A (en) * | 2023-11-13 | 2024-01-09 | 齐鲁工业大学(山东省科学院) | Electrocardiogram QRS waveform detection method for wearable equipment |
| CN117770832A (en) * | 2024-02-28 | 2024-03-29 | 泰州市新起点创意科技有限公司 | Electrocardiosignal error marking training sample identification method based on cross verification |
| CN117357129B (en) * | 2023-11-13 | 2024-06-04 | 齐鲁工业大学(山东省科学院) | Electrocardiogram QRS waveform detection method for wearable equipment |
-
2023
- 2023-04-19 CN CN202310417011.9A patent/CN116712083A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117357129A (en) * | 2023-11-13 | 2024-01-09 | 齐鲁工业大学(山东省科学院) | Electrocardiogram QRS waveform detection method for wearable equipment |
| CN117357129B (en) * | 2023-11-13 | 2024-06-04 | 齐鲁工业大学(山东省科学院) | Electrocardiogram QRS waveform detection method for wearable equipment |
| CN117770832A (en) * | 2024-02-28 | 2024-03-29 | 泰州市新起点创意科技有限公司 | Electrocardiosignal error marking training sample identification method based on cross verification |
| CN117770832B (en) * | 2024-02-28 | 2024-04-26 | 泰州市新起点创意科技有限公司 | Electrocardiosignal error marking training sample identification method based on cross verification |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108714026B (en) | Fine-grained electrocardiosignal classification method based on deep convolutional neural network and online decision fusion | |
| CN107837082B (en) | Automatic electrocardiogram analysis method and device based on artificial intelligence self-learning | |
| CN110840402B (en) | Atrial fibrillation signal identification method and system based on machine learning | |
| Javadi et al. | Classification of ECG arrhythmia by a modular neural network based on mixture of experts and negatively correlated learning | |
| Cao et al. | Atrial fibrillation detection using an improved multi-scale decomposition enhanced residual convolutional neural network | |
| Kumar et al. | Fuzz-ClustNet: Coupled fuzzy clustering and deep neural networks for Arrhythmia detection from ECG signals | |
| CN109124620B (en) | Atrial fibrillation detection method, device and equipment | |
| Kang et al. | A method of denoising multi-channel EEG signals fast based on PCA and DEBSS algorithm | |
| CN106419898A (en) | Method removing electrocardiosignal baseline drift | |
| Agrawal et al. | ECG-iCOVIDNet: Interpretable AI model to identify changes in the ECG signals of post-COVID subjects | |
| Peláez et al. | Ischemia classification via ECG using MLP neural networks | |
| CN116712083A (en) | QRS complex wave detection method based on U-Net network | |
| Tseng et al. | Sliding large kernel of deep learning algorithm for mobile electrocardiogram diagnosis | |
| CN116361688A (en) | Multi-mode feature fusion model construction method for automatic classification of electrocardiographic rhythms | |
| An et al. | Effective data augmentation, filters, and automation techniques for automatic 12-lead ECG classification using deep residual neural networks | |
| EP4041073A1 (en) | Systems and methods for electrocardiogram diagnosis using deep neural networks and rule-based systems | |
| CN113180688B (en) | Coronary heart disease electrocardiogram screening system and method based on residual error neural network | |
| CN113180685B (en) | Electrocardio abnormity discrimination system and method based on morphological filtering and wavelet threshold | |
| CN113384277A (en) | Electrocardiogram data classification method and classification system | |
| Allam et al. | A deformable CNN architecture for predicting clinical acceptability of ECG signal | |
| Berger et al. | Generative adversarial networks in electrocardiogram synthesis: Recent developments and challenges | |
| Kar et al. | A technical review on statistical feature extraction of ECG signal | |
| CN111528833B (en) | Rapid identification and processing method and system for electrocardiosignals | |
| Sathawane et al. | Prediction and analysis of ECG signal behaviour using soft computing | |
| CN116035593B (en) | Electrocerebral noise reduction method based on generation countermeasure type parallel neural network |
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 |