Electrocardiogram classification method and system based on information particles

Abstract

The invention discloses an electrocardiogram classification method and system based on information particles, and relates to the technical field of electrocardiogram classification. The method comprises the steps that electrocardiogram data are obtained and preprocessed, and an electrocardiogram time sequence is obtained; determining a data key point by adopting a local extreme value and a standard deviation; the key points are applied to an information particle reasonable granularity criterion and a slope information criterion in an information increment form, and a granulation process is completed to obtain an information particle form of data; the information particles are divided into different types according to a classification algorithm, candidate prominent information particles are screened out through intra-class similarity and inter-class similarity calculation, and then the information particle with the most significant feature is selected from the same type to serve as the final prominent information particle; and calculating the distance between the highlighted information particles and the time sequence data, and judging the class label of the time sequence according to the minimum distance criterion to obtain a final classification result. The electrocardiogram classification accuracy can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of electrocardiogram classification, and in particular to an electrocardiogram classification method and system based on information granules. Background Art

[0002] Time series refers to a data sequence in which the same statistical indicator is arranged in the order of occurrence of real phenomena. It is widely used in fields such as medicine, astronomy, and economics. For example, when doctors analyze the electrocardiogram data of patients, it helps them make more accurate decisions; by analyzing and forecasting weather data, they can issue warnings for extreme weather; based on stock data, they can observe the rise and fall of the stock market and even predict the stock market situation in the next stage. In-depth research on data in these fields can obtain theoretical support for rapid decision-making and promote social progress and development.

[0003] An electrocardiogram is a graph that records the electrical signals generated by the heart cycle and plots them in a specified format. Electrocardiograms have been widely used in clinical examinations and early warnings of cardiovascular diseases. With the rise of remote monitoring, electrocardiogram analysis is also developing outside the hospital, for sub-healthy people, and for long-term monitoring, which has led to the need for automated diagnosis. Electrocardiogram classification is an important auxiliary means for doctors to diagnose heart disease. Accurate and fast electrocardiogram classification can provide doctors with effective reference information. In addition, electrocardiogram classification has important research significance in wearable electrocardiogram equipment, intensive care, and other aspects. Electrocardiogram data can be regarded as a time series data, and electrocardiogram classification can be regarded as a time series classification problem.

[0004] The patent (CN110179453A) discloses an electrocardiogram classification method based on a convolutional neural network and a long short-term memory network. The method realizes the automatic classification of arrhythmias through multi-lead electrocardiogram data and a convolutional neural network and a long short-term memory network (CNN-LSTM) combined model. On the one hand, multi-lead electrocardiograms contain more information than single-lead electrocardiograms. On the other hand, the CNN-LSTM combined model combines the advantages of CNN and LSTM and has unique advantages in learning spatial data structures and time series structures. Using multi-lead electrocardiogram data to train CNN-LSTM can improve the learning efficiency of the network and the accuracy of electrocardiogram recognition.

[0005] The patent (CN110211684A) discloses an electrocardiogram classification method based on a BP neural network optimized by a genetic algorithm, comprising the following steps: selecting a data set from a database, reading the data set through a Matlab program, and saving the data set in a mat format file; performing crossover mutation on the data set using a genetic algorithm to select the optimal offspring, and passing the obtained result as a training sample into the input layer of the BP neural network; training the BP neural network, wherein the BP neural network comprises an input layer, multiple hidden layer units, and an output layer connected in sequence; reading the electrocardiogram to be classified into a mat format file through a Matlab program, and then inputting the trained BP neural network into the electrocardiogram to be classified for classification.

[0006] The patent (CN112560784A) discloses an electrocardiogram classification method based on a dynamic multi-scale convolutional neural network, which includes the following steps: obtaining an electrocardiogram data set and preprocessing it to obtain an electrocardiogram time series; generating a set of variable-length convolution kernels according to the electrocardiogram time series and randomly initialized convolution kernels; inputting the electrocardiogram time series into a convolution layer using a variable-length convolution kernel to obtain multi-scale time series features; inputting the obtained multi-scale time series features into a max-overtime pooling layer to select the most discriminative features; inputting the most discriminative features into a classification layer to obtain the final classification results, forming a dynamic multi-scale convolutional neural network; and training the constructed dynamic multi-scale convolutional neural network using a cross entropy loss function. The dynamic multi-scale convolutional neural network can learn the length of the convolution kernel through data without relying on prior knowledge, and it can learn sample-specific multi-scale time series features for each electrocardiogram time series.

[0007] In summary, the existing ECG methods mainly focus on deep learning classification methods, which require a large amount of training data for model training and optimization. The training of some small data sets cannot meet the set requirements, resulting in inaccurate classification results. Therefore, there is an urgent need for an ECG classification method that can output more reasonable and accurate results to facilitate medical staff to make quick and accurate decisions. Summary of the invention

[0008] The purpose of the present invention is to provide an electrocardiogram classification method based on information granules, which can effectively improve the accuracy of electrocardiogram classification.

[0009] To achieve the above object, the present invention provides the following solutions:

[0010] An electrocardiogram classification method based on information granules, comprising:

[0011] Obtain an electrocardiogram data set and preprocess it to obtain an electrocardiogram time series;

[0012] Use local extreme values ​​and standard deviations to determine key data points;

[0013] Apply key points in the form of information increments to the reasonable granularity criterion of information granules and the slope information criterion to complete the overall granulation process and divide the data into multiple information granules;

[0014] According to the nearest neighbor classification algorithm, the information particles are divided into several different types, and then the candidate prominent information particles are screened out by calculating the intra-class similarity and inter-class similarity. Then, the information particles with the most significant characteristics are selected from the same type as the final prominent information particles;

[0015] The distance between the prominent information particles and the time series data is calculated, and the class label of the time series is determined according to the minimum distance criterion to obtain the final classification result.

[0016] Optionally, the adopting of local extreme values ​​and standard deviations to determine data key points specifically includes:

[0017] First, using the formula Identify a set of local extreme points. Extreme points can be due to outliers, so they are important for identifying and analyzing anomalies in data. In time series analysis, periodicity and trends in data can be identified by finding extreme points.

[0018] Then, determine the set of standard deviation salient feature points. When the standard deviation is large, it means that the distance between the data points and the average value is far and the distribution is wide. In this case, the data is more variable and may fluctuate greatly. Take the median of the interval with a large standard deviation as the standard deviation salient feature point. The specific steps of selection are as follows:

[0019] Set the sliding window length to l and calculate the standard deviation T of each sliding window i Then determine the window corresponding to the maximum standard deviation among the m adjacent windows, and use the center point of its window as the prominent feature point S s , S s ={x s |x s =med(T j ),T j =aremax[STD(T i )...STD(T i+m )]}, where med() is the median function and STD() is the standard deviation function.

[0020] Finally, the union of the local extreme point set and the standard deviation prominent feature point set is taken as the key point S k , that is, S k =S m ∪S s .

[0021] Optionally, the key points are applied to the reasonable granularity criterion of information granules and the slope information criterion in the form of information increments to complete the overall granulation process and divide the data into a plurality of information granules, specifically including:

[0022] In the process of completing granulation using the reasonable granularity criterion of information granules, given a univariate time series X = {m1, m2, ..., m t}, granulate it into time intervals of T1, T2, .., T h h non-overlapping time series subsequences Y1, Y2, ...Y h For the i-th subsequence Y i , according to the granulation judgment criteria, an interval-based information granule Ω is formed i =[a i ,b i ].use The quality index of the information granule is used to represent the quality index of the information granule. On this basis, an incremental granulation method based on key points is proposed. This method only needs to judge whether the quality of the information granule with multiple new data points ending at the key point is greater than the quality of the original information granule. If it is true, the multiple new data points ending at the key point are assigned to the original information granule, and the representation of the original information granule is adjusted. Otherwise, a new information granule is generated, and then the number of information granules is increased. This granulation method does not require repeated calculations, thereby reducing the amount of calculations and improving the efficiency of granulation. Using the formula Q(Ω i (t+1))<Q(Ω i (t)) is used to determine whether the reasonable granularity criterion of information particles is met.

[0023] In the process of granulation using slope information, the trend of subsequence data is calculated. Assuming that the time series X has different linear trends, the time series can be divided into multiple segments according to the trend. The trend estimation of the sequence can be done using the linear function Y i =k i x+b i , a≤x≤b, where k i and b i is the local slope and intercept of the i-th linear interval. According to the idea of ​​incremental granulation mentioned above, the slope of the information granule is calculated in the form of increments. i (t+1) and the original trend k i (t) is compared. If the trends are consistent, it proves that they belong to the same information particle. Then the slope is updated and k i (t) is changed to k i (t+1), otherwise, a new information particle will be added. i )≠sgn(k i+1 ) to determine whether the trend information criterion is met.

[0024] Optionally, the information particles are divided into several different types according to the nearest neighbor classification algorithm, specifically including:

[0025] Relying on the first neighbor of the information particle to discover the adjacent chain between information particles, use the formula To achieve clustering of information particles. rj1 ) represents the information particle X(Ω i ). By This equation converts each information particle X(Ω i ) is connected to its first neighbor; by To achieve a symmetrical effect, Information particles with the same neighbors [X(Ω i ),X(Ω j )] are also connected, and finally a symmetric and sparse matrix is ​​returned.

[0026] Optionally, the step of screening out candidate prominent information particles by calculating intra-class similarity and inter-class similarity specifically includes:

[0027] The similarity between classes is used to extract high-quality information particles. Since the information particles are generated according to the rules, the generated information particles are of different lengths. Therefore, DTW is used to calculate the similarity of information particles. Because the higher the similarity of information particles of the same type, the better they can represent the characteristics of the class, so the formula is used Calculate the minimum distance between information particles, where I Tra is the index set of information particles in the same time series, Ω i is a specific information particle, Ω j It refers to other information particles of the same type except for specific information particles.

[0028] The key to improving the accuracy of time series classification is to select or extract features that can effectively distinguish different categories. Calculate the distance between different types of information particles. The smaller the similarity, the better. Ter is a set of indicators of information particles in different time series, Ω i is a specific information particle, Ω j For information particles in different classes.

[0029] Optionally, the step of selecting the information grain with the most significant features among the same type as the final prominent information grain specifically includes:

[0030] Divide the intra-class similarity obtained above by the inter-class similarity And select the information particle with the smallest quotient as the candidate information particle.

[0031] According to the type obtained by the nearest neighbor classification, an optimal information particle is selected to represent the type, thereby obtaining the prominent information particle

[0032] Optionally, the calculating of the distance between the salient information granule and the time series data and determining the class label of the time series according to the minimum distance criterion to obtain the final classification result specifically includes:

[0033] The prominent information particles are divided according to categories, the distance between the information particles and X is calculated, the predicted labels are obtained, and the final classification results are obtained.

[0034] A computer device comprises: a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the electrocardiogram classification method based on prominent information particles.

[0035] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0036] The present invention provides an electrocardiogram classification method and system based on prominent information granules, which obtain an electrocardiogram data set and perform preprocessing to obtain an electrocardiogram time series, use local extreme values ​​and standard deviations to determine data key points, apply the key points in the form of information increments to the information granule reasonable granularity criterion and slope information criterion, complete the granulation process of the overall electrocardiogram data set, divide the data into a plurality of information granules, divide the information granules into several different types according to the nearest neighbor classification algorithm, and then screen out candidate prominent information granules by calculating intra-class similarity and inter-class similarity, and then select the information granules with the most significant features in the same type as the final prominent information granules, calculate the distance between the prominent information granules and the time series data, and judge the class label of the electrocardiogram time series according to the minimum distance criterion to obtain the final classification result. The complete time series is granulated into several subsequences using reasonable granularity criteria. These subsequences are called information granules. It can be seen that the formation of information granules has a rigorous theoretical basis, so the information quality is very high. In addition, due to the characteristics of subsequences, they are highly interpretable. Using information granules to classify time series is not only conducive to improving the classification accuracy of time series, but also converting the complete time series into subsequences, realizing dimensionality reduction of the data and reducing the complexity of classification processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0038] Figure 1 A flow chart of the electrocardiogram classification method based on salient information granules provided for this application;

[0039] Figure 2 A specific flow chart of the electrocardiogram classification method based on salient information granules provided in this application. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] The present invention proposes an electrocardiogram classification method and system based on information granules. First, an electrocardiogram data set is obtained and preprocessed to obtain an electrocardiogram time series; secondly, the data key points are determined by using local extreme values ​​and standard deviations, and the key points are applied to the reasonable granularity criterion and slope information criterion of information granules in the form of information increments to complete the overall granulation process and divide the data into the form of multiple information granules; then, the information granules are divided into several different types according to the nearest neighbor classification algorithm, and then the candidate prominent information granules are screened out by calculating the intra-class similarity and inter-class similarity, and then the information granules with the most significant features are selected from the same type as the final prominent information granules; finally, the distance between the prominent information granules and the time series data is calculated, and the class label of the time series is judged according to the minimum distance criterion to obtain the final classification result. The present invention realizes electrocardiogram classification and solves the problem that the existing model fails to effectively grasp the electrocardiogram classification.

[0042] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0043] In an exemplary embodiment, the present application provides an electrocardiogram classification method based on salient information granules. Figure 1 and Figure 2 As shown, the electrocardiogram classification method based on salient information granules provided in an embodiment of the present invention includes the following steps 1 to 7.

[0044] Step 1: Obtain the ECG dataset and preprocess it to obtain the ECG time series.

[0045] An electrocardiogram data set is obtained, and the electrocardiogram data set is preprocessed to obtain n electrocardiogram time series, where n is a positive integer greater than or equal to 1.

[0046] Step 2: Use local extreme values ​​and standard deviations to determine key data points.

[0047] First, determine the set of local extreme points. Extreme points may be caused by outliers, so they are very important for identifying and analyzing anomalies in data. In time series analysis, the periodicity and trend of data can be identified by finding extreme points. The set of local extreme points S m The definition is as follows:

[0048]

[0049] Then, determine the set of standard deviation salient feature points. When the standard deviation is large, it means that the distance between the data points and the average value is far and the distribution is wide. In this case, the data is more variable and may fluctuate greatly. Take the median of the interval with a large standard deviation as the standard deviation salient feature point. The specific steps of selection are as follows:

[0050] Set the sliding window length to l and calculate the standard deviation T of each sliding window i Then determine the window corresponding to the maximum standard deviation among the m adjacent windows, and use the center point of its window as the prominent feature point S s , defined as follows:

[0051] S s ={x s |x s =med(T j ),T j =aremax[STD(T i )...STD(T i+m )]} (2)

[0052] Where med() is the median function and STD() is the standard deviation function.

[0053] Finally, the union of the local extreme point set and the standard deviation prominent feature point set is taken as the key point S k , that is, S k =S m ∪S s .

[0054] Step 3: Apply the key points in the form of information increments to the reasonable granularity criterion of information granules and the slope information criterion to complete the overall granulation process and divide the data into multiple information granules.

[0055] Given a univariate time series X = {m1,m2,...,m t}, granulate it into time intervals T1, T2, ..., T h h non-overlapping time series subsequences Y1,Y2,...,Yh For the i-th subsequence Y i , according to the granulation judgment criteria, an interval-based information granule Ω is formed i =[a i ,b i ]. In the process of information particle formation, two conditions must be followed: one is rationality, and the other is specificity.

[0056] Rationality: To ensure information granularity i In terms of experimental rationality, the more data contained in the particles, the more fully they can reflect the characteristics of the time series, which can be achieved through Ω i quantified by the cardinality, i.e. card{m q |m q ∈Ω i}, which means it falls on Ω i The amount of data in the range.

[0057] Concreteness: The requirement for information particles to have a clear semantic interpretation, Ω i Should be as specific as possible, the interval m(Ω i )=|b i -a i The length of | should be as small as possible so that the information particle Ω can be better quantified i Therefore, the smaller the particle length, the higher the specificity.

[0058] It can be seen that the two requirements described above are conflicting. The method to reasonably deal with these two conflicting requirements is to use function definition. This function is defined as the product of two conflicting functions, that is, f = f1·f2, where f1 is {m q |m q ∈Ω i}Base quantization increasing function, f2 is the interval m(Ω i )=|b i -a i | length decreasing function. f1 and f2 are defined as follows:

[0059] f1(u)=u (3)

[0060] f2(u)=exp(-αu) (4)

[0061] Transformed into the form of f = f1·f2, the quality index of the information particle is defined as follows:

[0062]

[0063] Where Ω i The volume of Ω i The covariance matrix of is determined and is defined as:

[0064]

[0065] where α g Ω i The covariance matrix Σ i The g-th eigenvalue of Σ i It can be calculated as:

[0066]

[0067] where z i Represents Ω i The number of data points in X Ωi Represents Ω i The data matrix, β i Ω i The mean of E represents a matrix whose elements are all equal to 1. Therefore, from the perspective of granulation quality, the information granule Ω i The number of data points z can be used i ,Ω i The mean value of β i and Ω i The covariance matrix Σ i Completely expressed.

[0068] On this basis, an incremental granulation method based on key points is proposed. This method only needs to judge whether the mass of the information granules of multiple new data points added to the key point is greater than the mass of the original information granules. If so, the multiple new data points ending at the key point are assigned to the original information granules, and the representation of the original information granules is adjusted. Otherwise, a new information granule is generated, and then the number of information granules is increased. This granulation method does not require repeated calculations, thus reducing the amount of calculations and improving the efficiency of granulation. This granulation strategy is divided into two steps: parameter update and granulation formation.

[0069] Parameter update: When time t+1, read the second key point position t+s, then the new s data points {x t+1 ,x t+2 ,...,x t+s} is used to update parameters.

[0070] Current information particle Ω i At time t, the expression (β i (t),Σ i (t),z i (t)) can be updated recursively in the following steps. First, by updating Ω i Number of data points in :

[0071] z i (t+1)=z i (t)+s (8)

[0072] Input data points and information granules Ω i The difference between the means is calculated as:

[0073] d i (t+1)=x t+s -β i (t) (9)

[0074] Then, the mean is updated as follows:

[0075]

[0076] Next, the statistics of the covariance matrix can be obtained:

[0077] S i (t+1)=S i (t)+d i (t+1)(x t+s -β i (t+1)) T (11)

[0078] Finally, the covariance matrix is ​​updated as follows:

[0079]

[0080] In order to make the obtained time series information granules more informative and interpretable, the quality index of the information granules is an important metric. t+1 ,x t+2 ,...,x t+s}, the mass of the information particle with the new sample Q(Ω i (t+1)) and the mass Q(Ω of the original information particle i (t)) is compared. When the newly obtained information granule is smaller than the original information granule, it means that the addition of the new sample has caused some loss in quality, so it is not suitable to be added to the original information granule. At this time, a new information granule needs to be formed, otherwise the newly added multiple data points are assigned to the current information granule. Thus, the granulation judgment criterion is obtained:

[0081] Q(Ω i (t+1))<Q(Ω i (t)) (13)

[0082] Calculate the trend of subsequence data, assuming that the time series X has different linear trends, that is, the time series can be divided into multiple segments according to the trend. The trend estimation of the sequence can be described by the following linear function:

[0083] Y i =k i x+b i,a≤x≤b (14)

[0084] where k i and b i are the local slope and intercept of the ith linear interval, k i and b i It can be determined using the least squares method.

[0085] According to the idea of ​​incremental granulation mentioned above, the slope of the information granule is calculated in the form of increments. i (t+1) and the original trend k i (t) is compared. If the trends are consistent, it proves that they belong to the same information particle. Then the slope is updated and k i (t) is changed to k i (t+1), otherwise, a new information particle will be added. The trend judgment criteria are as follows:

[0086] sgn(k i )≠sgn(k i+1 ) (15)

[0087] Step 4: Divide the information particles into several different types according to the nearest neighbor classification algorithm.

[0088] The first neighbor of the information particle is used to discover the adjacency chain between the information particles, so as to achieve clustering of the information particles. Based on the integer index of the first neighbor, an adjacency matrix can be directly defined:

[0089]

[0090] Among them, Represents information particle X(Ω i ). By This equation converts each information particle X(Ω i ) is connected to its first neighbor; by To achieve a symmetrical effect, Information particles with the same neighbors [X(Ω i ),X(Ω j )] are also connected, and finally a symmetric and sparse matrix is ​​returned.

[0091] Step 5: Screen out candidate salient information particles by calculating intra-class similarity and inter-class similarity.

[0092] The similarity between classes is used to extract high-quality information particles. Since information particles are generated according to rules, the generated information particles are of unequal length. Therefore, this paper uses DTW to calculate the similarity of information particles. Because the higher the similarity of information particles of the same type, the better they can represent the characteristics of the class, so it is necessary to calculate the minimum distance between information particles. The calculation formula is as follows:

[0093]

[0094] Among them I Tra is the index set of information particles in the same time series, Ω i is a specific information particle, Ω j It refers to other information particles of the same type except for specific information particles.

[0095] The key to improving the accuracy of time series classification is to select or extract features that can effectively distinguish different categories. When calculating with different types of information particles, the smaller the similarity, the better. The calculation formula is as follows:

[0096]

[0097] Among them I Ter is a set of indicators of information particles in different time series, Ω i is a specific information particle, Ω j For information particles in different classes.

[0098] Step 6: Select the most significant information particles of the same type as the final prominent information particles.

[0099] Divide the intra-class similarity obtained above by the inter-class similarity and select the information particle with the smallest quotient as the candidate information particle. The calculation formula is as follows:

[0100]

[0101] From the types obtained by the nearest neighbor classification, an optimal information particle is selected to represent the type, thereby obtaining a prominent information particle. The calculation formula is as follows:

[0102]

[0103] Step 7: Calculate the distance between the prominent information particles and the time series data, and determine the class label of the time series according to the minimum distance criterion to obtain the final classification result.

[0104] The prominent information particles are divided according to categories, the distance between the information particles and X is calculated, the predicted labels are obtained, and the final ECG classification results are obtained.

[0105] The present invention and other methods are used to classify the electrocardiogram data set, and the classification accuracy results are compared in Tables 1 and 2.

[0106] Table 1 Comparison of classification accuracy of the present invention and other methods on the AtrialFibrillation dataset

[0107] method The present invention DKN SVP-T MR-PETSC SMATE MTSC_FF Accuracy 0.667 0.467 0.400 0.400 0.133 0.533

[0108] Table 2 Comparison of classification accuracy of the present invention and other methods on the StandWalkJump dataset

[0109] method The present invention DKN SVP-T MR-PETSC SMATE MTSC_FF Accuracy 0.667 0.533 0.467 0.400 0.533 0.467

[0110] The present invention obtains an electrocardiogram time series by preprocessing the acquired electrocardiogram data set, and uses the most prominent information granule set of the extracted electrocardiogram time series for classification, which improves the accuracy of the electrocardiogram while being interpretable, and can provide a certain reference for medical staff to make reasonable and accurate decisions. The present invention proposes the following three innovations in the classification of electrocardiogram time series: 1. A key point extraction method based on local extreme values ​​and prominent standard deviations is proposed. The key points extracted by this method can effectively represent the prominent features of data changes and fluctuations, better express the changing laws of time series, carry more high-quality information, and are conducive to improving the classification accuracy of electrocardiogram time series. 2. An incremental information granulation strategy is proposed. This strategy combines the granulation judgment criterion and the trend judgment criterion to make it have more significant information and stronger interpretability. At the same time, this strategy only performs incremental judgment on the extracted key points, which improves the efficiency of the granulation process. 3. A prominent information granule screening method is proposed. This method uses intra-class similarity and inter-class similarity to screen out prominent information granules, thereby obtaining significant features that effectively represent various types of time series data. In this process, the existence of similar information granules is effectively avoided and the number of information granules is reduced.

[0111] Compared with the prior art, the present invention has the following advantages: 1. The existing method requires feature selection in the network model, and the obtained features lack interpretability. The present invention uses a granulation strategy in the feature selection of the electrocardiogram data set, so that the obtained subsequence has stronger interpretability and the effect of feature selection is better. 2. The existing method requires continuous optimization of the model, and the model cannot be well optimized on a small data set, resulting in poor final classification effect. The present invention uses a screening structure in feature selection, and selects high-quality feature subsequences through inter-class similarity and intra-class similarity, thereby improving the accuracy of the classification results and outputting more accurate electrocardiogram classification results.

[0112] In an exemplary embodiment, the present invention further provides a computer device, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0113] In an exemplary embodiment, the present invention further provides a computer-readable storage medium storing a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0114] In an exemplary embodiment, the present invention further provides a computer program product, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0115] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in the present invention are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant legal provisions.

[0116] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided by the present invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0117] The database involved in each embodiment provided by the present invention may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided by the present invention may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0118] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0119] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. An electrocardiogram classification method based on information granules, characterized in that: include: Obtain an electrocardiogram data set and preprocess it to obtain an electrocardiogram time series; Use local extreme values ​​and standard deviations to determine key data points; Apply key points in the form of information increments to the reasonable granularity criterion of information granules and the slope information criterion to complete the overall granulation process and divide the data into multiple information granules; According to the nearest neighbor classification algorithm, the information particles are divided into several different types, and then the candidate prominent information particles are screened out by calculating the intra-class similarity and inter-class similarity. Then, the information particles with the most significant characteristics are selected from the same type as the final prominent information particles; The distance between the prominent information particles and the time series data is calculated, and the class label of the time series is determined according to the minimum distance criterion to obtain the final ECG classification result.

2. The electrocardiogram classification method based on salient information granules according to claim 1 is characterized in that: Use local extreme values ​​and standard deviations to determine key data points; specifically: First, using the formula Identify a set of local extreme points. Extreme points can be due to outliers, so they are important for identifying and analyzing anomalies in data. In time series analysis, periodicity and trends in data can be identified by finding extreme points. Then, determine the set of standard deviation salient feature points. When the standard deviation is large, it means that the distance between the data points and the average value is far and the distribution is wide. In this case, the data is more variable and may fluctuate greatly. Take the median of the interval with a large standard deviation as the standard deviation salient feature point. The specific steps of selection are as follows: Set the sliding window length to l and calculate the standard deviation T of each sliding window i Then determine the window corresponding to the maximum standard deviation among the m adjacent windows, and use the center point of its window as the prominent feature point S s , S s ={x s |x s =med(T j ),T j =aremax[STD(T i )...STD(T i+m )]}, where med() is the median function and STD() is the standard deviation function. Finally, the union of the local extreme point set and the standard deviation prominent feature point set is taken as the key point S k , that is, S k =S m ∪S s .

3. The electrocardiogram classification method based on salient information granules according to claim 1 is characterized in that: Apply the key points in the form of information increment to the reasonable granularity criterion of information granules and the slope information criterion to complete the overall granulation process and divide the data into multiple information granules, including: In the process of completing granulation using the reasonable granularity criterion of information granules, given a univariate time series X = {m1, m2, ..., m t }, granulate it into time intervals T1, T2, ..., T h h non-overlapping time series subsequences Y1,Y2,...,Y h For the i-th subsequence Y i , according to the granulation judgment criteria, an interval-based information granule Ω is formed i =[a i ,b i ].use The quality index of the information granule is used to represent the quality index of the information granule. On this basis, an incremental granulation method based on key points is proposed. This method only needs to judge whether the quality of the information granule with multiple new data points ending at the key point is greater than the quality of the original information granule. If it is true, the multiple new data points ending at the key point are assigned to the original information granule, and the representation of the original information granule is adjusted. Otherwise, a new information granule is generated, and then the number of information granules is increased. This granulation method does not require repeated calculations, thereby reducing the amount of calculations and improving the efficiency of granulation. Using the formula Q(Ω i (t+1))<Q(Ω i (t)) is used to determine whether the reasonable granularity criterion of information particles is met. In the process of granulation using slope information, the trend of subsequence data is calculated. Assuming that the time series X has different linear trends, the time series can be divided into multiple segments according to the trend. The trend estimation of the sequence can be done using the linear function Y i =k i x+b i , a≤x≤b, where k i and b i is the local slope and intercept of the i-th linear interval. According to the idea of ​​incremental granulation mentioned above, the slope of the information granule is calculated in the form of increments. i (t+1) and the original trend k i (t) is compared. If the trends are consistent, it proves that they belong to the same information particle. Then the slope is updated and k i (t) is changed to k i (t+1), otherwise, a new information particle will be added. i )≠sgn(k i+1 ) to determine whether the trend information criterion is met.

4. The electrocardiogram classification method based on salient information granules according to claim 1 is characterized in that: According to the nearest neighbor classification algorithm, information particles are divided into several different types, including: Relying on the first neighbor of the information particle to discover the adjacent chain between information particles, use the formula To achieve clustering of information particles. Represents information particle X(Ω i ). By This equation converts each information particle X(Ω i ) and its first neighbor; by To achieve a symmetrical effect, Information particles with the same neighbors [X(Ω i ),X(Ω j )] are also connected, and finally a symmetric and sparse matrix is ​​returned.

5. The electrocardiogram classification method based on salient information granules according to claim 1 is characterized in that: By calculating the intra-class similarity and inter-class similarity, candidate prominent information particles are screened out, including: The similarity between classes is used to extract high-quality information particles. Since the information particles are generated according to the rules, the generated information particles are of different lengths. Therefore, DTW is used to calculate the similarity of information particles. Because the higher the similarity of information particles of the same type, the better they can represent the characteristics of the class, so the formula is used Calculate the minimum distance between information particles, where I Tra is the index set of information particles in the same time series, Ω i is a specific information particle, Ω j It refers to other information particles of the same type except for specific information particles. The key to improving the accuracy of time series classification is to select or extract features that can effectively distinguish different categories. Calculate the distance between different types of information particles. The smaller the similarity, the better. Ter is a set of indicators of information particles in different time series, Ω i is a specific information particle, Ω j For information particles in different classes.

6. The electrocardiogram classification method based on salient information granules according to claim 1, characterized in that: The most prominent information particles are selected from the same type as the final prominent information particles, including: Divide the intra-class similarity obtained above by the inter-class similarity And select the information particle with the smallest quotient as the candidate information particle. According to the type obtained by the nearest neighbor classification, an optimal information particle is selected to represent the type, thereby obtaining the prominent information particle 7. The electrocardiogram classification method based on salient information granules according to claim 1 is characterized in that: Calculate the distance between the prominent information particles and the time series data, and determine the class label of the time series according to the minimum distance criterion to obtain the final classification result, which includes: The prominent information particles are divided according to categories, the distance between the information particles and X is calculated, the predicted labels are obtained, and the final classification results are obtained.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the electrocardiogram classification method based on prominent information particles as described in any one of claims 1 to 7.

Images