CN115607166B - Intelligent electrocardiosignal analysis method and system and intelligent electrocardio auxiliary system - Google Patents
Intelligent electrocardiosignal analysis method and system and intelligent electrocardio auxiliary system Download PDFInfo
- Publication number
- CN115607166B CN115607166B CN202211220314.3A CN202211220314A CN115607166B CN 115607166 B CN115607166 B CN 115607166B CN 202211220314 A CN202211220314 A CN 202211220314A CN 115607166 B CN115607166 B CN 115607166B
- Authority
- CN
- China
- Prior art keywords
- electrocardiosignal
- electrocardio
- model
- critical value
- abnormality
- 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
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]
- A61B5/346—Analysis of electrocardiograms
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/316—Modalities, i.e. specific diagnostic methods
- A61B5/318—Heart-related electrical modalities, e.g. electrocardiography [ECG]
- A61B5/346—Analysis of electrocardiograms
- A61B5/349—Detecting specific parameters of the electrocardiograph cycle
-
- 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
- A61B5/7271—Specific aspects of physiological measurement analysis
- A61B5/7275—Determining trends in physiological measurement data; Predicting development of a medical condition based on physiological measurements, e.g. determining a risk factor
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Abstract
The application discloses an electrocardiosignal intelligent analysis method and system and an intelligent electrocardiosignal auxiliary system, wherein the method comprises the following steps: acquiring a plurality of electrocardiosignals, other information of a patient related to the electrocardiosignals, an electrocardiosignal abnormality label and an electrocardiosignal critical value label, and constructing a sample data set; dividing a sample data set into a training set, a verification set and a test set, constructing a training electrocardio abnormality recognition model and an electrocardio critical value recognition model based on the training set, inputting an electrocardio signal to be analyzed into the trained electrocardio abnormality recognition model to obtain an electrocardio abnormality prediction probability, inputting the electrocardio signal and other information of a patient related to the electrocardio signal into the electrocardio critical value recognition model when the electrocardio abnormality prediction probability is higher than a preset threshold, and outputting the electrocardio critical value prediction probability. The method and the system can give out the prediction probability by using the electrocardiosignal and other information of the patient related to the electrocardiosignal, and improve the accuracy of the identification of the electrocardiosignal abnormality and the identification of the electrocardiosignal critical value.
Description
Technical Field
The application relates to the technical field of artificial intelligence, in particular to an electrocardiosignal intelligent analysis method and system and an intelligent electrocardiosignal auxiliary system.
Background
The electrocardiosignal records the change of the electrical activity of the heart, is one of the most commonly used clinical examinations, and can help prompt the possibility of heart abnormality such as arrhythmia, myocardial ischemia, heart enlargement, hypertrophy and the like by analyzing the electrocardiosignal. Electrocardiogram critical values refer to the performance of an electrocardiogram recorded potentially life threatening and leading to severe hemodynamic abnormalities. The identification of the critical value of the electrocardio is particularly important in clinic, and the early identification and timely treatment can effectively reduce the occurrence of serious cardiovascular events such as sudden death and the like.
At present, the identification of the electrocardiosignal abnormality and critical value recorded in the electrocardiosignal requires the manual judgment of an electrocardiosignal interpretation doctor, and the data volume of the electrocardiographic examination of China is huge every year, so that the demand for the electrocardiosignal analysis is very large and is difficult to meet only by manual work. In addition, the identification and judgment of the abnormal electrocardiosignal and critical value conditions are directly affected by the technical level of doctors, but the current high-level doctors lack of resources, particularly in basic medical institutions in China, which further limit the identification and early warning of the abnormal electrocardiosignals and critical values.
In recent years, deep learning has remarkable development in the field of electrocardiographic anomaly recognition and higher accuracy than that of a traditional algorithm, but most of leading edge research results belong to abroad, the domestic research progress is relatively backward, particularly, the electrocardiographic critical value recognition and early warning are carried out, the traditional intelligent recognition method generally only analyzes electrocardiographic signals or measured values, comprehensive and effective utilization of other information of patients such as past history, medicine taking history and the like is relatively lacking, and the problem of insufficient recognition model accuracy exists.
Disclosure of Invention
The application provides an electrocardiosignal intelligent analysis method which can comprehensively and effectively utilize electrocardiosignals and other information of a patient related to the electrocardiosignals to carry out electrocardiosignal intelligent analysis to obtain electrocardiosignal abnormal prediction probability and electrocardiosignal critical value prediction probability and solve the problems of insufficient effective utilization of other information of the patient and insufficient model precision in the existing intelligent identification method.
The technical scheme also provides an electrocardiosignal intelligent analysis system which can realize automatic acquisition and utilize electrocardiosignals and other information of patients related to the patients to carry out electrocardiosignal intelligent analysis to obtain electrocardiosignal abnormal probability and electrocardiosignal critical value prediction probability and solve the problem of insufficient precision of the existing recognition model.
The technical scheme also provides an intelligent electrocardio auxiliary system which solves the problems of low automation and intelligent degree of the existing system.
In order to achieve the above purpose, the specific technical scheme is as follows:
an electrocardiosignal intelligent analysis method comprises the following steps:
acquiring a plurality of electrocardiosignals, other information of a patient related to the electrocardiosignals, an electrocardiosignal abnormality label and an electrocardio critical value label, and constructing a sample data set D;
dividing the sample data set D into a first training setD train,1 Verification setD vali And test setD test ；
Constructing an electrocardiographic abnormality recognition model framework to form a first training setD train,1 Training to obtain an electrocardiographic abnormality recognition model for input;
constructing a second training set for training an electrocardiographic critical value recognition modelD train,2 ；
Constructing an electrocardio critical value recognition model frame, and inputting a second training setD train,2 Training to obtain electrocardio critical value identificationA model;
inputting the electrocardiosignal to be analyzed into a trained electrocardiosignal abnormality recognition model to obtain an electrocardiosignal abnormality prediction probability, inputting the electrocardiosignal to be analyzed and other information of a patient associated with the electrocardiosignal to be analyzed into the trained electrocardiosignal critical value recognition model when the electrocardiosignal abnormality prediction probability is higher than an electrocardiosignal abnormality probability preset threshold value, and outputting the electrocardiosignal critical value prediction probability of the electrocardiosignal to be analyzed.
Preferably, the electrocardiographic abnormality label represents whether the electrocardiographic signal is abnormal or not; the electrocardio critical value label represents whether critical value exists in the electrocardio signals or not; the electrocardiographic abnormality identification model frame is a deep neural network frame; the electrocardio critical value identification model framework comprises a depth feature extractor, a manual feature extractor and a matrix regression model.
Preferably, the training of the electrocardiographic critical value recognition model comprises the following steps:
second training setD train,2 The central electrocardiosignal is input into a depth feature extractor to obtain an electrocardiosignal depth feature matrix F;
second training setD train,2 Inputting other information of a patient associated with the electrocardiosignal into a manual feature extractor to obtain a manual feature vector Z;
with electrocardiosignal depth characteristic matrix F, manual characteristic vector Z and second training setD train,2 And (3) taking the central electrocardio critical value label as input, and training a matrix regression model.
Preferably, the second training setD train,2 The construction of (2) comprises the following steps:
first, a first training set is extractedD train,1 In which the samples with the electrocardiographic anomalies form a first training subsetD train,2，1 ；
Second, the first training setD train,1 Inputting the electrocardiosignals into a trained electrocardiosignal abnormality recognition model to obtain the prediction probability of whether each electrocardiosignal is abnormal or not, and carrying out a first training setD train,1 The samples in the training sequence are sampled in layers according to the prediction probability to obtain a second training subsetD train,2,2 ；
Finally, the first training subsetD train,2,1 And a second training subsetD train,2,2 Obtaining a second training set for training the electrocardio critical value recognition model after taking the union setD train,2 。
Preferably, the matrix regression model is trained to obtain the prediction probability Y of the electrocardiographic critical value, wherein y=sigmoid (< F, B > + < β, Z >), wherein Sigmoid (< j >) represents Sigmoid transformation, F and Z represent the depth feature matrix and the manual feature vector of the electrocardiographic signal respectively, the matrix B and the vector β are both learnable parameters, < F, B > represents the inner product of F and B, and < β, Z > represents the inner product of β and Z.
Preferably, the matrix B and the vector beta are obtained by updating the objective function l (B, beta) + in the training process of the matrix regression model𝜆 1 ||B|| * + 𝜆 2 ||β|| 1 Obtained by the method, wherein𝜆 1 And𝜆 2 is a super parameter.
Preferably, the deep neural network framework comprises a deep feature extractor framework, and further comprises 1 Reshape layer, 1 Dropout layer and 1 FC layer with a Sigmoid activation function, which are sequentially connected behind the deep feature extractor framework.
Preferably, the depth feature extractor frame comprises 1 Conv module and 3 Res-SE modules; the Conv module is used for inputting electrocardiosignals to obtain characteristics of the middle layer, after the Conv module, 3 Res-SE modules are sequentially connected in series, the output of the former module is used as the input of the next module, and the last Res-SE module is used for outputting an electrocardiosignal depth characteristic matrix F.
An electrocardiosignal intelligent analysis system comprises a model generation module and a service calculation module,
the model generation module is used for acquiring a sample data set, constructing a model frame and completing model training to obtain a trained electrocardio abnormality identification model and an electrocardio critical value identification model;
the service calculation module is used for receiving a new electrocardiosignal analysis request, automatically collecting electrocardiosignal data corresponding to the request and other information of an associated patient, automatically calling a trained model, and acquiring and storing electrocardiosignal abnormality prediction probability and electrocardiosignal critical value prediction probability corresponding to the request.
Preferably, the model generating module comprises a data acquisition engine, a sample library, a model training engine and a model library, wherein the data acquisition engine is used for acquiring electrocardiosignals, other information of a patient related to the electrocardiosignals, an electrocardiosignal abnormality label and an electrocardiosignal critical value label and generating a sample data set required by model training; the sample library is used for storing a sample data set; the model training engine is based on a sample data set stored in a sample library, and model training is completed; the model library is used for storing a trained electrocardiographic anomaly recognition model and an electrocardiographic critical value recognition model;
the service computing module comprises a service triggering engine, a data acquisition engine and a model computing engine; the service triggering engine is used for receiving a new electrocardiosignal analysis request and sending the new electrocardiosignal analysis request to the data acquisition engine; the data acquisition engine is used for automatically acquiring electrocardiosignal data corresponding to a new electrocardiosignal analysis request and other information of an associated patient and sending the electrocardiosignal data and other information of the associated patient to the model calculation engine; the model calculation engine is used for calling the trained electrocardio abnormality recognition model and the electrocardio critical value recognition model, obtaining electrocardio abnormality prediction probability and electrocardio critical value prediction probability corresponding to the request, and finishing prediction probability storage.
Preferably, the model calculation engine comprises an electrocardio abnormality calculation sub-engine and an electrocardio critical value calculation sub-engine; the electrocardio abnormality calculation sub-engine is used for calling a trained electrocardio abnormality identification model to obtain electrocardio abnormality prediction probability; the electrocardio critical value calculation sub-engine is used for calling a trained electrocardio critical value identification model to obtain electrocardio critical value prediction probability.
Preferably, the electrocardiosignal intelligent analysis system further comprises a front-end interaction module and a dynamic monitoring module;
the front-end interaction module comprises a prediction result presentation unit and an artificial tag storage unit; the prediction result presentation unit is used for visually prompting a user based on the model prediction result obtained by the service calculation module, wherein the prediction result comprises an electrocardiographic abnormality prediction probability and an electrocardiographic critical value prediction probability; the manual label storage unit is used for comprehensively judging and forming an electrocardiosignal final manual interpretation label corresponding to the request by referring to prompt contents and combining professional knowledge for an electrocardiograph interpretation technician, wherein the manual interpretation label comprises an electrocardiograph abnormal label and an electrocardiograph critical value label;
the dynamic monitoring module comprises a service monitoring evaluation unit and a service update triggering engine; the service monitoring and evaluating unit is used for evaluating the model prediction effect in real time based on the model prediction result generated in the automatic accumulation application process and the manual interpretation label data; and the service update triggering engine is used for automatically triggering the update of the model and the service when the model effect does not meet the preset condition, and realizing the dynamic optimization update of the model.
The intelligent electrocardio-assisted system comprises an electrocardio-signal intelligent analysis system, a data processing module, a knowledge module and a prompting module;
the data processing module is used for processing the electrocardio abnormality prediction probability and the electrocardio critical value prediction probability to obtain an electrocardio abnormality grade and an electrocardio critical value grade;
the knowledge module is used for storing the processing knowledge;
and the prompting module is used for calling the knowledge module to obtain the processing knowledge associated with the electrocardiosignal corresponding to the request and prompting a clinician after the data processing module gives the electrocardiosignal abnormal grade and the electrocardiosignal critical value grade.
An electronic device, comprising: a processor;
and the memory is used for storing a program, and the program is configured to realize the electrocardiosignal intelligent analysis method when being executed by the processor.
A non-transitory computer readable storage medium storing instructions that, when executed by a processor, cause the processor to perform the method of intelligent analysis of electrocardiographic signals.
The electrocardiosignal intelligent analysis method and system can comprehensively utilize the electrocardiosignal and other information of a patient related to the electrocardiosignal to give out prediction probability to the electrocardiosignal abnormality and the electrocardiosignal critical value existing in the electrocardiosignal, and improve the accuracy of electrocardiosignal abnormality identification and electrocardiosignal critical value identification.
Drawings
Fig. 1 is a flowchart of an intelligent analysis method for electrocardiosignals provided in embodiment 1 of the application;
FIG. 2 is a schematic diagram of an electrocardiographic critical value recognition model according to embodiment 1 of the present application;
FIG. 3 is a schematic diagram of a deep neural network model according to embodiment 1 of the present application;
FIG. 4 is a diagram of Res-SE module frames in the deep neural network model provided in embodiment 1 of the present application;
fig. 5 is a schematic diagram of an intelligent analysis system for electrocardiograph signals according to embodiment 1 of the present application;
fig. 6 is a schematic diagram of an intelligent analysis system for electrocardiograph signals according to embodiment 2 of the present application;
fig. 7 is a schematic diagram of an intelligent electrocardiograph auxiliary system provided in embodiment 2 of the present application;
fig. 8 is a schematic diagram of an electronic device according to embodiment 2 of the present application.
Detailed Description
The present application will be described in further detail with reference to the accompanying drawings, in order to make the objects, technical means and advantages of the present application more apparent. The following description of the embodiments of the present application will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present application, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.
The terms "comprising" and "having" and any variations thereof in the description and claims of the application and in the foregoing drawings are intended to cover non-exclusive inclusions. For example, a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those elements but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
The technical scheme of the application is described in detail below by specific examples. The following embodiments may be combined with each other, and some embodiments may not be repeated for the same or similar concepts or processes.
Example 1
As shown in fig. 1, fig. 1 is a flowchart of an intelligent analysis method for an electrocardiograph signal according to embodiment 1 of the present application, which includes the following steps:
step 101, acquiring a plurality of electrocardiosignals, other information of a patient related to the electrocardiosignals, an electrocardiosignal abnormality label and an electrocardiosignal critical value label, and constructing a sample data set D;
the electrocardiosignals collected in the embodiment are 12-lead electrocardiosignals, the sampling time length is 10s, and the sampling rate is 500Hz;
other information of the patient associated with the electrocardiograph signals, including but not limited to age, gender, clinical diagnosis, whether drugs possibly causing electrocardiographic changes are taken, whether electrolyte disorders exist, etc.; other information of the patient collected by the embodiment of the application comprises gender, age, current medical history, past operation history and medication condition; extendably, electrocardiographic measurements may also be obtained as auxiliary information including, but not limited to, ventricular rate, atrial rate, heart rate, P-wave width, P-wave electrical axis, T-wave width, T-wave electrical axis, PR interval, QRS interval, QT interval, QTc interval, etc.; the cardiac electrical measurement values collected by the embodiment of the application are 18 in total and comprise ventricular rate, atrial rate, heart rate, P wave width, P wave electric axis, T wave width, T wave electric axis, R wave electric axis, PR interval, QRS interval, QT interval, QTc interval, RV1, RV3, RV5, SV1, SV3 and SV5;
step 102, dividing the sample data set D into a first training setD train,1 Verification setD vali And test setD test ；
The dividing ratio of the sample data set D can be set as required, in this embodiment 80% is the first training setD train,1 10% is the validation setD vali 10% is the test setD test ；
Step 103, constructing an electrocardiographic abnormality recognition model framework by using a first training setD train,1 Training to obtain an electrocardiographic abnormality recognition model for input;
104, constructing a second training set for training the electrocardio critical value recognition modelD train,2 The second training setD train,2 Based on the first training setD train,1 Obtaining, wherein the number of samples is smaller than the first training setD train,1 Number of samples;
step 105, constructing an electrocardio critical value recognition model frame, and inputting a second training setD train,2 Training to obtain an electrocardio critical value recognition model;
and 106, inputting the electrocardiosignal to be analyzed into a trained electrocardiosignal abnormality recognition model to obtain an electrocardiosignal abnormality prediction probability, inputting the electrocardiosignal to be analyzed and other information of a patient associated with the electrocardiosignal to be analyzed into the trained electrocardiosignal critical value recognition model when the electrocardiosignal abnormality prediction probability is higher than an electrocardiosignal abnormality probability preset threshold value, and outputting the electrocardiosignal critical value prediction probability of the electrocardiosignal to be analyzed.
The electrocardio abnormality probability preset threshold value can be set manually or can be obtained by automatic calculation according to a certain rule; in the embodiment, the electrocardio abnormality probability preset threshold is a super parameter, and is obtained by obtaining the best F1 score on the verification set when training an electrocardio abnormality identification model; in the method, in order to obtain a better electrocardio abnormality recognition model and an electrocardio critical value recognition model, the trained electrocardio abnormality recognition model and the electrocardio critical value recognition model can be subjected to model super-parameter adjustmentPerforming optimization and model effect evaluation; in this embodiment, a verification set is usedD vali Performing optimal model super-parameter selection, and evaluating model effect as using test setD test Performing effect evaluation on the trained model; model super parameters include, but are not limited to, learning rate, weight decay rate, batch size, etc., model super parameter tuning methods include, but are not limited to, grid parameters, model effect evaluation indexes include, but are not limited to, accuracy rate, AUC, recall rate, accuracy rate, F1 score, etc.
In actual clinical work, the identification of the electrocardiographic abnormality and the judgment of critical value need to integrate multi-source information, and only depend on the condition that electrocardiographic signals or measured values are difficult to accurately judge, for example, for the condition within a certain range of critical values, the judgment is often needed by combining comprehensive factors such as gender, age and the like of a patient, and for example, the administration of certain medicines can lead to the change of electrocardiographic signals in a certain period and degree; the method integrates other information of the patient on the basis of the electrocardiosignal data, wherein the other information of the patient is the patient condition which must be known by an electrocardiosignal technician when judging an electrocardiogram, and the information has specific practical clinical significance on the analysis of the electrocardiosignal, so that a prediction model of the information is integrated, and the method has the potential of obtaining more accurate electrocardiosignal abnormality prediction probability and electrocardio critical value prediction probability.
The electrocardiosignal abnormality label represents whether the electrocardiosignal is abnormal or not; the electrocardio critical value label represents whether critical value exists in the electrocardio signals or not; the electrocardiographic abnormality identification model frame is a deep neural network frame; as shown in fig. 2, the electrocardiographic critical value identification model framework comprises a depth feature extractor, a manual feature extractor and a matrix regression model; the depth feature extractor is obtained when training an electrocardiographic abnormality recognition model.
The electrocardiosignal abnormal label is a label which is obtained by manually judging whether the electrocardiosignals are abnormal or not by an electrocardiosignal interpretation technician; the electrocardiographic abnormalities include a variety of common electrocardiographic abnormality categories including, but not limited to, sinus bradycardia, sinus tachycardia, atrial fibrillation, ventricular premature beat, atrial premature beat, left bundle branch block, right bundle branch block, etc.;
the electrocardio critical value label is characterized in that an electrocardio interpretation technician judges whether an electrocardio critical value exists in each electrocardio signal and gives a label; the electrocardio critical values comprise 4 major electrocardio critical values, suspected acute coronary syndrome, severe tachyarrhythmia, severe slow arrhythmia and others; each electrocardiosignal is marked by an electrocardio interpretation technician through an electrocardio abnormal label and an electrocardio critical value label.
As shown in fig. 2, the electrocardiographic critical value identification model framework comprises a depth feature extractor, a manual feature extractor and a matrix regression model, wherein the depth feature extractor is used for extracting a depth feature matrix of an electrocardiographic signal, the manual feature extractor is used for extracting manual feature vectors of other information of a patient associated with the electrocardiographic signal, and the matrix regression model is used for receiving the depth feature and the manual feature of the electrocardiographic signal and outputting electrocardiographic critical value prediction probability.
As shown in fig. 2, the training of the electrocardiographic critical value identification model includes the following steps:
second training setD train,2 The central electrocardiosignal is input into a depth feature extractor to obtain an electrocardiosignal depth feature matrix F;
second training setD train,2 Inputting other information of a patient associated with the electrocardiosignal into a manual feature extractor to obtain a manual feature vector Z;
with electrocardiosignal depth characteristic matrix F, manual characteristic vector Z and second training setD train,2 And (3) taking the central electrocardio critical value label as input, and training a matrix regression model.
The second training setD train,2 The construction of (2) comprises the following steps:
first, a first training set is extractedD train,1 In which the samples with the electrocardiographic anomalies form a first training subsetD train,2,1 ；
Second, first trainingCollection setD train,1 Inputting the electrocardiosignals into a trained electrocardiosignal abnormality recognition model to obtain the prediction probability of whether each electrocardiosignal is abnormal or not, and carrying out a first training setD train,1 The samples in the training sequence are sampled in layers according to the prediction probability to obtain a second training subsetD train,2,2 ；
Finally, the first training subsetD train,2,1 And a second training subsetD train,2,2 Obtaining a second training set for training the electrocardio critical value recognition model after taking the union setD train,2 。
The first training set is extractedD train,1 In the sample with electrocardio abnormality, refers to the first training setD train,1 In which the electrocardiosignal marked as being in the presence of an electrocardiosignal abnormality is extracted to form a first training subsetD train,2,1 The method comprises the steps of carrying out a first treatment on the surface of the The first training setD train,1 The step of hierarchically sampling the samples in the training set according to the prediction probability means that the first training setD train,1 The electrocardiosignal in the heart is input into a trained electrocardiosignal abnormality recognition model, and a first training set can be obtainedD train,1 Predicting probability of whether the electrocardiosignal in the heart is abnormal or not, and the value range of the predicting probability is [0,1 ]]The larger the prediction probability is, the higher the possibility that the electrocardiograph has abnormality is; for the first training set according to the prediction probability of the electrocardiographic abnormalityD train,1 Grouping to obtain m subgroups, sampling each subgroup to obtain a second training subsetD train,2,2 The method comprises the steps of carrying out a first treatment on the surface of the In this embodiment, the training set is divided into 10 subgroups, m=10, each subgroup is simply and randomly sampled, and the sampled samples form a second training subsetD train,2,2
Finally, the first training subsetD train,2,1 And a second training subsetD train,2,2 Obtaining a second training set for training the electrocardio critical value recognition model after taking the union setD train,2 . The method is used for training the electrocardio crisisThe training set of the value recognition model can cope with the situation that the sample size cannot be too large when the matrix regression model is trained, and can effectively relieve the phenomenon of imbalance of the electrocardio critical value label.
Training the matrix regression model to obtain the prediction probability Y of the electrocardio critical value, wherein Y=sigmoid<F,B>+<β,Z>) Wherein Sigmoid (- /) represents Sigmoid transformation, F and Z represent the depth feature matrix and the manual feature vector of the electrocardiographic signal, respectively, the matrix B and the vector beta are both learnable parameters,<F,B>represents the inner product of F and B,<β,Z>representing the inner product of beta and Z. There is no limit to the matrix B and the vector beta by updating the objective function l (B, beta) + in the training process of the matrix regression model𝜆 1 ||B|| * +𝜆 2 ||β|| 1 Obtaining, wherein l (B, beta) is a cross entropy loss function, |·|| * In order to be a core norm,𝜆 1 and𝜆 2 the value range is [0, + ] infinity for the super parameter; as shown in fig. 3, the deep neural network framework comprises a deep feature extractor framework, and further comprises 1 Reshape layer, 1 Dropout layer and 1 FC layer with Sigmoid activation function, which are sequentially connected behind the deep feature extractor framework. The depth feature extractor frame comprises 1 Conv module and 3 Res-SE modules; the Conv module is used for inputting electrocardiosignals to obtain characteristics of the middle layer, after the Conv module, 3 Res-SE modules are sequentially connected in series, the output of the former module is used as the input of the next module, and the last Res-SE module is used for outputting an electrocardiosignal depth characteristic matrix F.
The Conv module takes an electrocardiosignal matrix X as input, carries out convolution operation through 1 Conv layer, carries out regularization treatment through BN layer, carries out nonlinear transformation through ReLU, and outputs intermediate layer depth characteristics, wherein the output of the Conv module is used as the input of the 1 st Res-SE module;
as shown in FIG. 4, the Res-SE module includes 2 parts, namely a main line and a branch line, and the inputs of the main line and the branch line of the first Res-SE module are the outputs of the Conv module; each Res-SE module comprises 2 outputs, namely an output 1 and an output 2, wherein the output 1 is used as the input of the main line of the next Res-SE module, and the output 2 is used as the input of the branch line of the next Res-SE module; and so on, obtaining output 1 and output 2 of the 3 rd Res-SE module, wherein the output 1 of the 3 rd Res-SE module is the output characteristic finally obtained by the whole depth characteristic extractor, namely a depth characteristic matrix F;
the Reshape layer is used for converting the feature matrix corresponding to the output 1 of the last Res-SE module into a feature vector;
the Dropout layer is used for randomly selecting some neurons and temporarily discarding the neurons, so that the generalization capability of the model is improved; in this embodiment, the Dropout layer Dropout ratio is set to 0.5;
the FC layer with the Sigmoid activation function is used for obtaining the prediction probability of the electrocardiographic abnormality;
when the deep neural network frame is used as an electrocardiographic abnormality recognition model frame to carry out electrocardiographic abnormality recognition model training, the deep feature extractor can be synchronously obtained.
The application applies matrix regression to an electrocardiosignal critical value identification model, in particular to low-rank sparse matrix regression. In general, when a convolutional neural network is used for depth feature extraction, the network is generally composed of a plurality of convolutional layers and one or more fully-connected layers, and the last convolutional layer outputs a matrix or tensor and then is connected with the fully-connected layers through vectorization, which has the disadvantage of neglecting the spatial structure of the output matrix or tensor. The application introduces matrix regression, and retains the spatial information of the extracted depth features by a low-rank method, so that the spatial structure of the electrocardiosignal depth features can be still retained when the depth features are fused with other patient information features for critical value prediction, and the model is beneficial to obtaining better prediction precision. In addition, the sparse method is considered in the matrix regression, and the artificial features most relevant to the electrocardio critical value can be screened out while the space information is considered, so that the prediction accuracy is improved.
As shown in fig. 5, an electrocardiosignal intelligent analysis system comprises a model generation module and a service calculation module,
the model generation module is used for acquiring a sample data set, constructing a model frame and completing model training to obtain a trained electrocardio abnormality identification model and an electrocardio critical value identification model;
the service calculation module is used for receiving a new electrocardiosignal analysis request, automatically collecting electrocardiosignal data corresponding to the request and other information of an associated patient, automatically calling a trained model, and acquiring and storing electrocardiosignal abnormality prediction probability and electrocardiosignal critical value prediction probability corresponding to the request.
The electrocardiosignal intelligent analysis system is used for realizing an electrocardiosignal intelligent analysis method. The model generation module comprises a data acquisition engine, a sample library, a model training engine and a model library, wherein the data acquisition engine is used for acquiring electrocardiosignals, other information of a patient related to the electrocardiosignals, an electrocardiosignal abnormal label and an electrocardiosignal critical value label and generating a sample data set required by model training; the sample library is used for storing a sample data set; the model training engine is based on a sample data set stored in a sample library, and model training is completed; the model library is used for storing a trained electrocardiographic anomaly recognition model and an electrocardiographic critical value recognition model;
the service computing module comprises a service triggering engine, a data acquisition engine and a model computing engine; the service triggering engine is used for receiving a new electrocardiosignal analysis request and sending the new electrocardiosignal analysis request to the data acquisition engine; the data acquisition engine is used for automatically acquiring electrocardiosignal data corresponding to a new electrocardiosignal analysis request and other information of an associated patient and sending the electrocardiosignal data and other information of the associated patient to the model calculation engine; the model calculation engine is used for calling the trained electrocardio abnormality recognition model and the electrocardio critical value recognition model, obtaining electrocardio abnormality prediction probability and electrocardio critical value prediction probability corresponding to the request, and finishing prediction probability storage.
The model generation module and the service calculation module can be independently provided with a data acquisition engine, and can also share the data acquisition engine.
The model calculation engine comprises an electrocardiographic anomaly calculation sub-engine and an electrocardiographic critical value calculation sub-engine; the electrocardio abnormality calculation sub-engine is used for calling a trained electrocardio abnormality identification model to obtain electrocardio abnormality prediction probability; the electrocardio critical value calculation sub-engine is used for calling a trained electrocardio critical value identification model to obtain electrocardio critical value prediction probability.
The model calculation engine is divided into an electrocardio abnormality calculation sub-engine and an electrocardio critical value calculation sub-engine, the electrocardio abnormality calculation sub-engine is used for calling a trained electrocardio abnormality recognition model to obtain an electrocardio abnormality prediction probability, and when the electrocardio abnormality prediction probability is higher than an electrocardio abnormality probability preset threshold value, the electrocardio critical value calculation sub-engine calls the trained electrocardio critical value recognition model to obtain an electrocardio critical value prediction probability.
Example 2
As shown in fig. 6, the electrocardiosignal intelligent analysis system disclosed in the embodiment further includes a front-end interaction module and a dynamic monitoring module, wherein:
the front-end interaction module comprises a prediction result presentation unit and an artificial tag storage unit; the prediction result presentation unit is used for visually prompting a user based on the model prediction result obtained by the service calculation module, wherein the prediction result comprises an electrocardiographic abnormality prediction probability and an electrocardiographic critical value prediction probability; the manual label storage unit is used for comprehensively judging and forming an electrocardiosignal final manual interpretation label corresponding to the request by referring to prompt contents and combining professional knowledge for an electrocardiograph interpretation technician, wherein the manual interpretation label comprises an electrocardiograph abnormal label and an electrocardiograph critical value label;
the dynamic monitoring module comprises a service monitoring evaluation unit and a service update triggering engine; the service monitoring and evaluating unit is used for evaluating the model prediction effect in real time based on the model prediction result generated in the automatic accumulation application process and the manual interpretation label data; and the service update triggering engine is used for automatically triggering the update of the model and the service when the model effect does not meet the preset condition, and realizing the dynamic optimization update of the model.
As shown in fig. 7, the intelligent electrocardio-assisted system comprises an electrocardio-signal intelligent analysis system, a data processing module, a knowledge module and a prompting module;
the data processing module is used for processing the electrocardio abnormality prediction probability and the electrocardio critical value prediction probability to obtain an electrocardio abnormality grade and an electrocardio critical value grade, and specifically, grade division logic can be preset; for example, setting the probability division threshold of the electrocardiographic abnormality level to be 0.5, thereby obtaining the level of the electrocardiographic abnormality as yes or no, and setting the probability division threshold of the electrocardiographic abnormality level to be 0.3 or 0.6, thereby obtaining the level of the electrocardiographic critical value as low, medium or high;
the knowledge module is used for storing the processing knowledge; the processing knowledge is summarized related knowledge, processing advice and the like according to critical value grade and clinical working experience by doctors.
And the prompting module is used for calling the knowledge module to obtain the processing knowledge associated with the electrocardiosignal corresponding to the request and prompting a clinician after the data processing module gives the electrocardiosignal abnormal grade and the electrocardiosignal critical value grade.
The electrocardio signal intelligent analysis system is mainly used for obtaining electrocardio abnormality prediction probability and electrocardio critical value prediction probability, and the electrocardio abnormality prediction probability and the electrocardio critical value prediction probability cannot be used for diagnosing the health condition of a patient, optionally, a data processing module processes the electrocardio abnormality prediction probability and the electrocardio critical value prediction probability to obtain an electrocardio abnormality grade and an electrocardio critical value grade, and an electrocardio interpretation technician can comprehensively make decisions according to the electrocardio abnormality and critical value prediction probability or grade and comprehensive clinical expertise to give out final interpretation results including the judgment conditions of electrocardio abnormality and critical value, so that more efficient and more accurate analysis and judgment are realized, and even early recognition of heart disease risks is expected.
As shown in fig. 8, an embodiment of the present application further provides an electronic device, which may include a processor 801, where the processor 801 is configured to execute the steps of the above-described method for intelligently analyzing an electrocardiograph signal. As can also be seen from fig. 6, the electronic device provided by the above embodiment further comprises a non-transitory computer readable storage medium 802, on which non-transitory computer readable storage medium 802 a computer program is stored, which computer program, when being executed by the processor 801, performs the steps of the above-described method for intelligent analysis of cardiac electrical signals.
In particular, the non-transitory computer readable storage medium 802 can be a general-purpose storage medium, such as a removable disk, a hard disk, a FLASH, a read-only memory (ROM), an erasable programmable read-only memory (EPROM or FLASH memory), or a portable compact disc read-only memory (CD-ROM), etc., and the computer program on the non-transitory computer readable storage medium 802 can cause the processor 801 to perform the steps of the method for intelligent analysis of electrocardiographic signals described above when the computer program is executed by the processor 801.
In practice, the non-transitory computer readable storage medium 802 may be included in the apparatus/device/system described in the above embodiment, or may exist alone, and not be assembled into the apparatus/device/system. The computer-readable storage medium carries one or more programs that, when executed, are capable of performing the steps of the intelligent analysis method of electrocardiographic signals described above.
Claims (14)
1. An electrocardiosignal intelligent analysis method is characterized by comprising the following steps:
acquiring a plurality of electrocardiosignals, other information of a patient related to the electrocardiosignals, an electrocardiosignal abnormality label and an electrocardio critical value label, and constructing a sample data set D;
dividing the sample data set D into a first training set D train,1 Verification set D vali And test set D test ；
Constructing an electrocardiographic abnormality recognition model framework to obtain a first training set D train,1 Training to obtain an electrocardiographic abnormality recognition model for input;
constructing a second training set D for training an electrocardiographic critical value recognition model train,2 ；
Constructing an electrocardio critical value recognition model frame, and inputting a second training set D train,2 Training to obtain an electrocardio critical value identification model;
and inputting the electrocardiosignals to be analyzed into a trained electrocardiosignal abnormality recognition model to obtain an electrocardiosignal abnormality prediction probability, inputting the electrocardiosignals and other information of a patient associated with the electrocardiosignals into the trained electrocardiosignal critical value recognition model when the electrocardiosignal abnormality prediction probability is higher than an electrocardiosignal abnormality probability preset threshold value, and outputting the electrocardiosignal critical value prediction probability.
2. The method of claim 1, wherein the electrocardiographic anomaly tag indicates whether the electrocardiograph signal is abnormal; the electrocardio critical value label represents whether critical value exists in the electrocardio signals or not; the electrocardiographic abnormality identification model frame is a deep neural network frame; the electrocardio critical value identification model framework comprises a depth feature extractor, a manual feature extractor and a matrix regression model.
3. The method according to claim 2, wherein the training of the electrocardiographic critical value recognition model comprises the steps of:
will second training set D train,2 The central electrocardiosignal is input into a depth feature extractor to obtain an electrocardiosignal depth feature matrix F;
will second training set D train,2 Inputting other information of a patient associated with the electrocardiosignal into a manual feature extractor to obtain a manual feature vector Z;
by using an electrocardiosignal depth feature matrix F, a manual feature vector Z and a second training set D train,2 And (3) taking the central electrocardio critical value label as input, and training a matrix regression model.
4. The method of claim 3, wherein the second training set D train,2 The construction of (2) comprises the following steps:
first, a first training set D is extracted train,1 In which the samples with the electrocardiographic anomalies form a first training subset D train,2,1 ；
Second, the first training set D train,1 Inputting the electrocardiosignals into a trained electrocardiosignal abnormality recognition model to obtain the prediction probability of whether each electrocardiosignal is abnormal or not, and obtaining a first training set D train,1 The samples in the training sequence are sampled in layers according to the prediction probability to obtain a second training subset D train,2,2 ；
Finally, the first training subset D train,2,1 And a second training subset D train,2,2 Obtaining a second training set D for training the electrocardio critical value recognition model after taking the union set train,2 。
5. The method of claim 4, wherein the training of the matrix regression model is used to obtain a predictive probability Y of the cardiac crisis value, Y = Sigmoid (< F, B > + < β, Z >), wherein Sigmoid (·) represents Sigmoid transformation, F and Z represent the cardiac signal depth feature matrix and the manual feature vector, respectively, the matrix B and the vector β are each a learnable parameter, < F, B > represents an inner product of F and B, and < β, Z > represents an inner product of β and Z.
6. The method of claim 5, wherein the matrix B and the vector β are obtained by updating the objective function l (B, β) +λ during training of the matrix regression model 1 ||B|| * +λ 2 ||β|| 1 Obtained by lambda 1 And lambda (lambda) 2 Is a super parameter.
7. The method of claim 6, wherein the deep neural network framework comprises a deep feature extractor framework, and further comprising 1 Reshape layer, 1 Dropout layer, and 1 FC layer with Sigmoid activation function sequentially connected after the deep feature extractor framework.
8. The method of claim 7, wherein the depth profile extractor frame comprises 1 Conv module and 3 Res-SE modules; the Conv module is used for inputting electrocardiosignals to obtain characteristics of the middle layer, after the Conv module, 3 Res-SE modules are sequentially connected in series, the output of the former module is used as the input of the next module, and the last Res-SE module is used for outputting an electrocardiosignal depth characteristic matrix F.
9. An electrocardiosignal intelligent analysis system is characterized by comprising a model generation module and a service calculation module,
the model generation module is used for acquiring a sample data set, constructing a model frame and completing model training to obtain a trained electrocardio abnormality identification model and an electrocardio critical value identification model;
the service calculation module is used for receiving a new electrocardiosignal analysis request, automatically collecting electrocardiosignal data corresponding to the request and other information of an associated patient, automatically calling a trained model, and acquiring and storing electrocardiosignal abnormality prediction probability and electrocardiosignal critical value prediction probability corresponding to the request;
the model generation module comprises a data acquisition engine, a sample library, a model training engine and a model library, wherein the data acquisition engine is used for acquiring electrocardiosignals, other information of a patient related to the electrocardiosignals, an electrocardiosignal abnormality label and an electrocardiosignal critical value label and generating a sample data set required by model training; the sample library is used for storing a sample data set; the model training engine is based on a sample data set stored in a sample library, and model training is completed; the model library is used for storing a trained electrocardiographic anomaly recognition model and an electrocardiographic critical value recognition model;
the service computing module comprises a service triggering engine, a data acquisition engine and a model computing engine; the service triggering engine is used for receiving a new electrocardiosignal analysis request and sending the new electrocardiosignal analysis request to the data acquisition engine; the data acquisition engine is used for automatically acquiring electrocardiosignal data corresponding to a new electrocardiosignal analysis request and other information of an associated patient and sending the electrocardiosignal data and other information of the associated patient to the model calculation engine; the model calculation engine is used for calling the trained electrocardio abnormality recognition model and the electrocardio critical value recognition model, obtaining electrocardio abnormality prediction probability and electrocardio critical value prediction probability corresponding to the request, and finishing prediction probability storage.
10. The intelligent analysis system of electrocardiosignal according to claim 9, wherein the model calculation engine comprises an electrocardio abnormality calculation sub-engine and an electrocardio critical value calculation sub-engine; the electrocardio abnormality calculation sub-engine is used for calling a trained electrocardio abnormality identification model to obtain electrocardio abnormality prediction probability; the electrocardio critical value calculation sub-engine is used for calling a trained electrocardio critical value identification model to obtain electrocardio critical value prediction probability.
11. The system for intelligent analysis of electrocardiograph signals according to claim 10, further comprising a front-end interaction module and a dynamic monitoring module,
the front-end interaction module comprises a prediction result presentation unit and an artificial tag storage unit; the prediction result presentation unit is used for visually prompting a user based on the model prediction result obtained by the service calculation module, wherein the prediction result comprises an electrocardiographic abnormality prediction probability and an electrocardiographic critical value prediction probability; the manual label storage unit is used for comprehensively judging and forming a final manual interpretation label of the electrocardiosignals by referring to prompt contents and combining professional knowledge for an electrocardiosignal interpretation technician, and comprises an electrocardiosignal abnormal label and an electrocardiosignal critical value label;
the dynamic monitoring module comprises a service monitoring evaluation unit and a service update triggering engine; the service monitoring and evaluating unit is used for evaluating the model prediction effect in real time based on the model prediction result generated in the automatic accumulation application process and the manual interpretation label data; and the service update triggering engine is used for automatically triggering the update of the model and the service when the model effect does not meet the preset condition, and realizing the dynamic optimization update of the model.
12. The intelligent electrocardio-signal intelligent analysis system is characterized by comprising the electrocardio-signal intelligent analysis system as claimed in claims 9-11, and further comprising a data processing module, a knowledge module and a prompting module;
the data processing module is used for processing the electrocardio abnormality prediction probability and the electrocardio critical value prediction probability to obtain an electrocardio abnormality grade and an electrocardio critical value grade;
the knowledge module is used for storing the processing knowledge;
and the prompting module is used for calling the knowledge module to obtain the processing knowledge related to the electrocardiosignal and prompting a clinician after the data processing module gives the electrocardiosignal abnormal grade and the electrocardiosignal critical value grade.
13. An electronic device, comprising: a processor;
a memory storing a program configured to implement the method of intelligent analysis of cardiac electrical signals as claimed in any one of claims 1 to 8 when executed by the processor.
14. A non-transitory computer readable storage medium storing instructions that, when executed by a processor, cause the processor to perform the intelligent analysis method of an electrocardiograph signal according to any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211220314.3A CN115607166B (en) | 2022-10-08 | 2022-10-08 | Intelligent electrocardiosignal analysis method and system and intelligent electrocardio auxiliary system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211220314.3A CN115607166B (en) | 2022-10-08 | 2022-10-08 | Intelligent electrocardiosignal analysis method and system and intelligent electrocardio auxiliary system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115607166A CN115607166A (en) | 2023-01-17 |
| CN115607166B true CN115607166B (en) | 2023-08-22 |
Family
ID=84860673
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211220314.3A Active CN115607166B (en) | 2022-10-08 | 2022-10-08 | Intelligent electrocardiosignal analysis method and system and intelligent electrocardio auxiliary system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115607166B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116211316B (en) * | 2023-04-14 | 2023-07-28 | 中国医学科学院阜外医院 | Type identification method, system and auxiliary system for multi-lead electrocardiosignal |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107220518A (en) * | 2017-06-28 | 2017-09-29 | 湖南暄程科技有限公司 | A kind of medical pre-alert notification method and system |
| CN109846474A (en) * | 2019-03-26 | 2019-06-07 | 深圳市理邦精密仪器股份有限公司 | The processing method and processing device of electrocardiogram, the long-range processing method of electrocardiogram and system |
| CN111759304A (en) * | 2020-07-01 | 2020-10-13 | 杭州脉流科技有限公司 | Electrocardiogram abnormity identification method and device, computer equipment and storage medium |
| CN112353402A (en) * | 2020-10-22 | 2021-02-12 | 平安科技（深圳）有限公司 | Training method of electrocardiosignal classification model, electrocardiosignal classification method and device |
| CN112842354A (en) * | 2020-12-30 | 2021-05-28 | 武汉联影智融医疗科技有限公司 | Electrocardio data critical value analysis method, device and system |
| WO2021208902A1 (en) * | 2020-04-15 | 2021-10-21 | 华为技术有限公司 | Sleep report generation method and apparatus, terminal, and storage medium |
| CN113855038A (en) * | 2021-10-15 | 2021-12-31 | 南方医科大学 | Electrocardiosignal critical value prediction method and device based on multi-model integration |
| CN114010173A (en) * | 2021-11-01 | 2022-02-08 | 暨南大学附属第一医院（广州华侨医院） | System and method for evaluating non-valvular atrial fibrillation cardiogenic stroke based on neural network |
| WO2022052358A1 (en) * | 2020-09-09 | 2022-03-17 | 生物岛实验室 | Cardiac abnormality detection network training method and cardiac abnormality early warning method and apparatus |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| GB2560339B (en) * | 2017-03-07 | 2020-06-03 | Transf Ai Ltd | Prediction of cardiac events |
| GB2582124A (en) * | 2018-09-07 | 2020-09-16 | Transf Ai Ltd | Analysis of cardiac data |
-
2022
- 2022-10-08 CN CN202211220314.3A patent/CN115607166B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107220518A (en) * | 2017-06-28 | 2017-09-29 | 湖南暄程科技有限公司 | A kind of medical pre-alert notification method and system |
| CN109846474A (en) * | 2019-03-26 | 2019-06-07 | 深圳市理邦精密仪器股份有限公司 | The processing method and processing device of electrocardiogram, the long-range processing method of electrocardiogram and system |
| WO2021208902A1 (en) * | 2020-04-15 | 2021-10-21 | 华为技术有限公司 | Sleep report generation method and apparatus, terminal, and storage medium |
| CN111759304A (en) * | 2020-07-01 | 2020-10-13 | 杭州脉流科技有限公司 | Electrocardiogram abnormity identification method and device, computer equipment and storage medium |
| WO2022052358A1 (en) * | 2020-09-09 | 2022-03-17 | 生物岛实验室 | Cardiac abnormality detection network training method and cardiac abnormality early warning method and apparatus |
| CN112353402A (en) * | 2020-10-22 | 2021-02-12 | 平安科技（深圳）有限公司 | Training method of electrocardiosignal classification model, electrocardiosignal classification method and device |
| CN112842354A (en) * | 2020-12-30 | 2021-05-28 | 武汉联影智融医疗科技有限公司 | Electrocardio data critical value analysis method, device and system |
| CN113855038A (en) * | 2021-10-15 | 2021-12-31 | 南方医科大学 | Electrocardiosignal critical value prediction method and device based on multi-model integration |
| CN114010173A (en) * | 2021-11-01 | 2022-02-08 | 暨南大学附属第一医院（广州华侨医院） | System and method for evaluating non-valvular atrial fibrillation cardiogenic stroke based on neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115607166A (en) | 2023-01-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107837082B (en) | Automatic electrocardiogram analysis method and device based on artificial intelligence self-learning | |
| US11134880B2 (en) | Automatic method to delineate or categorize an electrocardiogram | |
| Liu et al. | Arrhythmia classification of LSTM autoencoder based on time series anomaly detection | |
| CN107714023B (en) | Static electrocardiogram analysis method and device based on artificial intelligence self-learning | |
| CN107951485B (en) | Ambulatory ECG analysis method and apparatus based on artificial intelligence self study | |
| CN107451417B (en) | Dynamic electrocardiogram analysis intelligent diagnosis system and method | |
| US20100217144A1 (en) | Diagnostic and predictive system and methodology using multiple parameter electrocardiography superscores | |
| CN111626114B (en) | Electrocardiosignal arrhythmia classification system based on convolutional neural network | |
| CN106815570B (en) | Electrocardiosignal ST-T segment identification method based on dynamic pattern identification | |
| WO2019161611A1 (en) | Ecg information processing method and ecg workstation | |
| CN110974214A (en) | Automatic electrocardiogram classification method, system and equipment based on deep learning | |
| US20220265199A1 (en) | Automatic method to delineate or categorize an electrocardiogram | |
| CN109360657B (en) | Time period reasoning method for selecting samples of hospital infection data | |
| CN115607166B (en) | Intelligent electrocardiosignal analysis method and system and intelligent electrocardio auxiliary system | |
| Patra et al. | Predictive analysis of rapid spread of heart disease with data mining | |
| CN112932498A (en) | T wave morphology classification system with strong generalization capability based on deep learning | |
| Rawi et al. | Classification and detection of ECG arrhythmia and myocardial infarction using deep learning: a review | |
| Qu et al. | ECG Heartbeat Classification Detection Based on WaveNet-LSTM | |
| CN116226724A (en) | Non-invasive coronary heart disease detection system | |
| Hong et al. | Gated temporal convolutional neural network and expert features for diagnosing and explaining physiological time series: a case study on heart rates | |
| Kovalchuk et al. | A novel feature vector for ECG classification using deep learning | |
| Yao et al. | Combining rhythm information between heartbeats and BiLSTM-Treg algorithm for intelligent beat classification of arrhythmia | |
| CN113616216B (en) | Electrocardiosignal classification method based on BiLSTM-Treg | |
| Ciocîrlan et al. | Classification and Feature Extraction of Biological Signals Using Machine Learning Techniques | |
| Radiuk et al. | A Novel Feature Vector for ECG Classification using Deep Learning |
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 |