CN111184508B - Electrocardiosignal detection device and analysis method based on joint neural network - Google Patents
Electrocardiosignal detection device and analysis method based on joint neural network Download PDFInfo
- Publication number
- CN111184508B CN111184508B CN202010063050.XA CN202010063050A CN111184508B CN 111184508 B CN111184508 B CN 111184508B CN 202010063050 A CN202010063050 A CN 202010063050A CN 111184508 B CN111184508 B CN 111184508B
- Authority
- CN
- China
- Prior art keywords
- neural network
- data
- electrocardiosignal
- intelligent mobile
- mobile equipment
- 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.)
- Active
Links
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—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
- A61B5/0002—Remote monitoring of patients using telemetry, e.g. transmission of vital signals via a communication network
- A61B5/0004—Remote monitoring of patients using telemetry, e.g. transmission of vital signals via a communication network characterised by the type of physiological signal transmitted
- A61B5/0006—ECG or EEG signals
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/316—Modalities, i.e. specific diagnostic methods
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/68—Arrangements of detecting, measuring or recording means, e.g. sensors, in relation to patient
- A61B5/6801—Arrangements of detecting, measuring or recording means, e.g. sensors, in relation to patient specially adapted to be attached to or worn on the body surface
- A61B5/6802—Sensor mounted on worn items
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7203—Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
-
- 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
- A61B5/7253—Details of waveform analysis characterised by using transforms
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
- A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
Abstract
The invention discloses an electrocardiosignal detection device and an analysis method based on a joint neural network, which comprises the steps of firstly, building a joint neural network algorithm on a machine learning server and training a model, extracting data space characteristics and obtaining space classification probability by the model through a residual error neural network module aiming at preprocessed electrocardio data, extracting time sequence characteristics of data on a space characteristic diagram after dimension reduction through a bidirectional long-short term memory neural network and an attention module and obtaining time sequence classification probability, and finally fusing the two classification probabilities to obtain a detection result; acquiring a small amount of electrocardiogram data of a patient from the wearable device, inputting the electrocardiogram data into a machine learning server for model fine-tuning after manual marking, and deploying a final model to intelligent mobile equipment; and finally, the wearable device and the intelligent mobile equipment are wirelessly transmitted to realize real-time anomaly detection. The wearable device for acquiring and detecting the electrocardiosignals in real time is developed, and an effective technical means is provided for the auxiliary diagnosis of heart diseases.
Description
Technical Field
The invention belongs to the field of electrocardiosignal abnormality detection, and particularly relates to an electrocardiosignal wearable device based on a joint neural network algorithm and an abnormality detection analysis method.
Background
The electrocardiosignal is drawn by transmitting current generated by heart activity to the body surface, can exactly reflect the activity state of the heart, and is one of the important means for doctors to diagnose the heart disease. The electrocardiosignal is a weak physiological signal with a small voltage value and a small time interval, so that the diagnosis efficiency of a doctor can be improved by extracting the electrocardiosignal characteristics and assisting detection by a computer. In recent years, with the continuous development and improvement of artificial intelligence algorithms, algorithms commonly used for electrocardiographic signal detection and classification are mainly classified into traditional machine learning methods and deep learning methods.
The method for recognizing electrocardiosignals based on traditional machine learning generally needs several processes of data preprocessing, feature analysis, feature extraction, classifier construction and the like. Document [1] proposes an MEES method to detect and locate myocardial infarction disease features in electrocardiosignals, and classifies the features based on the features by using a support vector machine as a classifier; document [2] de-noising data by using DWT, proposing a new electrocardiosignal characteristic by fitting a 20-order polynomial function with an electrocardiosignal, and finally realizing classification of the electrocardiosignal by using a decision tree; document [3] extracts 47 myocardial infarction features after DWT denoising, and finally inputs the features into a kNN algorithm as a final classifier. In the process of using the traditional machine learning method, each step needs manual analysis and regulation intervention, but because of complexity and variability of electrocardiosignal characteristics, manual extraction of the signal characteristics is difficult, and the effect of the classifier is influenced by each previous process, so that the classification effect is unstable and the process is complex.
Compared with the traditional machine learning method, the whole process of the deep learning method is automatic, and features do not need to be extracted manually. The common algorithm for detecting abnormal electrocardiosignals based on deep learning mainly comprises a model constructed based on a convolutional neural network and a model constructed based on a cyclic neural network, which can respectively extract the spatial characteristics and the time sequence characteristics of the electrocardiosignals. Documents [4,5] respectively construct 11 layers of CNN framework and 34 layers of residual network framework to realize myocardial infarction detection and arrhythmia classification; document [6] uses an LSTM framework to extract the time sequence characteristics of electrocardiosignals to realize the classification of arrhythmia; document [7,8] uses superimposed CNN and LSTM models, first uses the CNN architecture to extract data space features, on this basis uses LSTM to extract timing features, and realizes classification of electrocardiosignals according to the output of LSTM. Many studies build models using only one type of neural network, resulting in inefficient use of the electrocardiographic features. And although some researchers use the CNN and LSTM models at the same time, the superimposed structure can cause that most of errors are distributed to the LSTM network closer to the classification layer when the neural network algorithm reversely propagates the update parameters, so that the CNN network of the shallow extraction space cannot be updated to the maximum extent. In addition, many researchers do not consider the problem of patient specificity when training and testing algorithms, that is, training set and testing set data used by them may come from the same patient, which may result in poor generalization ability of the final model and failure to accurately detect electrocardiographic data of new patients.
Reference documents:
[1]L.Sharma,R.Tripathy,S.Dandapat,Multiscale energy and eigenspace approach to detection and localization of myocardial infarction,IEEE Trans.Biomed.Eng.62 (2015)1827–1837.
[2]B.Liu,J.Liu,G.Wang,K.Huang,F.Li,Y.Zheng,Y.Luo,F.Zhou,A novel electrocardiogramparameterization algorithm and its application in myocardial infarction detection,Comput.Biol.Med.61(2015)178–184.
[3]U.R.Acharya,H.Fujita,V.K.Sudarshan,S.L.Oh,M.Adam,J.E.Koh,J.H.Tan, D.N.Ghista,R.J.Martis,C.K.Chua,others,Automated detection and localization of myocardial infarction usingelectrocardiogram:a comparative study of different leads, Knowl.-Based Syst.99(2016)146–156.
[4]U.R.Acharya,H.Fujita,S.L.Oh,Y.Hagiwara,J.H.Tan,M.Adam,Application of deep convolutional neural network for automated detection of myocardial infarction using ECG signals,Inf.Sci.415(2017)190–198.
[5]A.Y.Hannun,P.Rajpurkar,M.Haghpanahi,G.H.Tison,C.Bourn,M.P.Turakhia, A.Y.Ng,Cardiologist-level arrhythmia detection and classification in ambulatory electrocardiograms usinga deep neural network,Nat.Med.25(2019)65.
[6]S.Saadatnejad,M.Oveisi,M.Hashemi,LSTM-Based ECG Classification for Continuous Monitoring on Personal Wearable Devices,IEEE J.Biomed.Health Inform.(2019).
[7]Q.Yao,R.Wang,X.Fan,J.Liu,Y.Li,Multi-class Arrhythmia detection from 12-lead varied-length ECG usingAttention-based Time-Incremental Convolutional Neural Network,Inf.Fusion.53(2020)174–182.
[8]J.H.Tan,Y.Hagiwara,W.Pang,I.Lim,S.L.Oh,M.Adam,R.San Tan,M.Chen, U.R.Acharya,Application of stacked convolutional and long short-term memory network for accurate identification of CAD ECG signals,Comput.Biol.Med.94 (2018)19–26.
disclosure of Invention
Aiming at the problems, the invention designs the efficient wearable acquisition device for the ground layer signals, which can acquire the single-lead electrocardiosignal in real time and realize real-time and accurate electrocardiosignal abnormity detection through wireless Bluetooth and mobile intelligent equipment; meanwhile, a parallel structure is used for introducing various types of neural network algorithms, the defects are overcome, a combined neural network model capable of automatically extracting electrocardio digital signal space characteristics and time sequence characteristics is constructed, data sets used for training and testing of the model come from different patients, final training of the model fine-tuning is completed by using new data collected by the device, and the detection capability of the model is improved.
The technical scheme adopted by the device of the invention is as follows: the utility model provides an electrocardiosignal detection device based on unite neural network which characterized in that: the electrocardiosignal acquisition unit is connected with the electrode through a communication network;
the electrodes are arranged on the surface of the human thorax;
the electrocardiosignal acquisition unit is connected with the electrode through a wire and is used for extracting a single-lead original signal of the electrocardiosignal, converting the single-lead original signal into an electrocardio digital signal after filtering, amplifying and A/D conversion processing operations and transmitting the electrocardio digital signal to the signal processing and transmitting unit;
the signal processing and transmission unit is used for denoising, resampling, segmenting and normalizing the electrocardio digital signals and realizing wireless communication with the intelligent mobile equipment through Bluetooth;
the intelligent mobile device is used for realizing real-time abnormal detection of the electrocardiogram data, giving an alarm in real time when abnormal data are encountered, and sending the data to a hospital for further judgment.
The method adopts the technical scheme that: an electrocardiosignal analysis method based on a combined neural network is characterized by comprising the following steps:
step 1: extracting a large number of single-lead electrocardio digital signals with labels from an electrocardio database containing a plurality of different patients and preprocessing the signals;
step 2: dividing the processed electrocardiogram data into a training set, a verification set and a test set, wherein the data of the three data sets are from different patients respectively;
and step 3: constructing a joint neural network on a machine learning server;
and 4, step 4: firstly, a training set, a verification set and a test set are used for completing preliminary training and testing of the combined neural network on a machine learning server;
and 5: acquiring and processing data by using an electrocardiosignal detection device based on a joint neural network, inputting the data after artificial marking into a machine learning server to perform fine-tuning on the joint neural network, and optimizing the joint neural network and then deploying the optimized joint neural network to intelligent mobile equipment;
step 6: the electrocardiosignal detection device based on the combined neural network is used for collecting electrocardiosignals of a human body in real time, and intelligent mobile equipment is used for analyzing and feeding back electrocardio digital signals.
The electrodes comprise RA and LA electrodes arranged below the left and right clavicles of the human body and RL electrodes arranged on the right abdomen part and used as reference electrodes, and the electrodes form a differential loop; the electrodes are connected to the ADS1298 for acquisition of cardiac electrical signals,
preferably, the electrocardiosignal acquisition unit is an ADS1298 chip.
Preferably, the signal processing and transmitting unit comprises an MCU chip and a BLE chip.
Preferably, the wearable device comprises an electrode, an electrocardiosignal acquisition unit, a signal processing and transmission unit and intelligent mobile equipment.
Preferably, the specific implementation of step 1 comprises the following sub-steps:
step 1.1: denoising the central electric digital signal of the electrocardiogram database by using wavelet transformation;
step 1.2: resampling the signal;
step 1.3: cutting the signal into samples of the same length;
step 1.4: normalization processing is performed for each sample.
Preferably, the specific implementation of step 3 comprises the following sub-steps:
step 3.1: setting a common rough extraction characteristic of the convolutional layer;
step 3.2: extracting spatial features by using a residual error module, then integrating a feature map by using a global mean pooling, and finally inputting a full connection layer and softmax to output spatial feature classification probabilities;
step 3.3: on the basis of spatial features, extracting data time sequence features by using a bidirectional long-short term memory network, then enhancing the extraction of local time features by using an attention mechanism and integrating the time sequence features, and finally inputting a full-connection layer and softmax to output a time sequence feature classification probability;
step 3.4: adding different weights to the classification probabilities of step 3.2 and step 3.3, respectively, and adding to output a final detection result according to the following formula:
Yo＝α*f1+(1-α)*f2，α∈[0，1]
wherein Y isoRepresenting the final classification output probability of the algorithm, f1And f2Denotes the softmax classification probability output based on temporal and spatial features, respectively, α and (1- α) denote their assigned weights, and α has a value between 0 and 1.
Preferably, in step 4, the best value of the hyper-parameter α is found according to the accuracy of the verification set.
Preferably, in step 5, the artificial mark is input into a machine learning server to perform fine-tuning on the joint neural network.
Compared with the prior art, the invention has prominent substantive characteristics and remarkable progress, and specifically comprises the following steps:
1. the invention combines a residual error neural network and a bidirectional long and short term memory neural network (BilSTM), respectively extracts the spatial characteristics and the time sequence characteristics of the electrocardio digital signals, simultaneously introduces an attention mechanism in the bidirectional long and short term memory network, adds different weights for different time points of the extracted time sequence characteristics, and enhances the extraction of the local time sequence characteristics; and then respectively distributing proper weights to the two characteristics based on the classification probabilities of the two characteristics, and finally fusing the two probabilities to obtain a final detection result. The whole feature extraction process is automatic, any waveform detection and manual feature analysis process is not needed, and meanwhile the branch architecture can ensure that errors can be distributed to all branches in parallel to fully update parameters when the neural network propagates in the reverse direction.
2. When the model is trained and tested, the training data and the testing data come from different patients, so that the trained and tested model can realize accurate detection on new patient data, and has stronger generalization capability.
3. The algorithm provided by the invention can realize accurate detection on the single-lead electrocardiogram data, effectively reduces the size of the model and the calculation complexity, enables the model to be deployed and operated on various intelligent devices such as an FPGA (field programmable gate array), an intelligent bracelet and an intelligent mobile phone, and meanwhile, the single-lead data relieves the delay problem of transmission between the intelligent mobile device and the wearable device.
4. According to the invention, the electrocardio data in the electrocardio database with the label is used for carrying out preliminary training and testing on the model on the machine learning server, and meanwhile, a small amount of data of the acquisition device is manually marked and then input into the machine learning server for finding-tuning the model, so that the model adapts to the characteristics of new data, the model detection accuracy is further effectively improved, and meanwhile, the training mode only needs to manually mark a small amount of acquired data, and the cost of manual marking is greatly reduced.
5. According to the method, the ADS1298 is used for acquiring electrocardiosignals, the MCU is used for carrying out edge calculation optimization processing on the acquired electrocardio digital signals by utilizing the characteristic of real-time high-performance calculation, and meanwhile, the wireless BLE 5.1 is combined to transmit data to the intelligent mobile equipment in real time for detection, so that the method has an accurate sound positioning function and improves the practicability.
Drawings
FIG. 1 is a flowchart of the overall method of an embodiment of the present invention;
FIG. 2 is a sample graph of processed ECG data according to an embodiment of the invention;
FIG. 3 is a block diagram of two residual error modules in an embodiment of the present invention;
FIG. 4 is a diagram of the operating principle of BilSTM in an embodiment of the present invention;
FIG. 5 is a schematic diagram of an attention mechanism in an embodiment of the present invention;
FIG. 6 is a graph illustrating the accuracy of validation sets under different hyper-parameters in an embodiment of the present invention;
FIG. 7 is a graph of spatial and temporal features extracted in an embodiment of the present invention;
FIG. 8 is a diagram illustrating a process of acquiring and detecting electrocardiographic data in real time according to an embodiment of the present invention.
Detailed Description
In order to facilitate the understanding and implementation of the present invention for those of ordinary skill in the art, the present invention is further described in detail with reference to the accompanying drawings and examples, it is to be understood that the embodiments described herein are merely illustrative and explanatory of the present invention and are not restrictive thereof.
As shown in fig. 1, the electrocardiographic signal abnormality detection method and wearable device based on the joint neural network provided by the present invention include the following steps:
step 1: preprocessing the single-lead electrocardiosignal in the PTB electrocardio database, and the specific process is as follows:
step 1.1: denoising the central electric digital signal of the electrocardiogram database by using wavelet transformation;
step 1.2: dividing the electrocardiosignals with labels in the database into two categories of positive abnormality and abnormal abnormality;
step 1.3: sampling the signal as 480Hz data points;
step 1.4: the electrocardio single-lead digital signal is divided into 5 seconds of samples, so that sufficient characteristics of the data are ensured;
step 1.5: each sample was normalized using the Z-score method.
In this embodiment, the collected data in the ecg database (the ecg database which is open source in hospital or on the internet and has undergone the processing operations of "filtering, amplifying, and a/D conversion") is used for the initial training and testing
The processed positive exception sample is shown in FIG. 2. Dividing the electrocardiogram data into a training set, a verification set and a test set, wherein data samples of the training set, the verification set and the test set are respectively from different patients, and generating more training set samples on the basis of the signal processing in the step 1 by using the following formula and keeping the relative balance of the number of normal and abnormal class samples;
where N and N represent the number of sample generation targets and the number of original samples, respectively, L represents the length of each sample, round represents rounding, and S represents the number of overlapping sample points.
A joint neural network algorithm is built on a machine learning server by using Tensorflow as a framework, and the algorithm mainly comprises the following structures:
firstly, setting a common convolutional layer for crude extraction of features, wherein the features are extracted in a coarse mode, the size of a convolutional kernel is 1x15, and the number of the convolutional kernels is 8;
the spatial features of the data are extracted using 8 residual blocks, two types of residual blocks are alternately connected, each residual block comprising two convolution layers, two Batch Normalization (BN) layers, two ReLU layers and one Dropout layer, as shown in fig. 3. The BN layer and the ReLU can prevent gradient explosion or disappearance and accelerate the training speed of the model. Dropout effectively prevents the network from overfitting. To reduce the dimension, the even number of residual blocks add a maxporoling layer with the convolution step size set to 2. After passing through each residual module, the size and the number of convolution kernels are changed along with the size of the feature map and the depth of the network, the size is changed into 1x15, 1x9, 1x5 and 1x3, and the number of the convolution kernels is changed into 8, 16, 32 and 64. Global mean pooling is then used instead of the traditional fully connected layer integrated spatial feature map, effectively reducing the parameters. Finally, obtaining spatial feature classification probability by using a full connection layer and a Softmax layer;
on the basis of the spatial features, the BilSTM is used to continuously extract the data time sequence features, and the principle is shown in FIG. 4. In contrast to conventional recurrent neural networks, BilSTM can predict the current time step in conjunction with past and future information from the time series data. The upper half of fig. 4 is a BiLSTM general working principle diagram, and the lower half is an internal state updating process at time t. The calculation steps at time step t when information is transferred in the forward direction are as follows:
Xtinput data representing the time step of t,
the output of the last time step is represented,
and
respectively representing the updating of the state of each part inside, and determining which information is deleted, added, updated at the moment and output to the next moment before and at the moment.
The weight and bias, tanh and σ, of each part represent the activation function. The output at each time step is then assigned a weight using an attention mechanism to make the local significant time points more prominent. As shown in fig. 5. The output vector of each time step passes through oneAfter learning the function, outputting a same-dimension weight, correspondingly multiplying the same-dimension weight by the output of each time step, adding and integrating each time step value in the weighted time sequence vector, and obtaining a time sequence feature classification probability by using a full connection layer and a Softmax layer;
and finally, adding different weights for the spatial feature probability and the time sequence feature probability respectively to adjust classification result scores, and adding the classification results to obtain the final classification result probability. As shown in the following equation;
Yo＝α*f1+(1-α)*f2，α∈[0，1]
wherein Y isoRepresenting the final classification output probability of the algorithm, f1And f2Denotes the softmax classification probability output based on temporal and spatial features, respectively, α and (1- α) denote their assigned weights, and α has a value between 0 and 1.
Under the machine learning server, using the training set and the verification set, using a random gradient descent algorithm as an optimizer, momentum is set to 0.9, epoch is set to 20, initial learning rate is set to 0.1, and epoch is multiplied by 0.1 every 5 times. The process is executed ten times, the fusion hyper-parameter is respectively equal to the values in the array [0.1,0.2,0.3,0.4,0.5,0.6,0.7,0.8,0.9], the best value in the ten groups of data is found according to the verification result, and the model is used as the optimal model. The accuracy of the validation set of different models is shown in fig. 6, and when the accuracy is 0.6, the model effect is the best.
Under the machine learning server, the trained model is tested using a test set. FIG. 7 is a diagram illustrating spatial and temporal features of a positive anomaly sample extracted in a joint neural network model test, wherein the shading in the temporal feature diagram is attention weight distribution. And different modules of the model are tested through an ablation experiment, and the test results are shown in the following table 1.
TABLE 1
| Pre.(％) | Rec.(％) | F1.(％) | Acc.(％) | |
| Based on spatial features | 87.71 | 87.94 | 87.82 | 93.11 |
| Based on time-series characteristics (without attention) | 92.31 | 91.92 | 91.32 | 95.44 |
| Based on time-series characteristics (with attention) | 91.55 | 94.49 | 92.93 | 95.86 |
| Feature classification probability fusion | 92.67 | 94.32 | 93.47 | 96.24 |
Experimental results show that each module plays an independent role in the detection process, the effect is gradually improved, and the complete model finally achieves 96.24% of detection accuracy.
A small amount of partial data are selected from the wearable device, are denoised by wavelet transformation, are cut into samples with the length of 5 seconds, are normalized by a Z-score mode, are manually marked and then are input into a machine learning server, and fine-tuning is carried out on the model subjected to preliminary test, so that the model adapts to the electrocardio data characteristics of a patient.
The parameters of the final model are 10 thousands, the calculation complexity is 0.09GFLOPs, the model is converted into a program in a TensorFlow Lite file format by using a TensorFlow Lite Converter, the size and the speed of the model are optimized, and the final model is deployed to the intelligent mobile device.
As shown in fig. 8, the electrocardiographic signal detection apparatus based on the joint neural network provided by the present invention includes an electrode, an electrocardiographic signal acquisition unit, a signal processing and transmission unit, and an intelligent mobile device; the electrode, the electrocardiosignal acquisition unit, the signal processing and transmission unit and the intelligent mobile device form a wearable device.
The electrodes are arranged on the surface of the human thorax; the electrocardiosignal acquisition unit is connected with the electrode through a wire and is used for extracting a single-lead original signal of the electrocardiosignal, converting the original signal into a digital signal after filtering, amplifying and A/D conversion processing operations and transmitting the digital signal to the signal processing and transmission unit; the signal processing and transmission unit is used for carrying out denoising, resampling, segmenting and normalization processing on the obtained electrocardio digital signal and realizing wireless communication with the intelligent mobile equipment through a network; and the intelligent mobile equipment is used for realizing real-time abnormal detection of the electrocardiogram data, giving an alarm in real time when abnormal data are encountered, and sending the data to the hospital for further judgment.
Consistent with a designed combined neural network algorithm, the differential electrocardiosignals are obtained in a bipolar single-lead mode in the embodiment, wherein the differential electrocardiosignals comprise RA and LA electrodes below the left and right clavicles of a human body and RL electrodes arranged on the right abdomen and used as reference electrodes, and the electrodes form a differential loop; the electrode is connected to the ADS1298 to collect the original electrocardio signal and convert the original electrocardio signal into an electrocardio digital signal, the analog power supply and the digital power supply are independently powered, and the analog ground and the digital ground are isolated by a 0R resistor, so that the interference of the front-end collection of the electrocardio signal can be greatly reduced; the MCU + BLE 5.1 integrated chip is designed, in order to be in butt joint with the input of a joint neural network algorithm, the MCU firstly carries out preprocessing operations of denoising, resampling, segmenting and normalizing processing on an electrocardio digital signal, then the wearable device realizes wireless transmission with the intelligent mobile device through BLE 5.1, finally the real-time abnormal detection function of the electrocardio data is realized at the end of the intelligent mobile device, the abnormal data is detected to give an alarm in real time, and the data is sent to a hospital to further judge the disease cause.
It should be understood that parts of the specification not set forth in detail are well within the prior art.
Finally, it should be noted that the above embodiments are only used for illustrating the technical solutions of the present invention and not for limiting the same; although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art will understand that: modifications to the specific embodiments of the invention or equivalent substitutions for parts of the technical features may be made; without departing from the spirit of the present invention, it is intended to cover all aspects of the invention as defined by the appended claims.
Claims (7)
1. An electrocardiosignal analysis method based on a joint neural network adopts an electrocardiosignal detection device based on the joint neural network;
the method is characterized in that: the device comprises an electrode, an electrocardiosignal acquisition unit, a signal processing and transmission unit and intelligent mobile equipment; the electrodes are arranged on the surface of the human thorax; the electrocardiosignal acquisition unit is connected with the electrode through a wire and is used for extracting single-lead original electrocardiosignals, converting the electrocardiosignals into electrocardio digital signals after filtering, amplifying and A/D conversion processing operations and transmitting the electrocardio digital signals to the signal processing and transmitting unit; the signal processing and transmission unit is used for denoising, resampling, segmenting and normalizing the electrocardio digital signals and realizing wireless communication with the intelligent mobile equipment through Bluetooth; the intelligent mobile equipment is used for realizing real-time abnormal detection of the electrocardiogram data, giving an alarm in real time when abnormal data are encountered, and sending the data to a hospital for further judgment;
the method comprises the following steps:
step 1: extracting a large number of single-lead electrocardio digital signals with labels from an electrocardio database containing a plurality of different patients and preprocessing the signals;
step 2: dividing the processed electrocardiogram data into a training set, a verification set and a test set, wherein the data of the three data sets are from different patients respectively;
and step 3: constructing a joint neural network on a machine learning server;
the specific implementation of the step 3 comprises the following substeps:
step 3.1: setting a common rough extraction characteristic of the convolutional layer;
step 3.2: extracting spatial features by using a residual error module, then integrating a feature map by using a global mean pooling, and finally inputting a full connection layer and softmax to output spatial feature classification probabilities;
step 3.3: on the basis of spatial features, extracting data time sequence features by using a bidirectional long-short term memory network, then enhancing the extraction of local time features by using an attention mechanism and integrating the time sequence features, and finally inputting a full-connection layer and softmax to output a time sequence feature classification probability;
step 3.4: adding different weights to the classification probabilities of step 3.2 and step 3.3, respectively, and adding to output a final detection result according to the following formula:
Yo＝α*f1+(1-α)*f2，α∈[0，1]
wherein Y isoRepresenting the final classification output probability of the algorithm, f1And f2Respectively representing the probability of softmax classification based on the output of the time sequence and spatial features, alpha and (1-alpha) representing the weight assigned to them, and the value of alpha is between 0 and 1;
and 4, step 4: firstly, a training set, a verification set and a test set are used for completing preliminary training and testing of the combined neural network on a machine learning server;
and 5: acquiring and processing data by using an electrocardiosignal detection device based on a joint neural network, inputting the data after artificial marking into a machine learning server to perform fine-tuning on the joint neural network, and deploying the optimized joint neural network to intelligent mobile equipment;
step 6: the electrocardiosignal detection device based on the combined neural network is used for collecting electrocardiosignals of a human body in real time, and intelligent mobile equipment is used for analyzing and feeding back the electrocardiosignals.
2. The method of claim 1, wherein: the electrodes comprise RA and LA electrodes arranged below the left and right clavicles of the human body and RL electrodes arranged on the right abdomen part and used as reference electrodes, and the electrodes form a differential loop; the electrodes are connected to the ADS1298 for collecting electrocardiosignals.
3. The method of claim 1, wherein: the electrocardiosignal acquisition unit is an ADS1298 chip.
4. The method of claim 1, wherein: the signal processing and transmitting unit comprises an MCU chip and a BLE chip.
5. The method according to any one of claims 1 to 4, wherein: the wearable device is composed of an electrode, an electrocardiosignal acquisition unit, a signal processing and transmission unit and intelligent mobile equipment.
6. The electrocardiosignal analysis method based on the joint neural network as claimed in claim 1, wherein the step 1 is realized by the following substeps:
step 1.1: denoising the central electric digital signal of the electrocardiogram database by using wavelet transformation;
step 1.2: resampling the signal;
step 1.3: cutting the signal into samples of the same length;
step 1.4: normalization processing is performed for each sample.
7. The electrocardiosignal analysis method based on the joint neural network as claimed in claim 1, wherein: and 4, finding the optimal value of the hyper-parameter alpha according to the accuracy of the verification set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010063050.XA CN111184508B (en) | 2020-01-19 | 2020-01-19 | Electrocardiosignal detection device and analysis method based on joint neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010063050.XA CN111184508B (en) | 2020-01-19 | 2020-01-19 | Electrocardiosignal detection device and analysis method based on joint neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111184508A CN111184508A (en) | 2020-05-22 |
| CN111184508B true CN111184508B (en) | 2021-05-18 |
Family
ID=70686014
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010063050.XA Active CN111184508B (en) | 2020-01-19 | 2020-01-19 | Electrocardiosignal detection device and analysis method based on joint neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111184508B (en) |
Families Citing this family (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111711585B (en) * | 2020-06-11 | 2021-06-22 | 西安交通大学 | Real-time signal sequence detection method based on deep learning |
| CN111916215B (en) * | 2020-06-17 | 2022-04-29 | 扬州大学附属医院 | Cluster/chronic disease early automatic alarm, preliminary feature analysis and risk assessment system |
| CN112022142B (en) * | 2020-08-07 | 2023-10-17 | 上海联影智能医疗科技有限公司 | Electrocardiosignal type identification method, device and medium |
| CN112137613B (en) * | 2020-09-01 | 2024-02-02 | 沈阳东软智能医疗科技研究院有限公司 | Determination method and device of abnormal position, storage medium and electronic equipment |
| CN112161173B (en) * | 2020-09-10 | 2022-05-13 | 国网河北省电力有限公司检修分公司 | Power grid wiring parameter detection device and detection method |
| CN111920375B (en) * | 2020-09-30 | 2021-01-15 | 平安科技（深圳）有限公司 | Vascular Endothelial Growth Factor (VEGF) resistance curative effect prediction device and method |
| CN112270240B (en) * | 2020-10-22 | 2024-02-09 | 微医云(杭州)控股有限公司 | Signal processing method, device, electronic equipment and storage medium |
| CN112244853B (en) * | 2020-10-26 | 2022-05-13 | 生物岛实验室 | Edge computing node manufacturing method and edge computing node |
| CN112508110A (en) * | 2020-12-11 | 2021-03-16 | 哈尔滨理工大学 | Deep learning-based electrocardiosignal graph classification method |
| CN112957052B (en) * | 2021-01-25 | 2023-06-23 | 北京工业大学 | Multi-lead electrocardiosignal classification method based on NLF-CNN lead fusion depth network |
| CN113080990B (en) * | 2021-03-25 | 2023-01-10 | 南京蝶谷健康科技有限公司 | Heart beat anomaly detection method based on CycleGAN and BilSTM neural network method |
| CN113349793A (en) * | 2021-03-30 | 2021-09-07 | 北京芯动卫士科技有限公司 | Arrhythmia classification method based on layered attention model |
| CN113191399B (en) * | 2021-04-12 | 2022-05-27 | 普赛微科技(杭州)有限公司 | Method for improving yield of semiconductor chips based on machine learning classifier |
| CN113288163A (en) * | 2021-06-04 | 2021-08-24 | 浙江理工大学 | Multi-feature fusion electrocardiosignal classification model modeling method based on attention mechanism |
| CN113965632A (en) * | 2021-11-03 | 2022-01-21 | 武汉大学 | ECG analysis system and method based on cloud-equipment terminal cooperation |
| CN114224354B (en) * | 2021-11-15 | 2024-01-30 | 吉林大学 | Arrhythmia classification method, arrhythmia classification device, and readable storage medium |
| CN114129171B (en) * | 2021-12-01 | 2022-06-03 | 山东省人工智能研究院 | Electrocardiosignal noise reduction method based on improved residual error dense network |
| CN114176602B (en) * | 2021-12-30 | 2024-04-26 | 重庆康如来科技有限公司 | Method for simultaneously positioning electrocardiograph P wave, QRS wave and T wave based on deep learning multi-target detection |
| CN117393153B (en) * | 2023-12-11 | 2024-03-08 | 中国人民解放军总医院 | Shock real-time risk early warning and monitoring method and system based on medical internet of things time sequence data and deep learning algorithm |
| CN117598711B (en) * | 2024-01-24 | 2024-04-26 | 中南大学 | QRS complex detection method, device, equipment and medium for electrocardiosignal |
Family Cites Families (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105595990A (en) * | 2016-01-27 | 2016-05-25 | 浙江铭众科技有限公司 | Intelligent terminal device for evaluating and distinguishing quality of electrocardiosignal |
| CN105769173B (en) * | 2016-02-29 | 2018-08-28 | 浙江铭众科技有限公司 | A kind of cardioelectric monitor system with electrocardiosignal noise removal function |
| US11432778B2 (en) * | 2017-01-24 | 2022-09-06 | General Electric Company | Methods and systems for patient monitoring |
| US20190175017A1 (en) * | 2017-05-03 | 2019-06-13 | Narmadha Kuppuswami | Maternity Monitoring System and Method |
| CN108416277A (en) * | 2018-02-11 | 2018-08-17 | 广州市碳码科技有限责任公司 | A kind of cardioelectric monitor method, apparatus, terminal and computer readable storage medium |
| CN108577831B (en) * | 2018-03-19 | 2021-02-09 | 武汉海星通技术股份有限公司 | Single-guide-core-paste data long-range monitoring and diagnosing system and processing method thereof |
| CN109998532A (en) * | 2019-06-04 | 2019-07-12 | 广州视源电子科技股份有限公司 | Electrocardiosignal recognition methods and device based on the more topology convergence networks of multi-lead |
| CN110251120B (en) * | 2019-06-06 | 2021-03-30 | 山东省计算中心（国家超级计算济南中心） | Electrocardiogram classification method of depth residual error neural network based on attention mechanism |
-
2020
- 2020-01-19 CN CN202010063050.XA patent/CN111184508B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN111184508A (en) | 2020-05-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111184508B (en) | Electrocardiosignal detection device and analysis method based on joint neural network | |
| CN108836302B (en) | Intelligent electrocardiogram analysis method and system based on deep neural network | |
| CN111990989A (en) | Electrocardiosignal identification method based on generation countermeasure and convolution cyclic network | |
| CN112741638B (en) | Medical diagnosis auxiliary system based on EEG signal | |
| CN109009102B (en) | Electroencephalogram deep learning-based auxiliary diagnosis method and system | |
| CN104523266A (en) | Automatic classification method for electrocardiogram signals | |
| CN111759345B (en) | Heart valve abnormality analysis method, system and device based on convolutional neural network | |
| CN110638430B (en) | Method for building cascade neural network ECG signal arrhythmia classification model | |
| CN113095302B (en) | Depth model for arrhythmia classification, method and device using same | |
| CN111544256A (en) | Brain-controlled intelligent full limb rehabilitation method based on graph convolution and transfer learning | |
| CN111887858B (en) | Ballistocardiogram signal heart rate estimation method based on cross-modal mapping | |
| CN109907753B (en) | Multi-dimensional ECG signal intelligent diagnosis system | |
| CN103886184A (en) | Construction method for heart pathology recognition model | |
| CN111460951A (en) | Electrocardiosignal automatic analysis method based on deep learning | |
| CN112906748A (en) | 12-lead ECG arrhythmia detection classification model construction method based on residual error network | |
| CN108577831B (en) | Single-guide-core-paste data long-range monitoring and diagnosing system and processing method thereof | |
| Arvanaghi et al. | Classification cardiac beats using arterial blood pressure signal based on discrete wavelet transform and deep convolutional neural network | |
| Fan et al. | Le-lwtnet: A learnable lifting wavelet convolutional neural network for heart sound abnormality detection | |
| CN113116300A (en) | Physiological signal classification method based on model fusion | |
| CN108836316B (en) | Electrocardiosignal R wave extraction method based on BP neural network | |
| JP2023104885A (en) | Electrocardiographic heart rate multi-type prediction method based on graph convolution | |
| CN116269212A (en) | Multi-mode sleep stage prediction method based on deep learning | |
| Aziz et al. | Detection of Myocardial Infarction using Spectral Textures through ECG | |
| Rajesh et al. | A Silent Cardiac Atrial Fibrillation Detection and Classification using Deep Learning Approach | |
| Bengherbia et al. | Real-time smart system for ecg monitoring using a one-dimensional convolutional 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |