CN116089862A - ECG arrhythmia classification method based on correction mechanism and self-adjusting ant colony algorithm - Google Patents
ECG arrhythmia classification method based on correction mechanism and self-adjusting ant colony algorithm Download PDFInfo
- Publication number
- CN116089862A CN116089862A CN202310068221.1A CN202310068221A CN116089862A CN 116089862 A CN116089862 A CN 116089862A CN 202310068221 A CN202310068221 A CN 202310068221A CN 116089862 A CN116089862 A CN 116089862A
- Authority
- CN
- China
- Prior art keywords
- ant
- pheromone
- self
- path
- ecg
- 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.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/316—Modalities, i.e. specific diagnostic methods
- A61B5/318—Heart-related electrical modalities, e.g. electrocardiography [ECG]
- A61B5/346—Analysis of electrocardiograms
- A61B5/349—Detecting specific parameters of the electrocardiograph cycle
- A61B5/352—Detecting R peaks, e.g. for synchronising diagnostic apparatus; Estimating R-R interval
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/316—Modalities, i.e. specific diagnostic methods
- A61B5/318—Heart-related electrical modalities, e.g. electrocardiography [ECG]
- A61B5/346—Analysis of electrocardiograms
- A61B5/349—Detecting specific parameters of the electrocardiograph cycle
- A61B5/363—Detecting tachycardia or bradycardia
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
- A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
-
- 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
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Cardiology (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Veterinary Medicine (AREA)
- Heart & Thoracic Surgery (AREA)
- Public Health (AREA)
- Animal Behavior & Ethology (AREA)
- Surgery (AREA)
- Pathology (AREA)
- Medical Informatics (AREA)
- Evolutionary Computation (AREA)
- Theoretical Computer Science (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Software Systems (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Fuzzy Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physiology (AREA)
- Psychiatry (AREA)
- Signal Processing (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides an ECG arrhythmia classification method based on a correction mechanism and a self-adjusting ant colony algorithm, which belongs to the technical field of ECG signal classification, improves the classification accuracy of a traditional machine learning method, compensates for the defects that a deep learning method depends on a large data set, has multiple iteration times and the like, and realizes ECG arrhythmia classification, and comprises the following steps: step 1: acquiring an ECG signal; step 2, preprocessing the ECG signal obtained in the step 1, finding out the position of an R wave peak value, and dividing the data into single periods; step 3, extracting the characteristics of the data preprocessed in the step 2; step 4, obtaining an arrhythmia pre-classification set through a self-adjustment ant colony aggregation algorithm based on a correction mechanism based on the characteristics obtained in the step 3; and 5, sorting all the results in the pre-sorting set obtained in the step 4 through a radix sorting algorithm to obtain an optimal arrhythmia sorting result.
Description
Technical Field
The invention relates to the technical field of ECG signal classification, in particular to an ECG arrhythmia classification method based on a self-adjusting ant colony algorithm of a correction mechanism.
Background
Cardiovascular diseases (CVDs) are among the most common causes of human death today, and arrhythmia is a representative type of cardiovascular disease, and sustained arrhythmias can lead to life hazards. For example, a long ventricular premature beat may become ventricular tachycardia or fibrillation, which may lead to heart failure. Different treatment measures are required according to different types of arrhythmia. Different types of arrhythmias correspond to different types of ECG, and obtaining a real-time electrocardiogram of the patient is helpful for rapidly judging the criticality of the patient. Clinically, doctors observe whether arrhythmia exists or not according to the main characteristics of the ECG, such as P waves, QRS wave groups, T waves and the like, and then judge the type of the arrhythmia. The arrhythmia classification auxiliary method based on the ECG can effectively improve the working efficiency of doctors and assist the doctors in diagnosis. Therefore, classification algorithms based on ECG arrhythmias have important research implications.
With the development of machine learning, ECG arrhythmia classification algorithms based on machine learning have become a current research hotspot. Generally, the conventional machine learning method is used in combination with feature engineering, and the classification method of the conventional machine learning method is improved by extracted medical features, statistical features, data morphology, and the like. The deep learning method has the capability of extracting hidden features and realizing nonlinear fitting between data. The convolutional neural network can extract spatial features and the recurrent neural network can capture temporal characteristics. The above methods have been widely used but have respective problems. The classification process and the classification accuracy based on the traditional machine learning method still need to be improved. The model of the method based on the convolutional neural network or the recurrent neural network is too complex and depends on a large data set and multiple iteration times.
Disclosure of Invention
Aiming at the problems, the invention provides an ECG arrhythmia classification method based on a self-adjusting ant colony algorithm of a correction mechanism, which improves the model time feature perception capability through a dynamic time planning algorithm, improves the classification accuracy of a traditional machine learning method, and overcomes the defects that a deep learning method depends on a large data set, multiple iteration times and the like through a self-adjusting ant transfer method to realize ECG arrhythmia classification.
An ECG arrhythmia classification method based on a self-adjusting ant colony algorithm of a correction mechanism comprises the following steps: the method comprises the following steps:
step 4, obtaining an arrhythmia pre-classification set through a self-adjustment ant colony aggregation algorithm based on a correction mechanism based on the characteristics obtained in the step 3;
and 5, sorting all the results in the pre-sorting set obtained in the step 4 through a radix sorting algorithm to obtain an optimal arrhythmia sorting result.
Further, step 2 includes the steps of:
step 2.1, removing high-frequency noise and baseline drift from the ECG signal obtained in the step 1 through wavelet transformation, and comparing decomposition results of different wavelet decomposition scales, wherein the R peak value of the ECG signal under the three scales of wavelet decomposition is most obvious, so that an R peak value point is detected under the scales;
step 2.2, searching the maximum value of the R wave based on the decomposition scale obtained in the step 2.1, finding out the point with the slope larger than 0, assigning 1, and assigning 0 to the rest points; searching an R wave minimum value, finding out a point with a slope smaller than 0, assigning 1, and assigning 0 to the rest points;
step 2.3, removing maximum value points and minimum value points with absolute values smaller than the detection threshold according to a preset detection threshold, and finding out the existence interval of suspected R wave crest value points;
and 2.4, performing missed detection and false detection and investigation of the R wave on the suspected R wave peak value point determined in the step 2.3, and finally determining the accurate R wave peak value point position.
Further, step 2.4 includes the steps of:
step 2.4.1, when the distance between adjacent suspected R wave crest points is smaller than a×mean (RR), indicating that a false detection phenomenon exists, wherein a is a multiple of the average distance, mean (RR) represents the average distance between adjacent R wave distances, and removing the suspected R wave crest point with the smallest absolute value under the decomposition scale obtained in step 2.1;
step 2.4.2, if the distance between adjacent R wave crest value points is greater than b×mean (RR), indicating that a missing detection phenomenon exists, wherein b is a multiple of the average distance, searching a maximum value and a minimum value pair with the maximum sum of absolute values in two suspected R wave intervals under the decomposition scale obtained in step 2.1, and positioning the zero crossing point of the found maximum value and minimum value pair as the position of the missing detection R wave peak value;
and 2.4.3, cutting the data into single periods according to the determined position of the R wave peak value.
Preferably, step 3 performs time feature extraction on the preprocessed ECG signal by a dynamic time planning algorithm to improve the perceptibility of the improved model time features.
Further, the dynamic time planning algorithm performs time feature extraction on the preprocessed ECG signal, which specifically includes the following steps:
first, a DTW distance matrix is determined by the following formula:
wherein:for the cumulative distance matrix of the kth test data and the center of the first cluster, the +.>Representing the distance between the ith point of the kth test data and the jth point of the center of the ith class cluster; />For the Euclidean distance between the ith feature of the feature vector x of the kth test data and the jth feature of the feature vector y of the center of the first class cluster, the calculation formula is shown as follows,
wherein: f is the number of features;
then, the DTW distance between the test data and the center of each cluster is calculated by the following formula:
wherein:represents the DTW distance of kth test data from the center of the kth cluster, +.>The i-th feature of the kth test data and the i-th feature of the center of the first cluster, respectively,/->And the DTW distance between the kth test data and the center of the ith class cluster is represented, f is the characteristic quantity, and e is the class cluster quantity.
Further, step 4 includes the steps of:
step 4.1: setting model parameters and setting the iteration times of the model as item num The ant amount is ant num The initial pheromone concentration of each path is tau;
step 4.2: randomly determining the position of a first ant at the beginning of a first iteration, and realizing arrhythmia pre-classification by a correction mechanism so as to reduce the classification error rate;
step 4.3: updating the pheromone concentration of the path;
step 4.4: the ants realize self-adjustment transfer according to the concentration of the pheromones and the path distance, and record the path length of the ants;
step 4.5: after one ant traverses all target points, determining a current path with the maximum pheromone concentration, wherein a random one of two end points of the path is used as a starting position of the next ant;
step 4.6: sequencing all ant traversing results after all ants realize one-time traversing;
step 4.7: when passing through iter num And after the iteration, obtaining a pre-class set.
Further, the method comprises the steps of,
step 4.2 comprises the steps of:
step 4.2.1: calculating Euclidean distance between the test data and the centers of various clusters;
step 4.2.2: the Euclidean distance between the standardized test data and the centers of various clusters;
step 4.2.3: determining a correction value in a correction mechanism based on the normalized Euclidean distance, and correcting a functionThe method comprises the following steps:
wherein:correction values calculated for the kth test data; e (E) * The Euclidean distance after the standardization of the maximum and minimum values; l is the number of types;
step 4.2.4: determining a class cluster corresponding to the minimum DTW distance;
wherein: t is the subscript of the minimum value in the DTW distance between the kth test data and the centers of various clusters; min is the minimum value in the DTW distance between the kth test data and the centers of various clusters; index () representation derives a subscript;the DTW distance between the kth test data and the center of the ith class cluster;
step 4.2.5: determining a correction coefficient b k ;
Wherein b k Representing a correction coefficient determined after the kth test data is compared with the centers of various clusters;
Preferably, in step 4.3, two convergence curves are adopted to form a convergence channel, so that dynamic updating of the pheromone volatilization coefficient ρ is faster and more stable.
Further, the pheromone volatility coefficient ρ is updated according to the following formula:
wherein ρ is 1 Representing the pheromone volatilization coefficient, ρ, determined according to the iteration number 2 Representing the pheromone volatilization coefficient determined according to the number of ants, and updating rho once by each ant based on the number of ants and the iteration number before the traversal starts;
wherein: ρ 1 Representing the pheromone volatilization coefficient, ρ, determined according to the iteration number x 2 Represents the pheromone volatilization coefficient determined according to the number x of ants, c 1 Is of formula ρ 1 Offset constant of c 2 Is of formula ρ 2 Is a constant of offset in (a); gamma is formulaρ 2 The coefficient of (2), gamma < 0,
updating the pheromone concentration according to formulas (31), (32) and (33);
τ ij (t+1)=(1-ρ)τ ij (t)+Δτ ij (31)
wherein: τ ij () For the pheromone on the iterative path (i, j), t represents the iterative times, and ρ is the pheromone volatilization coefficient; Δτ ij Is a pheromone increment;represents the pheromone left by the a-th ant on the path (i, j), n is the number of ants, L a For the total path length traversed by ant a in this iteration, through (i, j) represents the traversed path (i, j).
Further, the formula of the ant self-adjustment transition probability in step 4.4 is:
in the method, in the process of the invention,represents the transition probability of the a-th ant on the path (i, j), eta ij (t) heuristic factor representing the t-th iteration path (i, j), τ ij (t) is the pheromone, eta on the t-th iteration path (i, j) is (t) heuristic factor representing the t-th iteration path (i, s), τ is (t) represents the pheromone on the t-th iteration path (i, s), s represents the sample point, and J is the sample point set which is allowed to be selected by the ant a in the next step; alpha is the relative importance degree of the pheromone, and alpha is more than 0;beta is the relative importance of heuristic factors, beta is dynamically updated by taking path length as an independent variable, and the following formula is satisfied:
wherein: d, d ij Is the path length between (i, j).
Compared with the prior art, the invention has the beneficial effects that: aiming at the problems, the invention provides an ECG arrhythmia classification method based on a self-adjusting ant colony algorithm of a correction mechanism, which improves the model time feature perception capability through a dynamic time planning algorithm, improves the classification accuracy of a traditional machine learning method, and overcomes the defects that a deep learning method depends on a large data set, multiple iteration times and the like through a self-adjusting ant transfer method to realize ECG arrhythmia classification.
1. The data adopted by the invention comprise different subjects, a new data set is formed after the data are processed uniformly, and the new data set does not distinguish the subjects any more so as to fully consider the influence caused by the difference of the ECG signal characteristics among different individuals.
2. The invention improves the perception capability of the time features of the model by using the DTW, builds a correction mechanism according to the distance between the DTW and Euclidean distance, improves the classification accuracy of the original ant colony classification method, and solves the problem that the classification result is affected by continuously accumulating non-positive information because the difference between partial features is very small in the classification process. The classification error number of the present invention is reduced by an order of magnitude (10 1 )。
3. According to the principle that the flow velocity can be increased under the convergence channel, the invention constructs two different pheromone volatilization coefficients rho-type, namely rho, from two dimensions of whole and local 1 And ρ 2 The two sub-formulas form a dynamic update pheromone volatility coefficient rho formula. By taking into account p alone with a fixed value 1 Or ρ 2 The model using this method converges more stably and faster than the model of the model.
4. The invention constructs a truly self-adjusting ant transferring method. The negative correlation between the pheromone concentration and the path between the target nodes and the dynamic updating of the alpha and beta formulas result in ant selecting the next target point determined by the pheromone concentration and the distance between the nodes. By using the method, the total path length of convergence of the model traversal is better than that of most of comparison models, and the model traversal is more stable.
Drawings
FIG. 1 is a flow chart of an ECG arrhythmia classification method based on a self-adjusting ant colony algorithm constructed by the invention;
table 1 is the type of spatio-temporal features of the present invention for extracting pre-processed data;
FIG. 2 is a partial spatiotemporal feature presentation of the extraction of the present invention;
FIG. 3 is a schematic diagram of increasing flow rate under a converging channel;
FIG. 4 is ρ of the pheromone volatility coefficient ρ 1 A function image;
FIG. 5 is ρ of the pheromone volatility coefficient ρ 2 A function image;
FIG. 6 is a graph of relative importance α of pheromones versus relative importance β of heuristic factors;
FIG. 7 is a graph of [ eta ] derived from a dynamic update beta method] β Is a function of the image of the object.
Detailed Description
The ECG arrhythmia classification method based on the self-adjusting ant colony algorithm of the correction mechanism is further described in detail below with reference to the accompanying drawings and the specific implementation method.
The technical scheme adopted by the invention is an ECG arrhythmia classification method based on a self-adjusting ant colony algorithm of a correction mechanism, which is implemented according to the following steps:
step 4, obtaining an arrhythmia pre-classification set through the characteristics obtained in the step 3 and a self-adjusting ant colony classification algorithm based on a correction mechanism and a cardinal number ordering algorithm;
and 5, sorting all the results in the pre-sorting set obtained in the step 4 through a radix sorting algorithm to obtain an optimal arrhythmia sorting result.
The invention is also characterized in that:
and 2.1, removing high-frequency noise and baseline drift from the ECG signal obtained in the step 1 through 6 layers of wavelet transformation, and determining a wavelet transformation equation formula according to the formula (1) and the formula (2).
cA j+1 (n)=∑ m cA j (m)h(m-2n) (1)
cD j+1 (n)=∑ m cA j (m)g(m-2n) (2)
Wherein: cA (cA) j (m) mth data representing decomposition result of jth layer after passing through high frequency filter, cA j+1 (n) is the nth data of the decomposition result of the j+1th layer after passing through the high frequency filter, h (m-2 n) represents the m-2n sample point after passing through the high frequency filter, cD j+1 (n) is the nth data of the decomposition result of the j+1th layer after passing through the low frequency filter, and g (m-2 n) represents the m-2n sample points after passing through the low frequency filter. A reconstruction formula is determined according to equation (3).
cA j (n)=∑ m cA j+1 (m)h(n-2m)+cD j+1 (m)g(n-2m) (3)
Comparing the decomposition results of different wavelet decomposition scales, the R wave peak value of the ECG signal under the wavelet decomposition three-time scale is most obvious, so that the R wave peak value point is detected under the scale.
Here, the R-peak value of the ECG signal at the three scales of wavelet decomposition is most remarkable, and thus the decomposition result at the three scales is selected to detect the R-peak value point.
And 2.2, searching the maximum value of the R wave based on the decomposition scale obtained in the step 2.1, finding out the point with the slope larger than 0, assigning 1, and assigning 0 to the rest points. Searching an R wave minimum value, finding out a point with a slope smaller than 0, assigning 1, and assigning 0 to the rest points;
step 2.3, removing maximum value points and minimum value points with absolute values smaller than the detection threshold according to the preset detection threshold to obtain a suspected existence section of the R peak value point, wherein zero crossing points of adjacent maximum and minimum value pairs are suspected R peak value points; here, the detection threshold is set to be an average value of one third of the adjacent signal periods at the three-time resolution scale.
And 2.4, performing error detection and omission detection on the R wave to the suspected R wave peak value point determined in the step 2.3, and finally determining the accurate R wave peak value point position. When the distance between adjacent suspected R wave crest value points is smaller than a×mean (RR), a shows that a false detection phenomenon exists, a shows a multiple of the average distance, a is 0.4, mean (RR) shows the average distance between adjacent R wave distances, the suspected R wave crest value point with the smallest absolute value under the decomposition scale obtained in the step 2.1 is removed, if the distance between adjacent R wave crest value points is larger than b×mean (RR), b shows that a false detection phenomenon exists, b shows a multiple of the average distance, b is 1.6, a maximum and minimum value pair with the largest absolute value is found in two suspected R wave intervals under the decomposition scale obtained in the step 2.1, and the zero crossing point of the found maximum and minimum value pair can be positioned as the position of the false detection R wave peak. Finally, the data is sliced into individual periods according to the determined R-wave position. The present invention does not distinguish between subjects, and the segmented dataset contains monocycle data for all subjects, from which 100 data of the normal heart rhythm (N), left bundle branch block (L), right bundle branch block (R), atrial premature beat (a) and ventricular premature beat (V) types each were randomly selected as model test data for a total of 500 test data.
In step 3, the spatial features and temporal features of the extracted ECG signal are shown in table 1, and the partial features are shown in fig. 2. Wherein the spatial features are classified into medical features and statistical features. The invention applies DTW (Dynamic Time Warping, dynamic time planning) to improve the perceptibility of model time features;
spatial and temporal feature names of ECG signals extracted in Table 1
Step 3.1, extracting spatial features of the preprocessed ECG signal obtained in the step 2:
3.1.1, extracting medical characteristics. The R-wave is determined by the positive and negative maxima and the detection threshold. The positions of the first three extreme points are determined as the Q-wave start points and the positions of the last three extreme points are determined as the S-wave end points based on the R-wave positions. Based on the QRS complex position, the forward 2/3RR interval of the starting point of the Q wave is used as the interval of the determined P wave, the backward 2/3RR interval of the end point of the S wave is used as the interval of the determined T wave, the maximum extremum pair is searched in the two intervals, the zero crossing point of the maximum extremum pair is found to be determined as the peak point of the P wave and the T wave, and the maximum value point and the minimum value point in the interval are respectively determined as the starting point and the end point of the P wave and the T wave. And determining each medical characteristic according to the obtained starting points and peak points of the R wave, the QRS wave group, the P wave and the T wave. Determining the medical characteristics (R wave area, R wave amplitude, R wave width value, QS area, QRS time limit, T wave amplitude, P time limit, PR interval, R wave left slope, R wave right slope, R wave left width, R wave right width, R wave left angle, R wave right angle and R wave angle) of the ECG signal according to the obtained starting points and peak points of the R wave, the QRS wave group, the P wave and the T wave;
and 3.1.2, extracting statistical characteristics by using a statistical formula. Calculating a bias state according to formula (4); counting data which are larger than or equal to 0 to obtain a positive area, counting data which are smaller than 0 to obtain a negative area, and counting data which are larger than 0 to obtain a zero crossing signal; calculating an average value according to the formula (5), calculating a standard deviation according to the formula (6), and obtaining a ratio of the average value to the standard deviation according to the formula (7); calculating a variance according to equation (8); the full-length (very poor) of the output according to formula (9); calculating a lower quartile according to formula (10), calculating a quartile according to formula (11), calculating an upper quartile according to formula (12), calculating a quartile range according to formula (13), and calculating a three-mean according to formula (14); calculating an upper cutoff point according to formula (15), and calculating a lower cutoff point according to formula (16); calculating kurtosis according to formula (17); obtaining the maximum value, the minimum value, the maximum absolute value and the binary number of the data;
wherein: skew is biased; mu is the expected; sigma is the standard deviation.
R=X max -X min (9)
Q 1 =(n+1)×0.25 (10)
Q 2 =(n+1)×0.5 (11)
Q 3 =(n+1)×0.75 (12)
Wherein: q (Q) 1 A lower quartile position; q (Q) 2 Is a binary number position; q (Q) 3 Is the position of the upper quartile.
IQR=Q 3 -Q 1 (13)
T up =Q 3 +1.5IQR (15)
T low =Q 1 -1.5IQR (16)
Wherein: k is kurtosis.
And 3.2, extracting the time characteristics of the preprocessed ECG signal obtained in the step 2, and improving the perceptibility of the time characteristics of the model by applying DTW. The DTW distance matrix is determined according to equations (18) (19) (20).
Wherein:for the cumulative distance matrix of the kth test data and the center of the first cluster, the +.>Representing the distance between the ith point of the kth test data and the jth point of the center of the ith class cluster; />For the Euclidean distance between the ith feature of the feature vector x of the kth test data and the jth feature of the feature vector y of the center of the first class cluster, the calculation formula is shown as follows,
wherein: f is the number of features;
then, the DTW distance between the test data and the center of each cluster is calculated by the following formula:
wherein:represents the DTW distance of kth test data from the center of the kth cluster, +.>The i-th feature of the kth test data and the i-th feature of the center of the first cluster, respectively,/->The DTW distance of the kth test data from the center of the first cluster is represented by f, which is the number of features, where f is 36, and e is the number of clusters, where e is 5.
And 4, applying a self-adjusting ant colony algorithm based on a correction mechanism, and extracting the spatial characteristics and the temporal characteristics of the ECG signals according to the step 3 to realize arrhythmia classification. The specific process is as follows:
step 4.1, setting the iteration number iter of the model num Number of ants ant =200 num Initial pheromone concentration for each path is τ=0.1, =30;
step 4.2, randomly determining the position of a first ant at the beginning of a first iteration, wherein the ants realize arrhythmia pre-classification through the spatial characteristics and the temporal characteristics obtained in the step 3 and a correction mechanism determined according to Euclidean distance and DTW;
4.2.1, calculating Euclidean distances between the test data and the centers of various clusters through each feature and the formula (22) in the step 3.2;
4.2.2, normalizing the Euclidean distance between the test data and the centers of 5 class clusters through a maximum and minimum normalization formula. Determining a maximum and minimum normalization formula according to formula (23);
wherein x is * Represents normalized data, x represents raw data, min (x) represents the minimum value in x, max (x) represents the maximum value in x,
4.2.3 determining correction values in the correction mechanism by the Euclidean distance normalized in step 4.2.2, determining a correction function according to equation (24)
Wherein:correction values calculated for the kth test data; e (E) * The Euclidean distance after the standardization of the maximum and minimum values; l is the number of types, and the value of l is 5.
4.2.4, determining a class cluster corresponding to the minimum DTW distance according to the DTW distance between the test data obtained in the step 4.2.1 and the centers of the 5 class clusters, wherein the class cluster is represented by a formula (25);
wherein: t is the subscript of the minimum value in the DTW distance between the kth test data and the center of the 5 class clusters; min is the minimum value in the DTW distance between the kth test data and the center of 5 class clusters; index () representation derives a subscript;is the DTW distance of the kth test data from the center of the ith cluster.
4.2.5 determining the correction factor b from the value obtained in step 4.2.4 k ;
4.2.6, determining the classification result according to the values obtained in the steps 4.2.2, 4.2.3 and 4.2.5Formula (27);
And 4.3, updating the pheromone concentration of the path after the pre-classification is realized in the step 4.2. The invention adopts two convergence curves to form a convergence channel to realize dynamic update of the pheromone volatilization coefficient rho, and the rho is controlled to be rho as shown in figures 3, 4 and 5 1 And ρ 2 Equation (28), ρ 1 And ρ 2 All are convergence curves, as shown in fig. 4 and 5;
wherein: ρ 1 Representing the pheromone volatilization coefficient, ρ, determined according to the iteration number 2 Representing the pheromone volatility coefficient determined according to the number of ants.
4.3.1, the invention determines ρ according to the number of iterations 1 Equation (29), determining ρ based on the number of ants 2 Formula (30). Updating rho once by each ant based on the number of ants and the iteration times before the traversal starts;
wherein: c 1 Is of formula ρ 1 Offset constant of c 2 Is of formula ρ 2 Is used to determine the offset constant of the sample, the last two bits of the decimal point are reserved upwards; ter (iter) num Is 200 and gamma is the formula ρ 2 Is less than 0.
4.3.2 updating the pheromone concentration according to formulas (31) (32) (33);
τ ij (t+1)=(1-ρ)τ ij (t)+Δτ ij (31)
wherein: τ ij () For the pheromone on the iterative path (i, j), t represents the iterative times, and ρ is the pheromone volatilization coefficient; Δτ ij Is a pheromone increment;represents the pheromone left by the a-th ant on the path (i, j), n is the number of ants, L a For the total path length traversed by ant a in this iterative process, through (i, j) represents the traversed path [ ]i,j)。
And 4.4, after updating the pheromone in the step 4.3, the ants realize self-adjustment transfer according to the concentration of the pheromone and the path distance, and record the path length of the ants. Determining a transition probability formula according to equation (34);
in the method, in the process of the invention,represents the transition probability of the a-th ant on the path (i, j), eta ij (t) heuristic factor representing the t-th iteration path (i, j), τ ij (t) is the pheromone, eta on the t-th iteration path (i, j) is (t) heuristic factor representing the t-th iteration path (i, s), τ is (t) represents the pheromone on the t-th iteration path (i, s), s represents the sample point, and J is the sample point set which is allowed to be selected by the ant a in the next step; alpha is the relative importance degree of the pheromone, and alpha is more than 0; beta is the relative importance of the heuristic factor, usually +.>d ij Is the path length between (i, j).
4.4.1, redefining a heuristic factor eta calculation formula, and enabling beta to be smaller than 0. Determining a heuristic factor calculation formula according to formula (35);
η ij =d ij (35)
wherein eta ij Heuristic factor, d, representing path (i, j) ij Representing the path length between (i, j),
4.4.2, providing a relation between the relative importance degree alpha of the pheromone and the relative importance degree beta of the heuristic factor according to the negative correlation between eta obtained in the step 4.4.1 and the path between the concentration of the pheromone and (i, j), and determining a relation formula of alpha and beta according to a formula (36);
αβ=m (36)
wherein: m is a relation constant, m is less than 0, and the invention takes m= -20.
4.4.3 at a path length d between (i, j) ij Dynamic updating of beta is achieved for the argument. Determining the relative importance of the heuristic factor beta according to formula (37);
wherein: d, d ij Is the path length between (i, j).
4.4.4, obtaining a formula for dynamically updating alpha according to the concentration of the pheromone determined in the step 4.3.2 in the formula (31), the relational expression of alpha and beta determined in the step 4.4.2 in the formula (36) and the heuristic factor relative importance degree beta determined in the step 4.4.3 in the formula (37), wherein alpha is determined by the volatile coefficients of the pheromone and the pheromone;
wherein: ρ is the pheromone volatility coefficient; τ is a pheromone.
4.4.5 determining [ eta ] according to the formula (37) in the step 4.4.3] β Is related to the distance d, and [ eta ] is determined according to the formula (39)] β ;
Wherein, [ eta ]] β To the power beta of the heuristic factor eta, d represents distance,
4.4.6, determining [ tau ] according to equation (38) in step 4.4.4] α Is correlated with the pheromone concentration τ, and [ τ ] is determined according to the formula (40)] α ;
Step 4.5, after one ant traverses all target points, determining a current path with the maximum pheromone concentration, wherein a random endpoint in two endpoints of the path is used as a starting position of the next ant;
and 4.6, after all ants realize one traversal, sequencing all ant traversal results by using radix sequencing. Sorting the sorting accuracy in all ant traversing results from small to large according to the sorting results and the traversing total path length obtained in the step 4.2 and the step 4.4, sorting the traversing path total length from small to large, and finally selecting the comprehensive optimal result after sorting as the pre-sorting of the traversing;
step 4.7, when passing through the iter num After the iteration, obtaining a pre-class set;
and 5, sorting all the results in the pre-sorting set obtained in the step 4 through a radix sorting algorithm, sorting the sorting accuracy from small to large, sorting the total length of the traversing path from small to large, and finally selecting the sorted comprehensive optimal arrhythmia sorting result.
Comparing with arrhythmia classification results of other models, wherein the test results are shown in table 2;
table 2 arrhythmia classification results test comparison table
The invention advances α=1, β=1, α=1, β=2, α=2, β=3, α=1, β=4 and α=2, β=4 to α=1, β=2And (5) performing experimental comparison. When ρ=0.5, the classification accuracy is remarkably improved after adding the correction mechanism, the overall accuracy is improved by about 10%, the classification error number is reduced by one order of magnitude (10 1 );
The invention tests the effectiveness of the proposed flow rate- ρ increasing method under different α, β. As shown in table 2, the classification accuracy using the flow increasing rate- ρ method is higher than that of most methods, and the flow increasing rate- ρ method is smoother and the convergence rate is faster;
the invention analyzes the effectiveness of using the proposed dynamic update α and β methods at ρ=0.2. Compared to α=1, β=1, α=1, β=2 and α=2, β=4, the overall classification accuracy is improved by 0.8%, 0.2% and 0.2%. As shown in Table 2, the method provided by the invention has absolute advantages in classification accuracy and path length.
The above embodiments are provided to illustrate the technical concept and features of the present invention and are intended to enable those skilled in the art to understand the content of the present invention and implement the same, and are not intended to limit the scope of the present invention. All equivalent changes or modifications made in accordance with the spirit of the present invention should be construed to be included in the scope of the present invention.
Claims (10)
1. An ECG arrhythmia classification method based on a self-adjusting ant colony algorithm of a correction mechanism is characterized by comprising the following steps:
step 1, acquiring ECG signal data;
step 2, preprocessing the ECG signal data obtained in the step 1, finding out the position of an R wave peak value, and dividing the signal data into single periods;
step 3, extracting the characteristics of the signal data preprocessed in the step 2;
step 4, obtaining an arrhythmia pre-classification set through a self-adjustment ant colony aggregation algorithm based on a correction mechanism based on the characteristics obtained in the step 3;
and 5, sorting all the results in the pre-sorting set obtained in the step 4 through a radix sorting algorithm to obtain an optimal arrhythmia sorting result.
2. The method for classifying ECG arrhythmias based on self-adjusting ant colony algorithm of claim 1, wherein,
step 2 comprises the following steps:
step 2.1, removing high-frequency noise and baseline drift from the ECG signal obtained in the step 1 through wavelet transformation, and comparing decomposition results of different wavelet decomposition scales, wherein the R peak value of the ECG signal under the three scales of wavelet decomposition is most obvious, so that an R peak value point is detected under the scales;
step 2.2, searching the maximum value of the R wave based on the decomposition scale obtained in the step 2.1, finding out the point with the slope larger than 0, assigning 1, and assigning 0 to the rest points; searching an R wave minimum value, finding out a point with a slope smaller than 0, assigning 1, and assigning 0 to the rest points;
step 2.3, removing maximum value points and minimum value points with absolute values smaller than the detection threshold according to a preset detection threshold, and finding out the existence interval of suspected R wave crest value points;
and 2.4, performing error detection and omission detection on the R wave to the suspected R wave peak value point determined in the step 2.3, and finally determining the accurate R wave peak value point position.
3. The method for classifying ECG arrhythmias based on self-adjusting ant colony algorithm of claim 2, wherein,
step 2.4 comprises the steps of:
step 2.4.1, when the distance between adjacent suspected R wave crest points is smaller than a×mean (RR), indicating that a false detection phenomenon exists, wherein a is a multiple of the average distance, mean (RR) represents the average distance between adjacent R wave distances, and removing the suspected R wave crest point with the smallest absolute value under the decomposition scale obtained in step 2.1;
step 2.4.2, if the distance between adjacent R wave crest value points is greater than b×mean (RR), indicating that a missing detection phenomenon exists, wherein b is a multiple of the average distance, searching a maximum value and a minimum value pair with the maximum sum of absolute values in two suspected R wave intervals under the decomposition scale obtained in step 2.1, and positioning the zero crossing point of the found maximum value and minimum value pair as the position of the missing detection R wave peak value;
and 2.4.3, cutting the data into single periods according to the determined R wave crest value positions.
4. The method for classifying ECG arrhythmias based on self-adjusting ant colony algorithm of claim 1, wherein,
and 3, extracting time features of the preprocessed ECG signal through a dynamic time planning algorithm so as to improve the perceptibility of the time features of the model.
5. The self-adjusting ant colony algorithm ECG arrhythmia classification method based on correction mechanism of claim 4 wherein:
the dynamic time planning algorithm performs time feature extraction on the preprocessed ECG signal, and specifically comprises the following steps:
first, a DTW distance matrix is determined by the following formula:
wherein:for the cumulative distance matrix of the kth test data and the center of the first cluster, the +.>Representing the distance between the ith point of the kth test data and the jth point of the center of the ith class cluster; />For the Euclidean distance between the ith feature of the feature vector x of the kth test data and the jth feature of the feature vector y of the center of the first class cluster, the calculation formula is shown as follows,
wherein: f is the number of features;
then, the DTW distance between the test data and the center of each cluster is calculated by the following formula:
wherein:represents the DTW distance of kth test data from the center of the kth cluster, +.>The i-th feature of the kth test data and the i-th feature of the center of the first cluster, respectively,/->And the DTW distance between the kth test data and the center of the ith class cluster is represented, f is the number of features, and e is the number of class clusters.
6. The method for classifying ECG arrhythmias based on self-adjusting ant colony algorithm of claim 1, wherein,
step 4 comprises the following steps:
step 4.1: setting model parameters and setting the iteration times of the model as item num The ant amount is ant num The initial pheromone concentration of each path is tau;
step 4.2: randomly determining the position of a first ant at the beginning of a first iteration, and realizing arrhythmia pre-classification by a correction mechanism so as to reduce the classification error rate;
step 4.3: updating the pheromone concentration of the path;
step 4.4: the ants realize self-adjustment transfer according to the concentration of the pheromones and the path distance, and record the path length of the ants;
step 4.5: after one ant traverses all target points, determining a current path with the maximum pheromone concentration, wherein a random one of two end points of the path is used as a starting position of the next ant;
step 4.6: sequencing all ant traversing results after all ants realize one-time traversing;
step 4.7: when passing through iter num And after the iteration, obtaining a pre-class set.
7. The method for ECG arrhythmia classification based on self-regulating ant colony algorithm of claim 6 wherein,
step 4.2 comprises the steps of:
step 4.2.1: calculating Euclidean distance between the test data and the centers of various clusters;
step 4.2.2: the Euclidean distance between the standardized test data and the centers of various clusters;
step 4.2.3: determining a correction value in a correction mechanism based on the normalized Euclidean distance, and correcting a functionThe method comprises the following steps:
wherein:correction values calculated for the kth test data; e (E) * The Euclidean distance after the standardization of the maximum and minimum values; l is the number of types;
step 4.2.4: determining a class cluster corresponding to the minimum DTW distance;
wherein: t is the subscript of the minimum value in the DTW distance between the kth test data and the centers of various clusters; min is the minimum value in the DTW distance between the kth test data and the centers of various clusters; index () representation derives a subscript;the DTW distance between the kth test data and the center of the ith class cluster;
step 4.2.5: determining a correction coefficient b k ;
Wherein: b k Representing a correction coefficient determined after the kth test data is compared with the centers of various clusters;
8. The method for ECG arrhythmia classification based on self-regulating ant colony algorithm of claim 7 wherein,
and 4.3, forming a convergence channel by adopting two convergence curves, and realizing faster and more stable dynamic updating of the pheromone volatilization coefficient rho.
9. The method for ECG arrhythmia classification based on self-regulating ant colony algorithm of claim 8 wherein,
the pheromone volatility coefficient ρ is updated according to the following formula:
wherein ρ is 1 Representing the pheromone volatilization coefficient, ρ, determined according to the iteration number 2 Representing the pheromone volatilization coefficient determined according to the number of ants, and updating rho once by each ant based on the number of ants and the iteration number before the traversal starts;
wherein: ρ 1 (x) Representing the pheromone volatilization coefficient, ρ, determined according to the iteration number x 2 (x) Represents the pheromone volatilization coefficient determined according to the number x of ants, c 1 Is of formula ρ 1 Offset constant of c 2 Is of formula ρ 2 Is a constant of offset in (a); gamma is the formula ρ 2 Coefficient of gamma<0,
Updating the pheromone concentration according to formulas (31), (32) and (33);
τ ij (t+1)=(1-ρ)τ ij (t)+Δτ ij (31)
wherein: τ ij () For the pheromone on the iterative path (i, j), t represents the iterative times, and ρ is the pheromone volatilization coefficient; Δτ ij Is a pheromone increment;represents the pheromone left by the a-th ant on the path (i, j), n is the number of ants, L a For the total path length traversed by ant a in this iteration, through (i, j) represents the traversed path (i, j).
10. The method for classifying ECG arrhythmias based on self-adjusting ant colony algorithm of correction mechanism according to claim 9, wherein the formula of self-adjusting transition probability of ants in step 4.4 is:
in the method, in the process of the invention,represents the transition probability of the a-th ant on the path (i, j), eta ij (t) heuristic factor representing the t-th iteration path (i, j), τ ij (t) is the pheromone, eta on the t-th iteration path (i, j) is (t) heuristic factor representing the t-th iteration path (i, s), τ is (t) represents the t-th iterationPheromones on the generation path (i, s), s represents sample points, and J is a sample point set which is allowed to be selected by the ant a in the next step; alpha is the relative importance of the pheromone, alpha>0; beta is the relative importance of heuristic factors, beta is dynamically updated by taking path length as an independent variable, and the following formula is satisfied:
wherein: d, d ij Is the path length between (i, j).
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310068221.1A CN116089862B (en) | 2023-02-06 | 2023-02-06 | ECG arrhythmia classification method based on correction mechanism and self-adjusting ant colony algorithm |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310068221.1A CN116089862B (en) | 2023-02-06 | 2023-02-06 | ECG arrhythmia classification method based on correction mechanism and self-adjusting ant colony algorithm |
Publications (2)
Publication Number | Publication Date |
---|---|
CN116089862A true CN116089862A (en) | 2023-05-09 |
CN116089862B CN116089862B (en) | 2023-08-15 |
Family
ID=86213695
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310068221.1A Active CN116089862B (en) | 2023-02-06 | 2023-02-06 | ECG arrhythmia classification method based on correction mechanism and self-adjusting ant colony algorithm |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116089862B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117807551A (en) * | 2024-02-29 | 2024-04-02 | 深圳市魔样科技有限公司 | Heart rate abnormality capturing method and system based on intelligent ring |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103927347A (en) * | 2014-04-01 | 2014-07-16 | 复旦大学 | Collaborative filtering recommendation algorithm based on user behavior models and ant colony clustering |
CN105320969A (en) * | 2015-11-20 | 2016-02-10 | 北京理工大学 | A heart rate variability feature classification method based on multi-scale Renyi entropy |
CN106228002A (en) * | 2016-07-19 | 2016-12-14 | 北京工业大学 | A kind of high efficiency exception time series data extracting method based on postsearch screening |
CN106377247A (en) * | 2016-09-10 | 2017-02-08 | 天津大学 | Feature selection-based arrhythmia classification method |
US20190179835A1 (en) * | 2016-11-30 | 2019-06-13 | Business Objects Software Ltd. | Time series analysis using a clustering based symbolic representation |
CN111832537A (en) * | 2020-07-27 | 2020-10-27 | 深圳竹信科技有限公司 | Abnormal electrocardiosignal identification method and abnormal electrocardiosignal identification device |
CN112257518A (en) * | 2020-09-30 | 2021-01-22 | 西安交通大学第二附属医院 | ECG identity recognition method based on WT and WOA-PNN algorithm |
CN112312836A (en) * | 2018-04-27 | 2021-02-02 | 深圳市长桑技术有限公司 | System and method for determining type of arrhythmia |
CN114098743A (en) * | 2020-09-01 | 2022-03-01 | 伯恩森斯韦伯斯特(以色列)有限责任公司 | Arrhythmia classification for cardiac mapping |
-
2023
- 2023-02-06 CN CN202310068221.1A patent/CN116089862B/en active Active
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103927347A (en) * | 2014-04-01 | 2014-07-16 | 复旦大学 | Collaborative filtering recommendation algorithm based on user behavior models and ant colony clustering |
CN105320969A (en) * | 2015-11-20 | 2016-02-10 | 北京理工大学 | A heart rate variability feature classification method based on multi-scale Renyi entropy |
CN106228002A (en) * | 2016-07-19 | 2016-12-14 | 北京工业大学 | A kind of high efficiency exception time series data extracting method based on postsearch screening |
CN106377247A (en) * | 2016-09-10 | 2017-02-08 | 天津大学 | Feature selection-based arrhythmia classification method |
US20190179835A1 (en) * | 2016-11-30 | 2019-06-13 | Business Objects Software Ltd. | Time series analysis using a clustering based symbolic representation |
CN112312836A (en) * | 2018-04-27 | 2021-02-02 | 深圳市长桑技术有限公司 | System and method for determining type of arrhythmia |
CN111832537A (en) * | 2020-07-27 | 2020-10-27 | 深圳竹信科技有限公司 | Abnormal electrocardiosignal identification method and abnormal electrocardiosignal identification device |
CN114098743A (en) * | 2020-09-01 | 2022-03-01 | 伯恩森斯韦伯斯特(以色列)有限责任公司 | Arrhythmia classification for cardiac mapping |
CN112257518A (en) * | 2020-09-30 | 2021-01-22 | 西安交通大学第二附属医院 | ECG identity recognition method based on WT and WOA-PNN algorithm |
Non-Patent Citations (4)
Title |
---|
DAVID MARTENS ET AL.: "Classification with ant colony optimization", IEEE TRANSACTIONS ON EVOLUTIONARY COMPUTATION * |
MIROSLAV BURSA ET AL.: "Modified Ant Colony Clustering Method in Long-term Electrocardiogram Processing", 2007 29TH ANNUAL INTERNATIONAL CONFERENCE OF THE IEEE ENGINEERING IN MEDICINE AND BIOLOGY SOCIETY * |
张丽娜 等: "基于蚁群聚类算法的动脉硬化无创检测", 计算机工程与应用, pages 4 * |
王玲 等: "基于自适应增量学习的时间序列模糊聚类算法", 电子学报, pages 2 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117807551A (en) * | 2024-02-29 | 2024-04-02 | 深圳市魔样科技有限公司 | Heart rate abnormality capturing method and system based on intelligent ring |
CN117807551B (en) * | 2024-02-29 | 2024-05-24 | 深圳市魔样科技股份有限公司 | Heart rate abnormality capturing method and system based on intelligent ring |
Also Published As
Publication number | Publication date |
---|---|
CN116089862B (en) | 2023-08-15 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107822622B (en) | Electrocardiogram diagnosis method and system based on deep convolutional neural network | |
CN109171712B (en) | Atrial fibrillation identification method, atrial fibrillation identification device, atrial fibrillation identification equipment and computer readable storage medium | |
CN106725428B (en) | Electrocardiosignal classification method and device | |
CN109077715B (en) | Electrocardiosignal automatic classification method based on single lead | |
US8290575B2 (en) | Biomedical signal morphology analysis method | |
CN110522444B (en) | Electrocardiosignal identification and classification method based on Kernel-CNN | |
CN109758145B (en) | Automatic sleep staging method based on electroencephalogram causal relationship | |
WO2019214026A1 (en) | Ecg feature extraction method employing pcanet | |
WO2019161611A1 (en) | Ecg information processing method and ecg workstation | |
CN115486855B (en) | Electrocardiogram and heart beat classification method based on QRS complex indefinite cycle superposition | |
CN111772628B (en) | Electrocardiosignal atrial fibrillation automatic detection system based on deep learning | |
CN116089862B (en) | ECG arrhythmia classification method based on correction mechanism and self-adjusting ant colony algorithm | |
Saini et al. | Delineation of ecg wave components using k-nearest neighbor (knn) algorithm: Ecg wave delineation using knn | |
CN113057648A (en) | ECG signal classification method based on composite LSTM structure | |
CN113951893B (en) | Multi-lead electrocardiosignal characteristic point extraction method combining deep learning and electrophysiological knowledge | |
CN111956210B (en) | Electrocardiosignal R wave identification method based on EEMD and signal structure analysis | |
Zifan et al. | Automated ECG segmentation using piecewise derivative dynamic time warping | |
CN116503673B (en) | Arrhythmia identification and detection method and system based on electrocardiogram | |
CN113499079A (en) | Atrial fibrillation detection method in electrocardiogram | |
Zidelmal et al. | Heartbeat classification using support vector machines (SVMs) with an embedded reject option | |
CN113116300A (en) | Physiological signal classification method based on model fusion | |
CN115470832A (en) | Electrocardiosignal data processing method based on block chain | |
CN110169767A (en) | A kind of search method of electrocardiosignal | |
CN113069124A (en) | Electrocardio monitoring method based on CNN-ET model | |
CN117582235A (en) | Electrocardiosignal classification method based on CNN-LSTM model |
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 |