CN117322888B - Single-lead electrocardiograph data processing method and device based on time-frequency domain fusion - Google Patents
Single-lead electrocardiograph data processing method and device based on time-frequency domain fusion Download PDFInfo
- Publication number
- CN117322888B CN117322888B CN202311441893.9A CN202311441893A CN117322888B CN 117322888 B CN117322888 B CN 117322888B CN 202311441893 A CN202311441893 A CN 202311441893A CN 117322888 B CN117322888 B CN 117322888B
- Authority
- CN
- China
- Prior art keywords
- data
- network
- processing unit
- frequency domain
- characteristic value
- 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
- 230000004927 fusion Effects 0.000 title claims abstract description 88
- 238000003672 processing method Methods 0.000 title claims abstract description 9
- 238000000605 extraction Methods 0.000 claims abstract description 80
- 238000000034 method Methods 0.000 claims abstract description 27
- 230000009467 reduction Effects 0.000 claims abstract description 13
- 238000004364 calculation method Methods 0.000 claims description 23
- 238000011176 pooling Methods 0.000 claims description 21
- 238000004458 analytical method Methods 0.000 claims description 15
- 238000010276 construction Methods 0.000 claims description 11
- 238000005070 sampling Methods 0.000 claims description 11
- 208000019622 heart disease Diseases 0.000 claims description 7
- 230000008569 process Effects 0.000 claims description 5
- 230000010365 information processing Effects 0.000 abstract description 2
- 230000001746 atrial effect Effects 0.000 description 7
- 206010015856 Extrasystoles Diseases 0.000 description 5
- 208000000418 Premature Cardiac Complexes Diseases 0.000 description 5
- 208000024891 symptom Diseases 0.000 description 5
- 230000002861 ventricular Effects 0.000 description 5
- 230000033764 rhythmic process Effects 0.000 description 4
- 230000000747 cardiac effect Effects 0.000 description 3
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 3
- 206010006578 Bundle-Branch Block Diseases 0.000 description 2
- 206010047281 Ventricular arrhythmia Diseases 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 238000003745 diagnosis Methods 0.000 description 2
- 201000010099 disease Diseases 0.000 description 2
- 238000002565 electrocardiography Methods 0.000 description 2
- 206010047302 ventricular tachycardia Diseases 0.000 description 2
- 206010003130 Arrhythmia supraventricular Diseases 0.000 description 1
- 206010003658 Atrial Fibrillation Diseases 0.000 description 1
- 206010006580 Bundle branch block left Diseases 0.000 description 1
- 206010006582 Bundle branch block right Diseases 0.000 description 1
- 206010029470 Nodal rhythm Diseases 0.000 description 1
- 208000007888 Sinus Tachycardia Diseases 0.000 description 1
- 206010040741 Sinus bradycardia Diseases 0.000 description 1
- 208000003734 Supraventricular Tachycardia Diseases 0.000 description 1
- 206010042602 Supraventricular extrasystoles Diseases 0.000 description 1
- 206010047289 Ventricular extrasystoles Diseases 0.000 description 1
- 230000005856 abnormality Effects 0.000 description 1
- 206010003668 atrial tachycardia Diseases 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000005284 excitation Effects 0.000 description 1
- 210000002837 heart atrium Anatomy 0.000 description 1
- 201000001715 left bundle branch hemiblock Diseases 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000001615 p wave Methods 0.000 description 1
- 230000001314 paroxysmal effect Effects 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 201000007916 right bundle branch block Diseases 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
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
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Biomedical Technology (AREA)
- Medical Informatics (AREA)
- Molecular Biology (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- General Health & Medical Sciences (AREA)
- Public Health (AREA)
- Veterinary Medicine (AREA)
- Heart & Thoracic Surgery (AREA)
- Pathology (AREA)
- Biophysics (AREA)
- Psychiatry (AREA)
- Signal Processing (AREA)
- Physiology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Cardiology (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The invention relates to the technical field of electrocardiograph information processing, and particularly discloses a single-lead electrocardiograph data processing method and device for time-frequency domain fusion, wherein the method comprises the following steps: the method comprises the steps of obtaining original single-lead electrocardiograph data sampled by a user, generating a processing unit A, calculating the processing unit A according to a preset data MASK processing strategy to obtain a processing unit A1, calculating the processing unit A1 according to a set time domain feature extraction network, a set frequency domain feature extraction network and a set feature splicing network to obtain a fusion feature value, obtaining prediction data according to a preset data reduction processing strategy, and obtaining statistical data according to a preset heart rate statistical strategy. The invention can be used for extracting the time domain features and the frequency domain features of a relatively small amount of untagged electrocardio data as the basis of prediction and statistics, so that the related data which can embody the state of the heart can be obtained under the condition of less data processing amount, unnecessary time and cost investment are reduced, and the accuracy of electrocardio identification is ensured.
Description
Technical Field
The invention relates to the technical field of electrocardiographic information processing, in particular to a single-lead electrocardiographic data processing method and device based on time-frequency domain fusion.
Background
The heart is excited successively by the pacing points, atria, ventricles in each cardiac cycle, accompanied by bioelectric changes, called electrocardiograms. The pattern of various forms of potential changes drawn from the body surface by an electrocardiograph is called an electrocardiogram (ECG for short). Electrocardiography is an objective indicator of the occurrence, spread, and recovery process of cardiac excitation.
The electrocardiosignal is widely focused in recent years as a biological characteristic with high safety, and the biological characteristic has obvious difference among different individuals and can be used as an important basis for disease diagnosis. The electrocardiographic data is a time sequence with extremely strong regularity, and the marking of the electrocardiographic data is very complex, and one heart rhythm period comprises a plurality of characteristic waveforms, such as: p wave, PQ segment, PRS wave, ST segment, T wave, U wave, etc., so when used for disease diagnosis or heart condition monitoring, the selective processing of the electrocardiograph data is needed, the data processing amount is reduced as much as possible, otherwise, a great deal of time and cost are consumed.
Based on the above-mentioned problems, how to accurately represent and analyze the heart state by using a relatively small amount of electrocardiographic data is a problem to be solved.
Disclosure of Invention
Aiming at the technical problems in the prior art, the invention provides a single-lead electrocardio data processing method and device for time-frequency domain fusion.
The invention comprises a single-lead electrocardio data processing method of time-frequency domain fusion, which comprises the following steps:
acquiring original single-lead electrocardiograph data sampled by a user through a preset sampling frequency, and generating a processing unit A;
Constructing a time domain feature extraction network, a frequency domain feature extraction network and a feature splicing network;
The processing unit A is operated according to a preset data MASK processing strategy to obtain a processing unit A1, and the processing unit A1 is operated according to the time domain feature extraction network, the frequency domain feature extraction network and the feature splicing network to obtain a fusion feature value Cv1;
according to a preset data reduction processing strategy, carrying OUT operation on the fusion characteristic value Cv1 to obtain predicted data OUT1;
and calculating according to a preset heart rate statistics strategy and the fusion characteristic value Cv1 to obtain statistics data OUT2.
Further, the operation on the processing unit a according to a preset data MASK processing policy to obtain a processing unit A1 includes:
And uniformly dividing the processing unit A into M sections, randomly extracting one section of data Y1 in the middle section for zero setting operation, obtaining the processing unit A1 together with the rest data, and recording the section number M and the section of data Y1.
Further, according to a preset data reduction processing strategy and the fusion characteristic value Cv1, performing an operation to obtain predicted data OUT1, including:
Acquiring the first n data values and the last n data values of the number m, and performing the operation of a cross-attetion algorithm with the fusion characteristic value Cv1 to obtain first operation data;
and expanding the first operation data into one-dimensional data, and then performing full-connection network algorithm operation to obtain the predicted data OUT1.
Further, the method further comprises the following steps:
according to a loss value function Reversely updating network parameters in the time domain feature extraction network, the frequency domain feature extraction network and the feature splicing network;
wherein, ，/>Is an integer and/>。
Further, constructing a time domain feature extraction network, a frequency domain feature extraction network and a feature stitching network, including:
The time domain feature extraction network sequentially comprises the following steps from an input layer: a first convolution operation, a second convolution operation, a third convolution operation, an axis replacement operation, a fourth convolution operation, a first max pooling operation, a fifth convolution operation, a second max pooling operation, a sixth convolution operation, a third max pooling operation, a seventh convolution operation, a fourth max pooling operation, and Reshape function operations;
The frequency domain feature extraction network comprises, from an input layer: FFT operation;
the feature stitching network includes operations that perform Reshape functions in sequence and operations of the self-attetion algorithm.
Further, according to a preset heart rate statistics strategy and the fusion characteristic value Cv1, calculating to obtain statistics data OUT2, including:
carrying OUT Reshape function operation on the fusion characteristic value Cv1, and obtaining statistical data OUT2 through two-layer Bi-LSTM network operation; and in the operation process, taking the first t pieces of data from the first occurrence of 0 as an end, wherein t is the heart rate statistic value.
Further, the method further comprises the following steps:
according to a loss value function Reversely updating network parameters in the time domain feature extraction network, the frequency domain feature extraction network and the feature splicing network;
wherein, ，/>Is an integer and/>，/>，The heart rate label Y2 is the number of heart rate labels Y2, and the heart rate labels Y2 are a set of R peak coordinates in the original single-lead electrocardiograph data.
Further, the method further comprises the following steps:
calculating the processing unit A according to the time domain feature extraction network, the frequency domain feature extraction network and the feature splicing network to obtain a fusion feature value Cv0;
And analyzing and judging the fusion characteristic value Cv0 based on a preset heart disease judgment standard, and generating an analysis result label Y3.
Further, the method further comprises the following steps:
The fusion characteristic value Cv1 is sequentially calculated through a self-attetion algorithm and a Reshape function to obtain third operation data;
pooling the third operation data to obtain one-dimensional vector data, and converting the one-dimensional vector data into two-dimensional vector data through Reshape functions;
performing softmax operation on the two-dimensional vector data to obtain a probability vector OUT3;
and according to the loss value function Reversely updating the network parameters in the steps; wherein,。
The invention also comprises a single-lead electrocardio data processing device fused in time-frequency domain, which comprises an original data acquisition module, a network construction module, a fusion characteristic value calculation module, a prediction module and a statistics module, wherein:
The original data acquisition module is connected with the fusion characteristic value calculation module and is used for acquiring original single-lead electrocardiograph data sampled by a user through a preset sampling frequency and generating a processing unit A;
The network construction module is connected with the fusion characteristic value calculation module and is used for constructing a time domain characteristic extraction network, a frequency domain characteristic extraction network and a characteristic splicing network;
The fusion characteristic value calculation module is connected with the original data acquisition module, the network construction module, the prediction module and the statistics module, and is used for calculating the processing unit A according to a preset data MASK processing strategy to obtain a processing unit A1, and calculating the processing unit A1 according to the time domain characteristic extraction network, the frequency domain characteristic extraction network and the characteristic splicing network to obtain a fusion characteristic value Cv1;
The prediction module is connected with the fusion characteristic value calculation module and is used for calculating according to a preset data reduction processing strategy and the fusion characteristic value Cv1 to obtain prediction data OUT1;
The statistics module is connected with the fusion characteristic value calculation module and is used for calculating according to a preset heart rate statistics strategy and the fusion characteristic value Cv1 to obtain statistics data OUT2.
According to the time-frequency domain fused single-lead electrocardiograph data processing method and device, original single-lead electrocardiograph data sampled by a user are obtained through preset sampling frequency, a processing unit A is generated, the processing unit A is operated according to a preset data MASK processing strategy to obtain a processing unit A1, the processing unit A1 is operated according to a set time domain feature extraction network, a set frequency domain feature extraction network and a set feature splicing network to obtain a fused feature value Cv1, prediction data is obtained according to a preset data reduction processing strategy and the fused feature value Cv1, and statistical data is obtained according to a preset heart rate statistical strategy and the fused feature value Cv 1. According to the invention, the time domain features and the frequency domain features of a relatively small amount of untagged electrocardio data can be extracted, so that the two features of the time domain and the frequency domain are fused as the basis of prediction and statistics, the acquisition of relevant data capable of reflecting the heart state under the condition of less data processing amount is realized, the unnecessary time and cost investment is reduced, and the accuracy of electrocardio identification is ensured.
Drawings
For a clearer description of embodiments of the invention or of solutions in the prior art, the drawings which are used in the description of the embodiments or of the prior art will be briefly described, it being obvious that the drawings in the description below are only some embodiments of the invention, and that other drawings can be obtained from them without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart showing steps of a method for processing single-lead electrocardiograph data by time-frequency domain fusion according to an embodiment of the present invention;
Fig. 2 is a structural diagram of a single-conductive electrocardiograph data processing device with time-frequency domain fusion according to an embodiment of the present invention.
Detailed Description
The following description of the embodiments of the present invention 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 invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to fall within the scope of the invention.
The method for processing single-lead electrocardiograph data by time-frequency domain fusion in the embodiment of the invention, as shown in fig. 1, comprises the following steps:
step S10: and acquiring original single-lead electrocardiograph data sampled by a user through a preset sampling frequency, and generating a processing unit A.
The preset sampling frequency of this implementation step is denoted as Fs, and if the sampling duration of the original single-lead electrocardiograph data is T, the acquired original single-lead electrocardiograph data contains N data values in total, andThe processing unit A is denoted as。
For convenience of explanation of the embodiments, the embodiments of the present invention will take fs=360 hz and t=40s as examples, and a total of 360×40=14400 data values are obtained.
Step S20: and constructing a time domain feature extraction network, a frequency domain feature extraction network and a feature splicing network.
The time domain feature extraction network is used for extracting time domain features according to the data value of the processing unit A, the frequency domain feature extraction network is used for extracting frequency domain features according to the data value of the processing unit A, and the feature splicing network is used for splicing the time domain features and the frequency domain features and is used for subsequent data processing.
Specifically, the time domain feature extraction network in the embodiment of the present invention sequentially includes, from an input layer: the method comprises the following steps of a first convolution operation, a second convolution operation, a third convolution operation, an axis replacement operation, a fourth convolution operation, a first maximum pooling operation, a fifth convolution operation, a second maximum pooling operation, a sixth convolution operation, a third maximum pooling operation, a seventh convolution operation, a fourth maximum pooling operation and Reshape function operation. The frequency domain feature extraction network comprises, from an input layer: and (5) FFT operation. The feature stitching network includes operations that perform Reshape functions in sequence and operations of the self-attetion algorithm.
Taking fs=360 hz and t=40s as an example, the kernel size of the input layer of the time domain feature extraction network is (1,1,14400), if the kernel sizes of the first convolution operation, the second convolution operation and the third convolution operation are sequentially (1, 7), (1, 14), and the steps are sequentially (1, 3), (1, 2), the filter layers are respectively 8, 16 and 32, the output sizes after the sequential operations are respectively (8,1,100), (16,1,1600) and (32,1,800), and in fact, in the third convolution operation, the electrocardiographic segment of 14 x 3/360=0.35S can be fitted, and in the start-stop division of each wave of the electrocardiographic graph, the longest T wave is generally within 0.25S, so that the three convolution operations can be set to meet the requirement of data analysis.
After obtaining output data with the size of (32,1,800), performing axis replacement to obtain output data with the size of (1,32,800), and performing fourth convolution operation with the kernel size of (3, 3), the step of (1, 1) and the filter layer number of 32 to obtain output data with the size of (32,32,800); then, performing first maximum pooling operation with the kernel size of (2, 2) and the stepping of (2, 2) to obtain output data with the size of (32,16,400); then, obtaining output data with the size of (64,16,400) through fifth convolution operation with the kernel size of (3, 3) and the step of (1, 1) and the filter layer number of 64; then, obtaining output data with the size of (64,8,200) through a second maximum pooling operation with the core size of (2, 2) and the stepping of (2, 2); then, obtaining output data with the size of (128,8,200) through a sixth convolution operation with the kernel size of (3, 3) and the step of (1, 1) and the filter layer number of 128; then, obtaining output data with the size of (128,4,100) through a third maximum pooling operation with the core size of (2, 2) and the stepping of (2, 2); then, obtaining output data with the size of (256,4,100) through seventh convolution operation with the kernel size of (3, 3) and the step of (1, 1) and the filter layer number of 256; then, performing fourth maximum pooling operation with the kernel size of (2, 2) and the stepping of (2, 2) to obtain output data with the size of (256,2,50); finally, the output data with the size (512,50) is obtained through Reshape function operation.
For the frequency domain feature extraction network, the core size of the input layer is (1,1,14400), the step is (1,267) through the core size of (1, 1024), and the FFT operation with the filter layer number of 1 is also obtained, so that the output data with the size of (512,50) is obtained. In this step, the core size is set by taking into account that in general, the length of the FFT is preferably such that two complete heartbeats occur, so that the spectral energy component thereof can be analyzed, and 1024 pieces of data (i.e., 2.84 s) are selected as a set for calculation.
For the feature splicing network, splicing the time domain feature data (512,50) and the frequency domain feature data (512,50) to obtain a feature vector (512,100) with dimensions, wherein the vector is equivalent to the time domain and frequency domain features of the point with dimensions of 100 according to time arrangement, then carrying out Reshape function operation to obtain a vector (64,8,100) with dimensions, and obtaining a fusion feature value (64,8,100) through 8-head 4-layer self-attetion algorithm operation.
The above specific structures and settings of the related parameters regarding the time domain feature extraction network, the frequency domain feature extraction network, and the feature stitching network are only examples, and should not be construed as limiting the scope of the present invention.
Step S30: and operating the processing unit A according to a preset data MASK processing strategy to obtain a processing unit A1, and operating the processing unit A1 according to a time domain feature extraction network, a frequency domain feature extraction network and a feature splicing network to obtain a fusion feature value Cv1.
The embodiment of the invention relates to MASK processing and restoration of data, in particular to a step S30: the processing unit A is operated according to a preset data MASK processing strategy to obtain a processing unit A1, which comprises the following steps:
the processing unit A is uniformly divided into M sections, a section of data Y1 is randomly extracted from the middle section for zero setting operation, the processing unit A1 is obtained together with the rest data, and the section number M and the section of data Y1 are recorded.
Assuming that the value of M is 64, the total 14400 data of the processing unit a is uniformly divided into 64 segments, and each segment of data is 40/64=0.625 s. In this step, for example, in 4 to 61 sections, a section of 0.625s data (i.e., 14400/64=225 data in total) is randomly extracted and directly zeroed, and the new whole section of data is obtained together with the rest of data, and recorded as the processing unit A1, and the number m of the zeroed section and the section of data Y1 are recorded.
Based on the constructed time domain feature extraction network, frequency domain feature extraction network and feature splicing network, the processing unit A1 is used as input data, and the (64,8,100) dimensional fusion feature value Cv1 is finally obtained.
Step S40: and carrying OUT operation according to a preset data reduction processing strategy and the fusion characteristic value Cv1 to obtain predicted data OUT1.
Specifically, step S40: according to a preset data reduction processing strategy and a fusion characteristic value Cv1, carrying OUT operation to obtain predicted data OUT1, wherein the method comprises the following steps:
Step S401: and acquiring the first n data values and the last n data values of the number m, and performing the operation of a cross-attetion algorithm with the fusion characteristic value Cv1 to obtain first operation data.
The value of n is not specifically limited, but the value of n should not exceed the whole data size of the processing unit A1, and assuming that n=400, the step is to obtain the first 400 data values and the last 400 data values of the number m, obtain data in (8,100) dimension, and perform cross-attetion operation with the fusion characteristic value Cv1, so as to obtain first operation data in (8,100) dimension.
Step S402: and expanding the first operation data into one-dimensional data, and then performing full-connection network algorithm operation to obtain predicted data OUT1.
The first operation data in the (8,100) dimension is expanded into the (1, 800) dimension, and then a layer of full connection (800 x 225) is performed, so as to obtain predicted data OUT1, which contains 225 predicted values.
Step S50: and calculating according to a preset heart rate statistics strategy and the fusion characteristic value Cv1 to obtain statistics data OUT2.
Specifically, step S50: calculating according to a preset heart rate statistics strategy and a fusion characteristic value Cv1 to obtain statistics data OUT2, wherein the method comprises the following steps:
carrying OUT Reshape function operation on the fusion characteristic value Cv1, and obtaining statistical data OUT2 through two-layer Bi-LSTM network operation; and in the operation process, taking the first t pieces of data from the first occurrence of 0 as an end, wherein t is the heart rate statistic value.
As an example, the fused characteristic value Cv1 is subjected to Reshape function operation to obtain data in the (128,200) dimension, and then statistical data OUT2 in the (1, 200) dimension is obtained through two-layer Bi-LSTM network operation; in the operation process, the first occurrence of 0 is taken as the end, the first t are taken, and t is the heart rate statistic value obtained in the step.
Specifically, on the basis of the above embodiment, the present invention further includes:
Step S60: according to a loss value function And reversely updating network parameters in the time domain feature extraction network, the frequency domain feature extraction network and the feature splicing network.
Wherein,，/>Is an integer and/>。
When (when)Loss function/>. Loss function/>Equal to zero or approaching zero, indicating that the indiscriminate or minimal difference between the predicted data OUT1 and the data Y1 is the best network parameters in the domain feature extraction network, the frequency domain feature extraction network and the feature stitching network.
Specifically, on the basis of the above embodiment, the present invention further includes:
step S70: and reversely updating network parameters in the time domain feature extraction network, the frequency domain feature extraction network and the feature splicing network according to the loss value function.
Wherein,，/>Is an integer and/>，/>，The heart rate label Y2 is the number of heart rate labels Y2, and the heart rate label Y2 is a set of R peak coordinates in the original single-lead electrocardiograph data.
Since the accuracy of the R-R heart rate prediction in the conventional algorithm is high, the present embodiment can use the conventional algorithm to make the heart rate tag Y2, wherein the heart rate tag Y2 is a set of R peak coordinates in the original single-lead electrocardiograph data, and the total of the R peak coordinates includes s coordinates representing the occurrence ofHeart beat. When the heart rate statistic t and the heart rate label Y2 identified in the steps of the embodiment of the invention are/>When the values are inconsistent, the smaller value of the two values is selected as heart rate data/>。
Through the above loss functionReverse updating of network parameters in a time domain feature extraction network, a frequency domain feature extraction network and a feature stitching network is carried out, and the method comprises the following steps ofIs an empirical parameter.
Specifically, on the basis of the above embodiment, the present invention further includes:
Step S80: and operating the processing unit A according to the time domain feature extraction network, the frequency domain feature extraction network and the feature splicing network to obtain a fusion feature value Cv0.
Step S90: and analyzing and judging the fusion characteristic value Cv0 based on a preset heart disease judgment standard, and generating an analysis result label Y3.
The analysis and judgment of the fusion characteristic value Cv0 comprises the analysis and judgment of whether related symptoms occur in a preset heart disease judgment standard in the fusion characteristic value Cv0, wherein the heart disease judgment standard can comprise the following analysis contents: normal, occasional premature beats, frequent atrial premature beats, paroxysmal atrial beats, short-array atrial beats, occasional bivariate, occasional trigeminal, left bundle branch block, right bundle branch block, ventricular fusion beat, junctional premature beats, supraventricular premature beats, ventricular escape beats, paced beats, R-on-T, sinus tachycardia, sinus bradycardia, atrial premature beats, atrial cardiac rhythms, accelerated atrial rhythms, atrial tachycardia, atrial fibrillation, accelerated junctional rhythms, paroxysmal supraventricular tachycardia, ventricular premature beats, unique ventricular rhythms, accelerated unique ventricular rhythms, ventricular tachycardia, artificial atrial pacing rhythms, artificial ventricular pacing rhythms, primary atrial ventricular conduction blocks, and the like.
According to the number of types of symptoms, 100 possibilities are reserved for fusion characteristic values Cv0 in the step of the invention, for judging each symptom, judging results are represented by 0 and 1, 0 indicates that a certain symptom is not possessed, 1 indicates that a certain symptom is possessed, and finally an analysis result label Y3 is obtained and is a (1, 100) dimensional vector. Y3 may contain multiple 1 s because patients often have more than two electrocardiographic abnormalities. The analysis judgment standard of the embodiment of the invention should be calibrated by a professional, so as to ensure the accuracy of the final analysis result.
Specifically, on the basis of the above embodiment, the present invention further includes:
step S100: and (3) sequentially calculating the fusion characteristic value Cv1 through a self-attetion algorithm and a Reshape function to obtain third operation data.
Step S110: and carrying out pooling on the third operation data to obtain one-dimensional vector data, and converting the one-dimensional vector data into two-dimensional vector data through Reshape functions.
Step S120: and carrying OUT softmax operation on the two-dimensional vector data to obtain a probability vector OUT3.
An example of an embodiment of the present invention is:
The fusion characteristic value Cv1 sequentially passes through a self-attetion algorithm of 8 first 1 layers to obtain a characteristic value of the (64,8,100) dimension, and then third operation data of the (8,8,200) dimension is obtained through Reshape function calculation. And (3) pooling the third operation data (8 x 8) to obtain one-dimensional vector data (1,1,200), converting the one-dimensional vector data into two-dimensional vector data (2, 100) through a Reshape function, and performing softmax operation two by two to obtain a 100-dimensional probability vector OUT3. The probability vector OUT3 of the present embodiment characterizes the probability of suffering from a heart disease, as in step S90, but this step yields the probability of suffering from each disorder.
And, the present invention further includes step S130: according to a loss value functionReversely updating the network parameters in the steps; wherein/>。
The present step updates the specific parameters of the operational model in steps S100 to S120 by the cross entropy loss function. The loss function of this step does not update the network parameters of the time domain feature extraction network, the frequency domain feature extraction network, and the feature stitching network in the aforementioned step S20, because this part of data is a limited feature, and it is more advantageous to freeze the relevant parameters.
The embodiment of the invention also comprises a single-lead electrocardiograph data processing device fused in a time-frequency domain, as shown in fig. 2, the device comprises an original data acquisition module 101, a network construction module 102, a fusion characteristic value calculation module 103, a prediction module 104 and a statistics module 105, wherein:
The original data acquisition module 101 is connected with the fusion characteristic value calculation module 103, and the original data acquisition module 101 is used for acquiring original single-lead electrocardiograph data sampled by a user through a preset sampling frequency and generating a processing unit A; the sampling duration of the original single-lead electrocardiograph data is T, the original single-lead electrocardiograph data contains N data values, N=Fs is T, and the processing unit A= [ A1, A2, … …, aN ];
the network construction module 102 is connected with the fusion characteristic value calculation module 103, and the network construction module 102 is used for constructing a time domain characteristic extraction network, a frequency domain characteristic extraction network and a characteristic splicing network;
The fusion characteristic value calculation module 103 is connected with the original data acquisition module 101, the network construction module 102, the prediction module 104 and the statistics module 105, and the fusion characteristic value calculation module 103 is used for calculating the processing unit A according to a preset data MASK processing strategy to obtain a processing unit A1, and calculating the processing unit A1 according to a time domain characteristic extraction network, a frequency domain characteristic extraction network and a characteristic splicing network to obtain a fusion characteristic value Cv1;
The prediction module 104 is connected with the fusion characteristic value calculation module 103, and the prediction module 104 is used for calculating according to a preset data reduction processing strategy and the fusion characteristic value Cv1 to obtain predicted data OUT1;
the statistics module 105 is connected to the fusion eigenvalue calculation module 103, and the statistics module 105 is configured to calculate according to a preset heart rate statistics policy and the fusion eigenvalue Cv1 to obtain statistics data OUT2.
Specifically, the device of the embodiment of the invention further comprises a model parameter updating module, which is connected with the network construction module 102, the prediction module 104 and the statistics module 105, and is used for performing a function according to the loss valueThe method comprises the steps of reversely updating network parameters in a time domain feature extraction network, a frequency domain feature extraction network and a feature splicing network; wherein/>，/>Is an integer and/>. The model parameter updating module is also used for updating the model parameter according to the loss value function/>The method comprises the steps of reversely updating network parameters in a time domain feature extraction network, a frequency domain feature extraction network and a feature splicing network; wherein/>，/>Is an integer and/>，，/>The heart rate label Y2 is the number of heart rate labels Y2, and the heart rate labels Y2 are a set of R peak coordinates in the original single-lead electrocardiograph data.
Specifically, the device of the embodiment of the invention further comprises an analysis module, wherein the analysis module is connected with the fusion characteristic value calculation module 103, and the fusion characteristic value calculation module 103 is used for calculating the processing unit A according to the time domain characteristic extraction network, the frequency domain characteristic extraction network and the characteristic splicing network to obtain a fusion characteristic value Cv0; the analysis module is used for carrying out analysis and judgment on the fusion characteristic value Cv0 based on a preset heart disease judgment standard, and generating an analysis result label Y3.
Specifically, the analysis module of the embodiment of the invention is further configured to calculate the fusion eigenvalue Cv1 sequentially through self-attetion algorithm and Reshape function to obtain third operation data; pooling the third operation data to obtain one-dimensional vector data, and converting the one-dimensional vector data into two-dimensional vector data through Reshape functions; and carrying OUT softmax operation on the two-dimensional vector data to obtain a probability vector OUT3.
Specifically, the model parameter updating module of the embodiment of the invention is further used for reversely updating the network parameters in the above steps according to the loss value function; wherein, the method comprises the following steps of.
The above implementation of the related functions of the single-conductive electrocardiograph data processing device with the time-frequency domain fusion can be understood in conjunction with the foregoing method embodiments, and will not be described herein again.
According to the time-frequency domain fused single-lead electrocardiograph data processing method and device, original single-lead electrocardiograph data sampled by a user are obtained through preset sampling frequency, a processing unit A is generated, the processing unit A is operated according to a preset data MASK processing strategy to obtain a processing unit A1, the processing unit A1 is operated according to a set time domain feature extraction network, a set frequency domain feature extraction network and a set feature splicing network to obtain a fused feature value Cv1, the predicted data OUT1 is obtained according to a preset data reduction processing strategy and the fused feature value Cv1 through operation, and the statistical data OUT2 is obtained according to a preset heart rate statistical strategy and the fused feature value Cv1 through operation. According to the invention, the time domain features and the frequency domain features of a relatively small amount of untagged electrocardio data can be extracted, so that the two features of the time domain and the frequency domain are fused as the basis of prediction and statistics, the acquisition of relevant data capable of reflecting the heart state under the condition of less data processing amount is realized, the unnecessary time and cost investment is reduced, and the accuracy of electrocardio identification is ensured.
The invention has been further described with reference to specific embodiments, but it should be understood that the detailed description is not to be construed as limiting the spirit and scope of the invention, but rather as providing those skilled in the art with the benefit of this disclosure with the benefit of their various modifications to the described embodiments.
Claims (6)
1. A time-frequency domain fusion single-lead electrocardiograph data processing method is characterized by comprising the following steps:
acquiring original single-lead electrocardiograph data sampled by a user through a preset sampling frequency, and generating a processing unit A;
Constructing a time domain feature extraction network, a frequency domain feature extraction network and a feature splicing network;
The processing unit A is operated according to a preset data MASK processing strategy to obtain a processing unit A1, and the processing unit A1 is operated according to the time domain feature extraction network, the frequency domain feature extraction network and the feature splicing network to obtain a fusion feature value Cv1;
according to a preset data reduction processing strategy, carrying OUT operation on the fusion characteristic value Cv1 to obtain predicted data OUT1;
Calculating according to a preset heart rate statistics strategy and the fusion characteristic value Cv1 to obtain statistics data OUT2;
The processing unit A is operated according to a preset data MASK processing strategy to obtain a processing unit A1, which comprises the following steps:
Uniformly dividing the processing unit A into M sections, randomly extracting one section of data Y1 in the middle section for zero setting operation, obtaining the processing unit A1 together with the rest data, and recording the section number M and the section of data Y1;
according to a preset data reduction processing strategy, the fusion characteristic value Cv1 is operated to obtain predicted data OUT1, which comprises the following steps:
Acquiring the first n data values and the last n data values of the number m, and performing the operation of a cross-attetion algorithm with the fusion characteristic value Cv1 to obtain first operation data;
the first operation data are unfolded to one-dimensional data, and then full-connection network algorithm operation is carried OUT to obtain the prediction data OUT1;
Calculating according to a preset heart rate statistics strategy and the fusion characteristic value Cv1 to obtain statistics data OUT2, wherein the method comprises the following steps:
Carrying OUT Reshape function operation on the fusion characteristic value Cv1, and obtaining statistical data OUT2 through two-layer Bi-LSTM network operation; ending with the first occurrence of 0 in the operation process, taking the first t, wherein t is the heart rate statistic value;
Further comprises:
according to a loss value function Reversely updating network parameters in the time domain feature extraction network, the frequency domain feature extraction network and the feature splicing network;
wherein, ，/>Is an integer and/>。
2. The method for processing single-lead electrocardiographic data by time-frequency domain fusion according to claim 1, wherein constructing a time domain feature extraction network, a frequency domain feature extraction network and a feature stitching network comprises:
The time domain feature extraction network sequentially comprises the following steps from an input layer: a first convolution operation, a second convolution operation, a third convolution operation, an axis replacement operation, a fourth convolution operation, a first max pooling operation, a fifth convolution operation, a second max pooling operation, a sixth convolution operation, a third max pooling operation, a seventh convolution operation, a fourth max pooling operation, and Reshape function operations;
The frequency domain feature extraction network comprises, from an input layer: FFT operation;
the feature stitching network includes operations that perform Reshape functions in sequence and operations of the self-attetion algorithm.
3. The method for processing single-lead electrocardiographic data fused in time-frequency domain according to claim 2, further comprising:
according to a loss value function Reversely updating network parameters in the time domain feature extraction network, the frequency domain feature extraction network and the feature splicing network;
wherein, ，/>Is an integer and/>，/>，/>Is the number of heart rate labels Y2, wherein the heart rate labels Y2 are the set of R peak coordinates in the original single-lead electrocardiograph data,/>Is an empirical parameter.
4. The method for processing single-lead electrocardiographic data fused in time-frequency domain according to claim 1, further comprising:
calculating the processing unit A according to the time domain feature extraction network, the frequency domain feature extraction network and the feature splicing network to obtain a fusion feature value Cv0;
And analyzing and judging the fusion characteristic value Cv0 based on a preset heart disease judgment standard, and generating an analysis result label Y3.
5. The method for processing single-lead electrocardiographic data fused in time-frequency domain according to claim 4, further comprising:
The fusion characteristic value Cv1 is sequentially calculated through a self-attetion algorithm and a Reshape function to obtain third operation data;
pooling the third operation data to obtain one-dimensional vector data, and converting the one-dimensional vector data into two-dimensional vector data through Reshape functions;
performing softmax operation on the two-dimensional vector data to obtain a probability vector OUT3;
and according to the loss value function Reversely updating the network parameters in the steps; wherein,。
6. The single-lead electrocardio data processing device with time-frequency domain fusion is characterized by comprising an original data acquisition module, a network construction module, a fusion characteristic value calculation module, a prediction module and a statistics module, wherein:
The original data acquisition module is connected with the fusion characteristic value calculation module and is used for acquiring original single-lead electrocardiograph data sampled by a user through a preset sampling frequency and generating a processing unit A;
The network construction module is connected with the fusion characteristic value calculation module and is used for constructing a time domain characteristic extraction network, a frequency domain characteristic extraction network and a characteristic splicing network;
The fusion characteristic value calculation module is connected with the original data acquisition module, the network construction module, the prediction module and the statistics module, and is used for calculating the processing unit A according to a preset data MASK processing strategy to obtain a processing unit A1, and calculating the processing unit A1 according to the time domain characteristic extraction network, the frequency domain characteristic extraction network and the characteristic splicing network to obtain a fusion characteristic value Cv1;
The prediction module is connected with the fusion characteristic value calculation module and is used for calculating according to a preset data reduction processing strategy and the fusion characteristic value Cv1 to obtain prediction data OUT1;
the statistics module is connected with the fusion characteristic value calculation module and is used for calculating according to a preset heart rate statistics strategy and the fusion characteristic value Cv1 to obtain statistics data OUT2;
The processing unit A is operated according to a preset data MASK processing strategy to obtain a processing unit A1, which comprises the following steps:
Uniformly dividing the processing unit A into M sections, randomly extracting one section of data Y1 in the middle section for zero setting operation, obtaining the processing unit A1 together with the rest data, and recording the section number M and the section of data Y1;
according to a preset data reduction processing strategy, the fusion characteristic value Cv1 is operated to obtain predicted data OUT1, which comprises the following steps:
Acquiring the first n data values and the last n data values of the number m, and performing the operation of a cross-attetion algorithm with the fusion characteristic value Cv1 to obtain first operation data;
the first operation data are unfolded to one-dimensional data, and then full-connection network algorithm operation is carried OUT to obtain the prediction data OUT1;
Calculating according to a preset heart rate statistics strategy and the fusion characteristic value Cv1 to obtain statistics data OUT2, wherein the method comprises the following steps:
Carrying OUT Reshape function operation on the fusion characteristic value Cv1, and obtaining statistical data OUT2 through two-layer Bi-LSTM network operation; ending with the first occurrence of 0 in the operation process, taking the first t, wherein t is the heart rate statistic value;
Further comprises:
according to a loss value function Reversely updating network parameters in the time domain feature extraction network, the frequency domain feature extraction network and the feature splicing network;
wherein, ，/>Is an integer and/>。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311441893.9A CN117322888B (en) | 2023-11-01 | 2023-11-01 | Single-lead electrocardiograph data processing method and device based on time-frequency domain fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311441893.9A CN117322888B (en) | 2023-11-01 | 2023-11-01 | Single-lead electrocardiograph data processing method and device based on time-frequency domain fusion |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117322888A CN117322888A (en) | 2024-01-02 |
| CN117322888B true CN117322888B (en) | 2024-04-19 |
Family
ID=89290382
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311441893.9A Active CN117322888B (en) | 2023-11-01 | 2023-11-01 | Single-lead electrocardiograph data processing method and device based on time-frequency domain fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117322888B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113397555A (en) * | 2021-07-21 | 2021-09-17 | 南通大学附属医院 | Arrhythmia classification algorithm of C-LSTM for physiological parameter monitoring |
| CN114692698A (en) * | 2022-04-22 | 2022-07-01 | 中北大学 | One-dimensional electrocardiogram data classification method based on residual error network |
| KR102458657B1 (en) * | 2022-03-24 | 2022-10-25 | (주)나눔테크 | Artificial Intelligence Based ECG Analysis and Application for Automated External Defibrillator |
| CN115500841A (en) * | 2022-09-26 | 2022-12-23 | 浙江大学 | Ventricular premature beat positioning method for fusion of time domain and frequency domain feature deep learning |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP5342066B2 (en) * | 2009-04-22 | 2013-11-13 | カーディアック ペースメイカーズ， インコーポレイテッド | Dynamic selection of algorithms for detecting arrhythmias |
-
2023
- 2023-11-01 CN CN202311441893.9A patent/CN117322888B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113397555A (en) * | 2021-07-21 | 2021-09-17 | 南通大学附属医院 | Arrhythmia classification algorithm of C-LSTM for physiological parameter monitoring |
| KR102458657B1 (en) * | 2022-03-24 | 2022-10-25 | (주)나눔테크 | Artificial Intelligence Based ECG Analysis and Application for Automated External Defibrillator |
| CN114692698A (en) * | 2022-04-22 | 2022-07-01 | 中北大学 | One-dimensional electrocardiogram data classification method based on residual error network |
| CN115500841A (en) * | 2022-09-26 | 2022-12-23 | 浙江大学 | Ventricular premature beat positioning method for fusion of time domain and frequency domain feature deep learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117322888A (en) | 2024-01-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Deng et al. | Extracting cardiac dynamics within ECG signal for human identification and cardiovascular diseases classification | |
| US20220233129A1 (en) | Method for constructing intracardiac abnormal activation point location model based on cnn and lstm | |
| CN109480825B (en) | Electrocardio data processing method and device | |
| CN110236520B (en) | Electrocardiogram type recognition device based on double convolution neural network | |
| CN107451417B (en) | Dynamic electrocardiogram analysis intelligent diagnosis system and method | |
| CN106815570B (en) | Electrocardiosignal ST-T segment identification method based on dynamic pattern identification | |
| CN109171712A (en) | Auricular fibrillation recognition methods, device, equipment and computer readable storage medium | |
| CN108836314A (en) | A kind of ambulatory ECG analysis method and system based on network and artificial intelligence | |
| CN109480827B (en) | Vector electrocardiogram classification method and device | |
| CN111053551A (en) | RR interval electrocardio data distribution display method, device, computer equipment and medium | |
| JP7487965B2 (en) | Prediction method of electrocardiogram heart rate multi-type based on graph convolution | |
| CN110236529A (en) | A kind of multi-lead arrhythmia cordis intelligent diagnosing method based on MODWT and LSTM | |
| JP2020517334A5 (en) | ||
| Srinivasan et al. | A new phase space analysis algorithm for cardiac arrhythmia detection | |
| CN117322888B (en) | Single-lead electrocardiograph data processing method and device based on time-frequency domain fusion | |
| CN112545525B (en) | Electrocardiogram data classification method, device and system | |
| CN111528833B (en) | Rapid identification and processing method and system for electrocardiosignals | |
| Cai et al. | Rule-based rough-refined two-step-procedure for real-time premature beat detection in single-lead ECG | |
| CN113180688A (en) | Coronary heart disease electrocardiogram screening system and method based on residual error neural network | |
| JP2005080712A (en) | Calculation method of heart health index and classification method of specified cardiographic wave | |
| CN112022140B (en) | Automatic diagnosis method and system for diagnosis conclusion of electrocardiogram | |
| CN114711780A (en) | Multi-lead electrocardiogram signal processing method, device, equipment and storage medium | |
| CN115316996A (en) | Training method, device and equipment for abnormal heart rhythm recognition model and storage medium | |
| JP4402298B2 (en) | System for determining the state of the patient's heart | |
| CN113940682A (en) | Atrial fibrillation identification method based on statistical characteristics |
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 |