CN111667921A - Artificial intelligence ECG atrial fibrillation and arrhythmia detection system - Google Patents

Artificial intelligence ECG atrial fibrillation and arrhythmia detection system Download PDF

Info

Publication number
CN111667921A
CN111667921A CN202010516175.3A CN202010516175A CN111667921A CN 111667921 A CN111667921 A CN 111667921A CN 202010516175 A CN202010516175 A CN 202010516175A CN 111667921 A CN111667921 A CN 111667921A
Authority
CN
China
Prior art keywords
ecg
layer
atrial fibrillation
arrhythmia
detection
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
Application number
CN202010516175.3A
Other languages
Chinese (zh)
Inventor
顾正阳
于喜明
刘亚平
周琦
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Humeds Health Technologies Corp
Suzhou Mitai Artificial Intelligence Research Institute Co ltd
Suzhou Mite Xisaier Artificial Intelligence Co ltd
Original Assignee
Humeds Health Technologies Corp
Suzhou Mitai Artificial Intelligence Research Institute Co ltd
Suzhou Mite Xisaier Artificial Intelligence Co ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Humeds Health Technologies Corp, Suzhou Mitai Artificial Intelligence Research Institute Co ltd, Suzhou Mite Xisaier Artificial Intelligence Co ltd filed Critical Humeds Health Technologies Corp
Priority to CN202010516175.3A priority Critical patent/CN111667921A/en
Publication of CN111667921A publication Critical patent/CN111667921A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/80—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for detecting, monitoring or modelling epidemics or pandemics, e.g. flu
    • 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
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H10/00—ICT specially adapted for the handling or processing of patient-related medical or healthcare data
    • G16H10/60—ICT specially adapted for the handling or processing of patient-related medical or healthcare data for patient-specific data, e.g. for electronic patient records
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/30—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for calculating health indices; for individual health risk assessment
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H80/00—ICT specially adapted for facilitating communication between medical practitioners or patients, e.g. for collaborative diagnosis, therapy or health monitoring

Landscapes

  • Health & Medical Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Medical Informatics (AREA)
  • Public Health (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Pathology (AREA)
  • Artificial Intelligence (AREA)
  • Primary Health Care (AREA)
  • Epidemiology (AREA)
  • Physics & Mathematics (AREA)
  • Molecular Biology (AREA)
  • Veterinary Medicine (AREA)
  • Biophysics (AREA)
  • Surgery (AREA)
  • Animal Behavior & Ethology (AREA)
  • Signal Processing (AREA)
  • Psychiatry (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • Physiology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Computation (AREA)
  • Fuzzy Systems (AREA)
  • Mathematical Physics (AREA)
  • Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)

Abstract

The invention discloses an artificial intelligent ECG atrial fibrillation and arrhythmia detection system. The system comprises: the electrocardiogram detector, the smart phone, the mobile internet and the cloud automatic diagnosis platform. The electrocardiogram detector sends detected or monitored electrocardiosignals to the smart phone through Bluetooth or WiFi, the smart phone sends the ECG to the cloud automatic diagnosis platform through a mobile data network, and after automatic diagnosis is completed, a diagnosis report is sent back to a patient, family members of the patient and a designated doctor. The cloud automatic diagnosis platform comprises a server, a database, a user management system and artificial intelligence ECG atrial fibrillation and arrhythmia automatic diagnosis software.

Description

Artificial intelligence ECG atrial fibrillation and arrhythmia detection system
Technical Field
The invention relates to the field of artificial intelligence, Internet and mobile medical treatment, in particular to an artificial intelligence ECG atrial fibrillation and arrhythmia detection system.
Background
The latest data of the national cardiovascular disease center show that the cardiovascular disease becomes the first killer of the health of residents in China. Arrhythmia mainly caused by atrial fibrillation (short for atrial fibrillation) causes cardiac function deterioration and cerebral apoplexy, and is an important cause of death and disability of patients. According to the introduction of experts, cardiovascular disease patients are about 2.9 hundred million in China at present, the number of patients with atrial fibrillation is as high as 1000 ten thousand, and males are higher than females. Therefore, the task of examination, diagnosis and prevention of cardiovascular diseases and heart diseases is arduous. The Electrocardiogram (ECG) monitoring is the most convenient means for diagnosing heart diseases, the artificial intelligence analyzes the ECG, monitors the ECG in real time and is a simple and effective means for diagnosing and preventing heart diseases in time.
Volta Medical, french Medical, developed artificial intelligence software (AIFib) to help cardiologists detect Atrial Fibrillation (AF). Apple Heart application provided by Apple Watch (Apple Watch) achieves 84% of atrial fibrillation recognition accuracy. The international atrial fibrillation automatic identification competition held by PhysioNet has sensitivity and specificity respectively over 95 percent based on an artificial intelligence atrial fibrillation identification system of ECG. There are also large companies in the country developing various ECG-based atrial fibrillation recognizers.
Atrial fibrillation has three major hazards, including cerebral infarction (i.e., stroke), heart failure and decreased quality of life. In addition, patients with atrial fibrillation have twice the mortality rate of the normal population. Atrial fibrillation is a more dangerous condition in that atrial fibrillation can greatly increase the risk of thrombosis and cerebral infarction due to the loss of atrial contractile function and the increase of long-term heart rate, which can cause the enlargement of the heart and cardiac insufficiency. Atrial fibrillation is not only harmful but also has a high incidence rate. According to statistics, the overall prevalence rate of atrial fibrillation is about 0.7%, and the prevalence rate of people over 80 years old can be as high as 7.5%.
Electrocardiogram (ECG) is the gold standard for diagnosing atrial fibrillation, but since it can only be done in hospitals, patients are often diagnosed after life-threatening events such as strokes occur. However, many patients do not know their condition because of the lack of convenient and quick diagnostic means.
In order to solve the problems, the invention provides an artificial intelligence ECG atrial fibrillation and arrhythmia detection system, a patient uploads collected Electrocardiogram (ECG) data or pictures to a cloud platform (through a mobile phone APP), artificial intelligence ECG atrial fibrillation detection software deployed on the cloud platform automatically detects and analyzes the ECG of the patient, and then a diagnosis result is fed back to the patient, family members of the patient and a designated doctor. And the alarm is given in time under the critical condition of the patient.
The invention has the obvious benefits that the ECG on-line detection and monitoring report can be provided for the patient through the ECG detection and monitoring of the patient, the labor intensity of an electrocardiogram diagnostician is greatly reduced, and meanwhile, the rapid and accurate ECG on-line automatic detection and detection can be provided for the patient. Through atrial fibrillation monitoring, the life of a patient with a heart disease is saved in time, the risk of life quality reduction caused by cerebral infarction (namely common stroke), heart failure and the like is reduced, and the death rate is reduced. The large-scale popularization of the invention can save a large amount of manpower and material resources, cover wide crowds, and has important significance for improving the health of people and obtaining huge social and economic benefits.
Disclosure of Invention
The invention discloses an artificial intelligent ECG atrial fibrillation and arrhythmia detection system. The system comprises: the electrocardiogram detector, the smart phone, the mobile internet and the cloud automatic diagnosis platform. The electrocardiograph may be a portable ECG monitor to facilitate the patient's detection of ECG at any time and place or a dynamic ECG monitor (Holter) for continuous monitoring of the patient's ECG. The intelligent mobile phone is provided with an app and is used for being connected with the electrocardiogram detector through Bluetooth or WiFi to acquire ECG data of the intelligent mobile phone and transmitting the ECG data to the cloud automatic diagnosis platform through a mobile data network. The mobile internet is used for transmitting data to the cloud automatic diagnosis platform. The cloud automatic diagnosis platform comprises a server, a database, a user management system and artificial intelligent ECG atrial fibrillation and arrhythmia automatic diagnosis software.
According to an aspect of the invention, the ecg monitor is operated (manually or automatically) by the smartphone app to automatically detect or monitor the ecg of the patient. The detected electrocardiogram signals of the patient are transmitted to the smart phone through Bluetooth or WiFi, and then transmitted to the cloud automatic diagnosis platform through the smart phone. In the cloud automatic diagnosis platform, each patient registers an account, detected ECG data is stored in a database under the patient account according to detection time, the real-time detection data is sent to the artificial intelligent ECG atrial fibrillation automatic diagnosis software installed at a server terminal for automatic diagnosis and analysis, and a diagnosis report is formed. The diagnosis report and the diagnosis result can inform the patient, the family members of the patient and the appointed doctor in time by using WeChat and short message modes.
According to another aspect of the invention, besides the electrocardiograph acquisition, the electrocardiograph acquisition method can also directly photograph the electrocardiogram obtained by other means through the smartphone app. For example, the electrocardiogram results of hospital examinations can be directly photographed by a mobile phone. And the electrocardiogram photographed by the smartphone app is uploaded to a cloud automatic diagnosis platform through the mobile data network for analysis and diagnosis.
According to another aspect of the invention, the cloud automatic diagnosis platform is deployed on a public cloud, has enough storage space and cloud computing capacity, and is convenient to expand.
Preferably, the cloud platform provides registration, login management and user information management of the user.
Preferably, the cloud platform provides a sufficiently large data storage space for the user.
Preferably, the cloud platform can provide enough cloud computing power, run the artificial intelligence ECG atrial fibrillation automatic diagnosis software quickly and efficiently, and expand the expansion of the number of patients on line according to the needs.
According to another aspect of the invention, the automatic ECG diagnosis software automatically analyzes and diagnoses the ECG signal and is composed of an atrial fibrillation detector and an arrhythmia detector. First, the entire input patient ECG digital signal is evenly sliced into ECG segments of 10 second length. These 10 second ECG segments are then fed as input signals to the atrial fibrillation detector for classification diagnosis. The ECG segment, after passing through the atrial fibrillation detector, is automatically classified into 4 signals: atrial fibrillation, normal, other (non-atrial fibrillation), and noise. After the ECG segment is diagnosed as atrial fibrillation, normal or noise by the atrial fibrillation detector, the diagnosis is completed and the diagnosis result is output as atrial fibrillation, normal or noise. If the other time is diagnosed, the ECG segment signal is sent to an arrhythmia detector for arrhythmia diagnosis and classification.
According to a further aspect of the invention, the arrhythmia detector first cuts the ECG segment (10 second ECG segment) signal into a number of individual heartbeat beats (beats) based on the R-peak. Then, the arrhythmia detector formed by the one-dimensional convolutional neural network is used for carrying out classified diagnosis on the arrhythmia detector.
Preferably, the R peak is detected using the Pan-Tompkins algorithm.
Preferably, after the R peak is detected, the middle region [ TRn-1+ m, TRn +1-m ] of three consecutive R peaks is intercepted as the nth single-beat to be detected, where m is the set number of sampling points for adjusting the number of the parts including the repetition before and after the nth QRS complex is intercepted.
Preferably, the truncated cardiac rhythm is numbered according to the serial number of the ECG segment and the truncation serial number.
Preferably, each heartbeat beat is fed as an input signal to the arrhythmia detector for detection.
Preferably, the arrhythmia detector fully complies with the ANSI/AAMI: EC57 standard to classify each heartbeat beat as class 5: n, S, V, F, Q are provided. Wherein:
n: normal beats (including left/right bundle branch block);
s: supraventricular ectopic beats;
v: ventricular ectopic beats;
f: fusing the beats;
q: undetermined beats.
According to still another aspect of the present invention, the atrial fibrillation detector is used to diagnose whether a subject has atrial fibrillation, and classifies ECG diagnosis results of the patient into 4 categories: normal, atrial fibrillation, others, and noise. The atrial fibrillation detector firstly changes a time domain ECG segmented signal into a two-dimensional characteristic signal by using wavelet transformation. The features of the image are classified by a classifier formed by a convolutional neural network.
Preferably, the time domain ECG signal is first processed using a wavelet transform to facilitate analysis of the frequency components at a certain point on the time domain ECG signal. Features such as characterizing atrial fibrillation: the P wave disappears and replaces the F wave with a fast frequency, and the characteristic needs to perform time-frequency analysis on the signal at the P wave and extract frequency components to compare the difference of the normal frequency and the atrial fibrillation frequency at the P wave. And thus information can be efficiently extracted from the signal. The function or signal can be subjected to multi-scale detailed analysis through operation functions such as stretching and translation, and the like, so that a plurality of difficult problems which cannot be solved by Fourier transform are solved.
The wavelet transform is as follows:
Figure 595410DEST_PATH_IMAGE002
（1）
wherein
Figure 77338DEST_PATH_IMAGE004
Represents a scale factor of the image data to be displayed,
Figure 133019DEST_PATH_IMAGE006
which represents a factor in time, is,
Figure 700397DEST_PATH_IMAGE008
representing the wavelet basis functions. Scale factor
Figure 499726DEST_PATH_IMAGE010
Controlling the scaling, time factor, of wavelet basis functions
Figure 836161DEST_PATH_IMAGE012
The translation of the wavelet basis functions is controlled.
Preferably, a Morlet wavelet basis is used in the atrial fibrillation detector, and the Morlet wavelet basis is a single-frequency complex sine function under a Gaussian envelope:
Figure 62743DEST_PATH_IMAGE014
（2）
preferably, a non-uniform wavelet transform scale is used in the atrial fibrillation detector, which employs a denser scaling on the small scale, i.e., the high frequency scale, to fully "reveal" the characteristic that P-waves of atrial fibrillation disappear instead of F-waves of high frequency.
Preferably, the ECG segment signals are transformed into two-dimensional feature signals through wavelet transformation, and a two-dimensional convolutional neural network is used for classification, and any convolutional neural network classifier applied in two-dimensional image classification can be adopted. The structure of the system comprises a data Input layer/Input layer, a convolution calculation layer/CONV layer, a ReLU excitation layer/ReLU layer, a Pooling layer/Pooling layer and a full connection layer/FC layer.
Preferably, the convolutional neural network comprises a data Input layer/Input layer, at least one convolution calculation layer/CONV layer, ReLU excitation layer/ReLU layer, Pooling layer/Pooling layer, and fully-connected layer/FClayer.
Preferably, the convolutional neural network may include a plurality of convolutional calculation layers/CONV layer, ReLU excitation layer/ReLU layer, Pooling layer/Pooling layer, and fully-connected layer/FC layer.
Preferably, in the convolutional neural network, normalization may be performed at an input end, or normalization processing (local regularization) may be performed at other layers, and sample data is compressed to a [0,1] interval.
Preferably, in the convolutional neural network, the input data can be enhanced by horizontal and vertical flipping and random truncation.
Preferably, in each layer structure of the convolutional neural network, a dropping algorithm Dropout can be adopted, and some neurons are properly dropped (Dropout), so that the training speed is improved.
Preferably, the convolutional neural network constructs a supervised learning four-classifier for the atrial fibrillation detector through softmax.
According to yet another aspect of the invention, the arrhythmia detector is used to diagnose arrhythmia in a patient via an ECG signal. In the present invention, the arrhythmia detector fully complies with the ANSI/AAMI: EC57 standard to classify each heartbeat beat into 5 classes: n, S, V, F, Q are provided.
Preferably, the arrhythmia detector adopts a one-dimensional depth convolution neural network classifier. The input data of the convolutional neural network input layer is a heartbeat rhythm (one beat), namely only comprises a complete QRS complex. The intermediate layer includes at least one convolution calculation layer/CONV layer, ReLU excitation layer/ReLU layer, pooling layer/Poolinglayer, and a plurality of full connection layers/FC layers. The output is a 5 classifier constructed with Softmax for classification of the detected heart beat rhythm: n, S, V, F, Q are provided.
Preferably, said ECG segment signal, after each heart beat rhythm thereof is detected by the arrhythmia detector, marks the arrhythmia detection result of the heart beat rhythm at the QRS wave apex of the heart beat rhythm. After all heart beat rhythms of the ECG segment signal are detected and marked, the ECG segment signal is output as the electrocardiogram ECG for the final diagnosis.
The invention has the beneficial effects that:
the direct benefits of the invention are: the artificial intelligence ECG atrial fibrillation and arrhythmia detection system is provided, a patient can conveniently take a picture of an electrocardiogram through an electrocardiogram detector or by using an intelligent mobile phone app directly, and the electrocardiogram is uploaded to a cloud platform for electrocardiogram detection or monitoring. The detection system is simple, convenient, rapid and accurate to use, and the diagnosis result is fed back to the patient, the patient relative and the designated doctor through the mobile phone. By monitoring atrial fibrillation, the life of a patient with a heart disease can be saved in time, the risk of life quality reduction caused by cerebral infarction (namely common stroke), heart failure and the like is reduced, and the death rate is reduced. And the safe driving and protecting navigation is provided for the electrocardio health of the patient. The large-scale popularization of the invention can save a large amount of manpower and material resources, serve the public and obtain huge social and economic benefits.
Drawings
FIG. 1 is a schematic diagram of an artificial intelligence ECG atrial fibrillation and arrhythmia detection system architecture according to a preferred embodiment of the present invention;
FIG. 2 is a flow chart of artificial intelligence ECG atrial fibrillation and arrhythmia detection in accordance with a preferred embodiment of the present invention;
FIG. 3 is an atrial fibrillation detector architecture according to a preferred embodiment of the present invention;
FIG. 4 is a schematic diagram of an arrhythmia detector input beat rhythm intercept in accordance with a preferred embodiment of the present invention;
FIG. 5 is an arrhythmia ECG diagnostic label of a preferred embodiment of the present invention;
fig. 6 is an arrhythmia detector architecture of a preferred embodiment of the present invention.
Detailed Description
The invention will now be described in further detail with reference to the drawings and preferred embodiments. The drawings are simplified schematic diagrams illustrating the basic principles, basic structures and basic functions of the present invention in a schematic manner, and thus show only the constituents related to the present invention.
One of ordinary skill in the art may recognize certain variations and equivalents of the invention, which should not be construed as beyond the scope of the invention.
Fig. 1 is a schematic diagram of an architecture of an artificial intelligence ECG atrial fibrillation and arrhythmia detection system according to a preferred embodiment of the present invention, where 11 is a detector, 12 is an ECG report that ECG detection has been completed, 13 is a smartphone with app installed, 11 detects or monitors the ECG count of a patient and transmits the ECG count to 13 through bluetooth or WiFi, or 13 takes a photograph 12 directly to obtain the ECG count of the patient; the electrocardiogram data of the server 13 is sent to the databases 15, 15 of the cloud platform through the mobile internet, the artificial intelligence ECG atrial fibrillation and arrhythmia detection software in the server 16, 16 sends the electrocardiogram data to automatically analyze and diagnose the electrocardiogram sent by the server 15, and then the diagnosis result and report are given and returned to the server 13 through the server 14 to be displayed.
In the flow of artificial intelligence ECG atrial fibrillation and arrhythmia detection of the preferred embodiment of the present invention as shown in FIG. 2, 21 is the input ECG signal, 22 is entered to fen segmentation process for the ECG signal, 21 is input ECG signal is cut into ECG segmented signal of 10 seconds length; then the ECG segmented signals are sent to a 23 atrial fibrillation detector for detection, namely the ECG segmented signals are classified into atrial fibrillation, normal, noise and other four types, if the ECG segmented signal detection result is atrial fibrillation, normal or noise, the detection is finished, and 25 is result output; the ECG segmented signal is then sent 28 and a diagnostic report is given; if the classification result given at 23 is other, the ECG segment is sent to 24 arrhythmia detector for arrhythmia diagnosis; a classification diagnosis result 26 for each cardiac rhythm; the ECG segment signal is labeled with the diagnostic result of 26 (corresponding N, S, V, F, or Q, is labeled on top of the QRS complex of the cardiac rhythm being diagnosed).
As shown in FIG. 3, in the preferred embodiment of the atrial fibrillation detector architecture, the incoming ECG segment (10 second ECG) is wavelet transformed 301, where the wavelet transform is as follows:
Figure 101106DEST_PATH_IMAGE016
（3）
wherein
Figure DEST_PATH_IMAGE018
Represents a scale factor of the image data to be displayed,
Figure DEST_PATH_IMAGE020
which represents a factor in time, is,
Figure DEST_PATH_IMAGE022
representsWavelet basis functions. Scale factor
Figure 327294DEST_PATH_IMAGE018
Controlling the scaling, time factor, of wavelet basis functions
Figure 501924DEST_PATH_IMAGE020
The translation of the wavelet basis functions is controlled.
In the atrial fibrillation detector, a Morlet wavelet basis is used, which is a single-frequency complex sine function under a Gaussian envelope:
Figure DEST_PATH_IMAGE024
（4）
in addition, a non-uniform wavelet transform scale is used in the atrial fibrillation detector, which employs a denser scale transform on a small scale, i.e., a high frequency scale, to fully "reveal" the characteristic that P-waves of atrial fibrillation disappear instead of F-waves of high frequency. After wavelet transformation, processing the image into 3000x512 size; entering 302, the first convolutional layer of the ReseNet-34 network, with convolution kernel 7 x 7, convolution step 2, output size 1500 x 256, 64 profiles, which in this real-time example all include output Relu layers; 303 is the maximum pooling layer, the convolution kernel is 3 x 3, the convolution step is 2, the maximum value is taken, the output size is 750 x 128, and 64 feature graphs are obtained; 304 is the 2 nd convolutional layer, which is composed of 3 sets of residual networks, which are composed of 2 convolutional layers with convolution kernel 3 x 3 and convolution step size 1, and 64 feature maps, respectively, and the input of the first layer is directly added to the output of the second layer. After 304, the output size is 375 x 64, 64 characteristic graphs; 305 is the 3 rd convolutional layer, which is composed of 4 groups of residual error networks, the residual error networks are respectively composed of 2 convolutional kernels with 3 x 3, except the 1 st convolutional step of the first group of residual error networks with 2, the other convolutional steps with 1, 128 feature maps, and the input of the first layer is directly added with the output of the second layer. 305 has an output size of 188 x 32, 128 signatures; and 306 is a 4 th convolutional layer consisting of 6 groups of residual error networks, the residual error networks respectively consist of 2 convolutional kernels with 3 x 3, except that the 1 st convolutional step of the first group of residual error networks is 2, the other convolutional layers with 1 convolutional step and 256 feature maps, and the input of the first layer is directly added with the output of the second layer. 306 has an output size of 94 x 16, 256 signatures; 307 is the 5 th convolutional layer, which is composed of 3 groups of residual error networks, the residual error networks are respectively composed of 2 convolutional kernels with 3 x 3, except that the 1 st convolutional step of the first group of residual error networks is 2, the other convolutional steps are 1, 512 convolutional layers of characteristic diagrams, and the input of the first layer is directly added with the output of the second layer. 307 output sizes of 47 x 8, 512 feature maps; 308 is an average pooling layer, that is, an average output value is obtained after global average pooling is performed on each feature map (feature map) of 307, so that the number of nodes of the full connection layer is greatly reduced; 309 is a fully-connected layer, connecting 310 the corresponding node to a classifier (softmax) by at least one layer of fully-connected layer and Dropout processing, outputting by the classifier 4 classes of results: atrial fibrillation, normal, noisy, and others.
The arrhythmia detector input beat rhythm interception scheme of the preferred embodiment of the present invention as shown in fig. 4 first labels each R peak 401 of the ECG segment signal intercepted to 10 seconds by the Pan-Tompkins algorithm as shown in fig. 4. Intercepting the nth cardiac rhythm, firstly finding n-1, n and n +1, 3 continuous cardiac rhythms, setting m sampling points, wherein the value of m is adjustable, and the purpose is to ensure that the interception interval between 402 and 403 can completely contain the information of the nth cardiac rhythm. Here, it does not matter that some information of n-1 and n +1 is contained. Finally, the nth cardiac rhythm is truncated between sample points 402 and 403. And so on, starting with n =1 until the 10 second ECG segment is truncated. And in the intercepting process, numbering the intercepted electrocardio-rhythm according to the segment serial number and the intercepting serial number.
The arrhythmia ECG diagnostic label shown in fig. 5 is labeled N when the cardiac rhythm is identified as normal, S, V, F or Q respectively for abnormal cardiac rhythms, and v (ventricular) for ectopic ventricular beats as shown.
An arrhythmia detector architecture of a preferred embodiment of the present invention is shown in fig. 6. When the detection result of the atrial fibrillation detector on the ECG subsection is other, the ECG subsection is sent to the arrhythmia detector for arrhythmia classification detection, 601 is the detected ECG subsection, 602 detects and marks an R peak of the ECG subsection by using a Pan-Tompkins algorithm; 603 finishing single cardiac rhythm interception, wherein a specific algorithm is shown in figure 4; sending the single cardiac rhythm into 604, and a first layer of one-dimensional convolutional neural network; 605 is a batch normalization layer, which performs batch normalization processing to keep the input of each layer of neural network in the same distribution in the deep neural network training process; 606 is a pooling layer; 607 is a second layer convolutional neural network layer; 608 is a BatchNormalization layer; 609 is a pooling layer; 610 is a fully connected layer; 611 is the softmax layer, here a 5 classifier, with the output: n, S, V, F, or Q; reference numeral 612 is an ECG segment labeling and output, wherein the output pattern of the labeled ECG segment is shown in FIG. 5.
In light of the above-described embodiments of the present invention, it is clear that many modifications and variations can be made by the worker skilled in the art without departing from the scope of the present invention. The technical scope of the present invention is not limited to the content of the specification, and must be determined according to the scope of the claims.

Claims (8)

1. An artificial intelligence ECG atrial fibrillation and arrhythmia detection system, comprising: the electrocardiogram detector, the smart phone, the mobile internet and the cloud automatic diagnosis platform.
2. The system according to claim 1, wherein the smart phone is installed with an app to control the ECG monitoring device to obtain ECG monitoring data through bluetooth or WiFi, or directly take a photograph of the ECG to obtain ECG monitoring data, and upload the obtained ECG data to the cloud automated diagnosis platform through the mobile internet.
3. The system of claim 1, wherein the cloud automated diagnostic platform comprises a database and a server: the database is used for storing patient information and electrocardiogram data, and the server is used for running artificial intelligence ECG atrial fibrillation and arrhythmia detection software to finish atrial fibrillation detection and arrhythmia detection of ECG signals.
4. The artificial intelligence ECG atrial fibrillation and arrhythmia detection system of claim 1, wherein the detected ECG is processed by ECG segmentation and truncation according to a length of 10 seconds; the ECG segment includes a plurality of cardiac rhythms.
5. The system of claim 1, wherein the ECG segment is first subjected to atrial fibrillation detection by the atrial fibrillation detector, and if the detection result is atrial fibrillation, normal or noise, the ECG segment detection is terminated, and if the detection result is other, the arrhythmia detector is turned on, and then arrhythmia detection is performed on the ECG segment.
6. The system of claim 1, wherein the one-dimensional ECG segment signals inputted from the atrial fibrillation detector are wavelet-transformed into two-dimensional signals, the two-dimensional signals are then subjected to atrial fibrillation detection by a classifier composed of a two-dimensional deep convolutional neural network, and the ECG segment signals required for atrial fibrillation detection are inputted into the classifier and classified and outputted as atrial fibrillation, normal, noise or other four results.
7. The artificial intelligence ECG atrial fibrillation and arrhythmia detection system of claim 1, wherein the two-dimensional depth convolutional neural network is composed of an input layer, a hidden layer, and an output layer; the input layer is an image with a fixed size; the hidden layer comprises at least one standard layer, namely: convolutional layer, activation function, BachNormalization layer and pooling layer; the output layer consists of an average pooling layer, a full-link layer and a softmax layer.
8. The artificial intelligence ECG atrial fibrillation and arrhythmia detection system of claim 1, wherein the arrhythmia detector processes a single one-dimensional cardiac rhythm signal using a one-dimensional depth convolution neural network; the one-dimensional depth convolution neural network comprises a data input layer, at least one hidden layer and an output layer; the hidden layer consists of a convolutional layer, an activation function, a BachNormalization layer and a pooling layer; the output layer consists of an average pooling layer, a full-link layer and a softmax layer, and the output is classified according to the international standard 5 for arrhythmia: n, S, V, F, Q are provided.
CN202010516175.3A 2020-06-09 2020-06-09 Artificial intelligence ECG atrial fibrillation and arrhythmia detection system Pending CN111667921A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010516175.3A CN111667921A (en) 2020-06-09 2020-06-09 Artificial intelligence ECG atrial fibrillation and arrhythmia detection system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010516175.3A CN111667921A (en) 2020-06-09 2020-06-09 Artificial intelligence ECG atrial fibrillation and arrhythmia detection system

Publications (1)

Publication Number Publication Date
CN111667921A true CN111667921A (en) 2020-09-15

Family

ID=72387103

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010516175.3A Pending CN111667921A (en) 2020-06-09 2020-06-09 Artificial intelligence ECG atrial fibrillation and arrhythmia detection system

Country Status (1)

Country Link
CN (1) CN111667921A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113229825A (en) * 2021-06-22 2021-08-10 郑州大学 Deep neural network-based multi-label multi-lead electrocardiogram classification method
CN113440149A (en) * 2021-07-12 2021-09-28 齐鲁工业大学 ECG signal classification method based on twelve-lead electrocardiogram data two-dimensional multi-input residual error neural network
CN114469133A (en) * 2021-12-14 2022-05-13 中国科学院深圳先进技术研究院 Undisturbed atrial fibrillation monitoring method

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113229825A (en) * 2021-06-22 2021-08-10 郑州大学 Deep neural network-based multi-label multi-lead electrocardiogram classification method
CN113440149A (en) * 2021-07-12 2021-09-28 齐鲁工业大学 ECG signal classification method based on twelve-lead electrocardiogram data two-dimensional multi-input residual error neural network
CN113440149B (en) * 2021-07-12 2023-09-29 齐鲁工业大学 ECG signal classification method based on twelve-lead electrocardiograph data two-dimensional multi-input residual neural network
CN114469133A (en) * 2021-12-14 2022-05-13 中国科学院深圳先进技术研究院 Undisturbed atrial fibrillation monitoring method
CN114469133B (en) * 2021-12-14 2023-10-03 中国科学院深圳先进技术研究院 Undisturbed atrial fibrillation monitoring method

Similar Documents

Publication Publication Date Title
EP3801238B1 (en) Detecting abnormalities in ecg signals
CN110840402B (en) Atrial fibrillation signal identification method and system based on machine learning
Murugesan et al. Ecgnet: Deep network for arrhythmia classification
EP3803709B1 (en) Detecting abnormalities in ecg signals
Mousavi et al. ECGNET: Learning where to attend for detection of atrial fibrillation with deep visual attention
EP3698707B1 (en) Electrocardiogram information dynamic monitoring system, computer program and computer readable storage medium
US11324455B2 (en) Artificial intelligence-based interference recognition method for electrocardiogram
US20230143594A1 (en) Systems and methods for reduced lead electrocardiogram diagnosis using deep neural networks and rule-based systems
CN111667921A (en) Artificial intelligence ECG atrial fibrillation and arrhythmia detection system
WO2019161610A1 (en) Electrocardiogram information processing method and electrocardiogram workstation system
Raj et al. A personalized point-of-care platform for real-time ECG monitoring
CN113095302B (en) Depth model for arrhythmia classification, method and device using same
Jin et al. Classification of normal and abnormal ECG records using lead convolutional neural network and rule inference
CN110638430A (en) Multi-task cascade neural network ECG signal arrhythmia disease classification model and method
Khan et al. Electrocardiogram heartbeat classification using convolutional neural networks for the detection of cardiac Arrhythmia
EP4041073A1 (en) Systems and methods for electrocardiogram diagnosis using deep neural networks and rule-based systems
CN115062655A (en) Multi-lead electrocardiosignal analysis method and multi-lead myocardial infarction analysis system
Sakli et al. An efficient deep learning model to predict cardiovascular disease based on ECG signal
KR20200125256A (en) Apparatus and method for analyzing electrocardiogram
Oliveira et al. A novel arrhythmia classification method based on convolutional neural networks interpretation of electrocardiogram images
CN110537907B (en) Electrocardiosignal compression and identification method based on singular value decomposition
He et al. Automatic classification of arrhythmias by residual network and BiGRU with attention mechanism
Van Zaen et al. Atrial fibrillation detection from PPG interbeat intervals via a recurrent neural network
Yin et al. An algorithm for locating pvc and spb in wearable ecgs
Rajesh et al. A Silent Cardiac Atrial Fibrillation Detection and Classification using Deep Learning Approach

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