Novel single-lead atrial fibrillation classification and identification method, system, medium and equipment

Abstract

The invention belongs to the technical field of biomedical signal processing, and provides a novel single-lead atrial fibrillation classification and identification method aiming at the technical problem that the diagnosis efficiency is not high due to the accuracy reason of an electrocardiogram recorded by current single-conductor portable equipment. The method comprises the following steps: extracting features from a plurality of fields from a single-lead electrocardiogram signal, including time, frequency, time frequency, phase space and meta-level features, and classifying and identifying the extracted features by another random forest classifier by utilizing a feature selection method based on the random forest classifier to obtain an atrial fibrillation classification and identification result. The method has the advantages of being easy to implement, high in accuracy, good in portability and the like.

Description

Technical Field

[0001] The present invention belongs to the technical field of biomedical signal processing, and in particular relates to a novel single-lead atrial fibrillation classification and identification method, system, medium and equipment. Background Art

[0002] Atrial fibrillation is a common cardiac arrhythmia that can lead to serious complications. Prompt electrocardiogram (ECG) recording (at least 30 seconds) is a diagnostic and effective method. In ECG recordings, the absence of P waves and irregular spacing of QRS complexes are the main signs of atrial fibrillation. However, although traditional multi-lead electrocardiogram (ECG) analysis methods are highly accurate, they are limited in portability and application scenarios. ECGs recorded by single-conductor portable devices currently have low diagnostic efficacy due to their low accuracy. Summary of the Invention

[0003] The purpose of the present invention is to overcome the deficiencies of the prior art and to provide a novel single-lead atrial fibrillation classification and identification method, system, medium and device.

[0004] The first object of the present invention is to provide a novel single-lead atrial fibrillation classification and identification method, comprising the following steps:

[0005] S1. Acquire single-lead electrocardiogram signal;

[0006] S2. preprocessing the acquired single-lead electrocardiogram signal to obtain a processed signal;

[0007] S3. Extracting basic time-domain features and frequency-domain features from the processed signal to obtain extracted features, and constructing a feature matrix using seven features from the extracted features;

[0008] S4. A dataset is constructed by constructing a feature matrix and a corresponding label vector, and a base-level classifier is trained using the dataset to output meta-features.

[0009] S5, combining the remaining features in the extracted features with the meta-features to obtain a mixed feature vector;

[0010] S6. The model is trained and predicted by combining the mixed feature vector with the random forest algorithm to obtain the prediction results, and the prediction results are classified to obtain the atrial fibrillation classification and recognition results.

[0011] Preferably, in step S2, the preprocessing includes performing baseline correction and denoising on the single-lead electrocardiogram signal.

[0012] Preferably, in step S3, the basic level time domain features include RR interval, PR interval, RR interval mean, RR interval standard deviation, RR interval variation coefficient, P wave peak value, P wave mean value, P wave amplitude variance, T wave peak value, T wave mean value, P wave correlation coefficient and Renyi entropy feature;

[0013] The basic frequency domain features include heartbeat feature 1 and heartbeat feature 2.

[0014] Preferably, in step S3, the seven features are RR interval, RR interval variation coefficient, PR interval, P wave correlation coefficient, Renyi entropy feature, heartbeat feature 1 and heartbeat feature 2.

[0015] Preferably, in step S4, the base-level classifier is composed of a linear discriminant analysis classifier and a random forest classifier;

[0016] The expression of the posterior probability of the base-level classifier is:

[0017]

[0018] Where C LDA (C k |X) for each category C k Below, the posterior probability of the linear discriminant analysis classifier; C RF (C k |X) for each category C k Below, the posterior probability of the random forest classifier.

[0019] Preferably, in step S4, the meta-features are the posterior probability mean and the posterior probability standard deviation of the basic level classifier.

[0020] Preferably, in step S6, the expression of the prediction result is:

[0021] y pred，final =RF[X mixed , Y initial ]

[0022] In the formula, RF means training with random forest algorithm, X mixed is the mixed eigenvector, Y initial is the training result, y pred，final To predict the results;

[0023] The expression of atrial fibrillation classification and recognition results is:

[0024]

[0025] Where Abnormal Heart Rate is the atrial fibrillation classification result, y pred，final For the prediction results.

[0026] A second object of the present invention is to provide a novel single-lead atrial fibrillation classification and identification system, which is implemented by the above-mentioned method, including:

[0027] A signal acquisition module, used for acquiring a single-lead electrocardiogram signal;

[0028] A preprocessing module, used for preprocessing the acquired single-lead electrocardiogram signal to obtain a processed signal;

[0029] The feature extraction module is used to extract the basic time domain features and frequency domain features from the processed signal to obtain extracted features, and construct a feature matrix using the seven features in the extracted features;

[0030] The model training module is used to construct a dataset using a feature matrix and corresponding label vectors, and use the dataset to train a base-level classifier to output meta-features.

[0031] A hybrid feature vector construction module is used to combine the remaining basic features and meta-features in the extracted features to obtain a hybrid feature vector;

[0032] The classification and recognition module is used to train and predict the model by combining mixed feature vectors with the random forest algorithm to obtain prediction results, classify the prediction results, and obtain atrial fibrillation classification and recognition results.

[0033] The third object of the present invention is to provide a computer-readable storage medium, which includes a stored computer program; wherein, when the computer program is run, it controls the device where the computer-readable storage medium is located to perform the steps of the new single-lead atrial fibrillation classification and identification method as described in any one of claims 1 to 7.

[0034] The fourth object of the present invention is to provide a terminal device comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor implements the novel single-lead atrial fibrillation classification and identification method as described in any one of claims 1 to 7 when executing the computer program.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] (1) Reduce the number of patches: Traditional electrocardiograms usually require 12 lead electrodes to fully monitor the heart's electrical activity, while the present invention only requires 3 patches to complete the acquisition of electrical signals and the identification of atrial fibrillation. While eliminating limb leads, it also reduces the number of chest leads, significantly reducing the complexity of the device.

[0037] (2) Simple and timely operation: Traditional ECG equipment in clinical practice usually requires professionals to perform complex connections and placement. However, the three-patch design of the present invention can greatly reduce the number of operation steps. The user only needs to stick the patch on the chest, and the device will automatically collect data and transmit it via Bluetooth. The user can access the collected ECG data at any time without the assistance of professionals. This simplified operation process can improve the convenience of use, especially for patient self-testing or telemedicine scenarios.

[0038] (3) The present invention uses a single-lead electrocardiogram signal, eliminating the need for complex multi-lead equipment and simplifying the data acquisition process. Furthermore, the present invention is based on random forest feature selection and classifier design, eliminating the need for complex hyperparameter tuning and resulting in high model training efficiency. Furthermore, single-lead equipment directly reduces hardware complexity, and the preprocessing steps (baseline correction, denoising) employ classical algorithms (such as bandpass filtering) with low computational resource requirements. The parallelization properties of random forests make them naturally suitable for efficient implementation in feature selection and classification, eliminating the need for complex iterative optimization.

[0039] (4) The present invention extracts 12 basic features and 2 meta-features from time, frequency, time-frequency, phase space and meta-level features from a single-lead electrocardiogram signal, covering the multi-dimensional information of the signal. The present invention utilizes a two-level classifier architecture, specifically generating meta-features through the first-level basic classifier (linear discriminant analysis + random forest), and the second-level random forest fuses the basic features and meta-features to enhance the classification robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 The performance comparison results of various methods are shown in the figure. DETAILED DESCRIPTION

[0041] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described 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 making any creative efforts shall fall within the scope of protection of the present invention.

[0042] Example 1

[0043] The embodiment of the present invention provides a novel single-lead atrial fibrillation classification and identification method, which specifically includes the following steps:

[0044] S1. Obtain a single-lead ECG signal. First, place three electrodes on the patient's chest and use a three-lead handheld ECG device to collect ECG signals at a sampling rate of 250 Hz to obtain a single-lead ECG signal. The expression of the single-lead ECG signal is:

[0045] f(t)=ECG signal at time t

[0046] Where f(t) is the single-lead electrocardiogram signal;

[0047] S1. Preprocess the acquired single-lead electrocardiogram signal to obtain a processed signal; this is specifically achieved through baseline correction and denoising.

[0048] In the embodiment of the present invention, DC drift in a single-lead electrocardiogram signal is eliminated by baseline correction. The expression for baseline correction is:

[0049] f Corrected (t) = f(t) - mean(f(t))

[0050] Where f(t) is the single-lead ECG signal, mean(f(t)) is the mean of the single-lead ECG signal, and f Corrected (t) is the corrected single-lead ECG signal;

[0051] In the embodiment of the present invention, power frequency interference and high-frequency noise are removed by a bandpass filter. The noise removal expression of the bandpass filter is:

[0052]

[0053] Where H(f) is the filter function, f(t) is the single-lead ECG signal;

[0054] S2. Extracting basic time-domain features and frequency-domain features from the processed signal to obtain extracted features, and constructing a feature matrix using seven features from the extracted features;

[0055] In the embodiment of the present invention, the basic level time domain features include RR interval, PR interval, RR interval mean, RR interval standard deviation, RR interval variation coefficient, P wave peak value, P wave mean value, P wave amplitude variance, T wave peak value, T wave mean value, P wave correlation coefficient and Renyi entropy feature;

[0056] The basic frequency domain features include heartbeat feature 1 and heartbeat feature 2.

[0057] The present invention is achieved by the following method when specifically extracting:

[0058] (1) Basic-level time-domain feature extraction

[0059] Extraction of R peak, P peak and T wave peak: Detection of R wave, P wave and T wave is performed on the processed ECG signal H(f). Specifically, the waveform characteristics of the processed ECG signal H(f) are analyzed by the R wave detection algorithm to identify the position of the R wave. Then, the position of the R wave is marked on the processed signal (ECG signal H(f)) and the position of the i-th R wave is recorded. The position of the i-th R wave is t R，i = / arg max(sigbal(t)), and identify the positions of P waves and T waves on the processed signal through template matching, adaptive threshold method, morphology-based detection algorithm and other algorithms, mark the position of the P wave, and the position of the i-th P wave is t P，i = / arg max(sigbal(t)), and mark the position of the T wave, and the position of the i-th T wave is t T，i = / arg max(sigbal(t)). Based on the extracted R peak, P peak, and T peak, basic time domain features can be extracted and calculated. The expressions of each basic time domain feature are as follows:

[0060] The expression of RR interval is:

[0061] RR i =t R，i+1 -t R，i

[0062] Where RR i is the i-th RR interval, t R，i is the position of the i-th R wave, t R，i+1 is the position of i+1 R wave;

[0063] The expression of the PR interval is:

[0064] PR i =t P，i+1 -t R，i

[0065] Where PR i is the i-th PR interval, t R，i is the position of the i-th R wave, t P，i+1 is the position of the i+1th P wave;

[0066] The expression for the mean RR interval is:

[0067]

[0068] Where N is the number of RR intervals, RR mean is the average value of RR interval, RR i is the i-th RR interval;

[0069] The expression for the standard deviation of the RR interval is:

[0070]

[0071] Where, σ RR represents the standard deviation of the RR interval, RR i is the i-th RR interval, RR mean is the average value of RR intervals, and N is the number of RR intervals;

[0072] The expression of the coefficient of variation of the RR interval is:

[0073]

[0074] Where, CV RR represents the coefficient of variation of the RR interval, σ RR represents the standard deviation of the RR interval, RR mean is the average value of the RR interval;

[0075] The expression for the peak value of the P wave is:

[0076] P i =max(P(t))-min(P(t))

[0077] Where, P i is the peak value of the i-th P wave, P(t) is the P wave signal part, max(P(t)) is the highest signal of the P wave, and min(P(t)) is the lowest signal of the P wave;

[0078] The expression for the average value of the P wave is:

[0079]

[0080] Where N is the number of P waves, P mean is the average value of P wave, P i is the peak value of the i-th P wave;

[0081] The expression for the P wave amplitude variance is:

[0082]

[0083] Where Var(P) is the variance of the P wave amplitude, N is the number of P waves, and P mean is the average value of P wave, P i is the peak value of the i-th P wave;

[0084] The expression of the T wave peak is:

[0085] T i =max(T(t))-min(T(t))

[0086] Where, T i is the peak value of the i-th T wave, T(t) is the signal part of the T wave, max(T(t)) is the highest signal of the T wave, and min(T(t)) is the lowest signal of the T wave;

[0087] The expression for the average value of the T wave is:

[0088]

[0089] Where, T mean is the average value of T wave, N is the number of T waves, T i is the peak value of the i-th T wave;

[0090] The expression of the P wave correlation coefficient is:

[0091]

[0092] Where, X i represents the ith P wave peak, Y i represents the duration of the i-th P wave, and X i and Y i The average value of

[0093] For the P wave of the electrocardiogram signal, the P wave can be summarized into the following three probabilities:

[0094] P1: P wave peak probability. P1 means within 300s, P i Probability <0.1;

[0095] P2: probability of P wave duration. P2 represents the probability that the P wave duration is less than 0.12 within 300 seconds;

[0096] P3: The probability of change in the shape of the P wave. P3 indicates the probability of asymmetric P waves within 300 seconds.

[0097] According to the above three probabilities, the expression of the Rényi entropy characteristic can be obtained as follows:

[0098]

[0099] Where H2(x) is the Rényi entropy characteristic.

[0100] (2) Basic frequency domain feature extraction

[0101] The time domain signal f(t) is converted into the frequency domain signal F(t) using the Fast Fourier Transform (FFT). The expression of the frequency domain signal F(f) is:

[0102] F(f)=FFT(f(t))

[0103] Then the expression of power spectral density is:

[0104]

[0105] Where |F(f)| is the amplitude in the frequency domain, N is the total number of samples, and p(f) is the power density at frequency f;

[0106] Then, within different frequency ranges (Hz), heartbeat feature 1 and heartbeat feature 2 are calculated for each heartbeat. Heartbeat feature 1 represents the power of the low-frequency component, and heartbeat feature 2 represents the power of the high-frequency component. The calculation formulas for heartbeat feature 1 and heartbeat feature 2 are as follows:

[0107]

[0108] Where pf1 is heartbeat feature 1, pf2 is heartbeat feature 2;

[0109] The RR interval, RR interval variation coefficient, PR interval, P wave correlation coefficient, Rényi entropy feature, heartbeat feature 1 and heartbeat feature 2 in the extracted features are used to form a feature matrix X. The expression of the feature matrix X is as follows:

[0110]

[0111] Where n is the number of samples, f i，j is the jth feature of the i-th sample, i = 1, 2…n; j = 1, 2…7; specifically, each row corresponds to 7 features, such as f 1,1 、f 1,2 、f 1,3 、f 1,4 、f 1,5 、f 1,6 、f 1,7 They correspond to the RR interval, RR interval variation coefficient, PR interval, P wave correlation coefficient, Rényi entropy feature, heartbeat feature 1, and heartbeat feature 2 of the first sample respectively;

[0112] S3. Construct a dataset using the feature matrix and the corresponding label vector (the true category of each sample: presence or absence of atrial fibrillation), input the dataset into a base-level classifier to train the base-level classifier, and output meta-features.

[0113] The basic classifier in the embodiment of the present invention is composed of a linear discriminant analysis classifier, a quadratic discriminant analysis classifier, and a random forest classifier; the basic classifier is equally weighted, and the expression of the posterior probability of the basic classifier is:

[0114]

[0115] Where CLDA (C k |X) for each category C k Below, the posterior probability of the linear discriminant analysis classifier; C RF (C k |X) for each category C k Below, the posterior probability of the random forest classifier.

[0116] C LDA (C k |X) is expressed as:

[0117]

[0118] In the formula, P(C k |X) is in category C k Under these conditions, the probability density function of feature X, P(C k ) is Category C k The prior probability of ; P(X) is the total probability density function of feature X;

[0119] C RF (C k |X) is expressed as:

[0120]

[0121] Where, T t (X) represents the prediction result of the t-th tree, I is the indicator function, when T t (X) is Category C k When C RF (C k |X) is 1, otherwise it is 0;

[0122] The meta-features in the embodiment of the present invention are the posterior probability mean and the posterior probability standard deviation of the basic classifier, and the expressions of the posterior probability mean and the posterior probability standard deviation are:

[0123]

[0124] Where, P mean (C k ) is the mean of the posterior probability, P(C k |X i ) is the i-th sample pair category C k The posterior probability of N is the total number of samples, P std (C k ) is the standard deviation of the posterior probability.

[0125] S4. Combine the remaining basic features and meta-features in the extracted features to obtain a mixed feature vector. The expression of the mixed feature vector is:

[0126] X mixed

[0127] =[P i , P mean ,Var(P),T i , T mean , σ RR ,RR mean , P mean (C1),P mean (C2),P mean (C3),

[0128] P std (C1),P std (C2),P std (C3)]

[0129] Where, P i The peak value of the i-th P wave, P mean is the mean value of P wave, Var(P) is the variance of P wave amplitude, T i is the peak value of the i-th T wave, T mean is the average value of T wave, σ RR is the standard deviation of the RR interval, RR mean is the average value of RR interval, P mean (C1) is the mean posterior probability of the normal state, P mean (C2) is the mean posterior probability of atrial fibrillation, P mean (C3) is the unknown posterior probability mean, P std (C1) is the standard deviation of the posterior probability under normal conditions, P std (C2) is the standard deviation of the posterior probability of atrial fibrillation, P std (C3) is the unknown posterior probability standard deviation;

[0130] S5. The clinical data is trained and predicted by combining the mixed feature vector with the random forest algorithm to obtain the prediction results, and the prediction results are classified to obtain the atrial fibrillation classification and recognition results.

[0131] The expression of the above prediction result is:

[0132] y pred，final =RF[X mixed , Y initial ]

[0133] In the formula, RF means training with random forest algorithm, X mixed is the mixed eigenvector, Y initial is the training result, y pred，final is the predicted result; and Y initial The expression is:

[0134] Y initial =[0, 1, .... 1]

[0135] Where 0 represents normal heart rate and 1 represents atrial fibrillation;

[0136] The expression of atrial fibrillation classification and recognition results is:

[0137]

[0138] Where Abnormal Heart Rate is the atrial fibrillation classification result, y predfinal For the prediction results.

[0139] Example 2

[0140] An embodiment of the present invention provides a novel single-lead atrial fibrillation classification and identification system, which is implemented by the method described in Example 1 and includes:

[0141] A signal acquisition module, used for acquiring a single-lead electrocardiogram signal;

[0142] A preprocessing module, used for preprocessing the acquired single-lead electrocardiogram signal to obtain a processed signal;

[0143] The feature extraction module is used to extract the basic time domain features and frequency domain features from the processed signal to obtain extracted features, and construct a feature matrix using the seven features in the extracted features;

[0144] The model training module is used to construct a dataset using a feature matrix and corresponding label vectors, and use the dataset to train a base-level classifier to output meta-features.

[0145] A hybrid feature vector construction module is used to combine the remaining basic features and meta-features in the extracted features to obtain a hybrid feature vector;

[0146] The classification and recognition module is used to train and predict the model by combining mixed feature vectors with the random forest algorithm to obtain prediction results, classify the prediction results, and obtain atrial fibrillation classification and recognition results.

[0147] Example 3

[0148] An embodiment of the present invention provides a computer-readable storage medium, which includes a stored computer program; wherein, when the computer program is run, it controls the device where the computer-readable storage medium is located to execute the steps of the new single-lead atrial fibrillation classification and identification method as described in Example 1.

[0149] Example 4

[0150] An embodiment of the present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When executing the computer program, the processor implements the new single-lead atrial fibrillation classification and identification method as described in Example 1.

[0151] The following uses the MIT-BIH atrial fibrillation database (containing 25 atrial fibrillation patients and 25 normal heart rhythm data) to study the prediction performance of the method proposed in the embodiment of the present invention.

[0152] 1. Experimental Setup

[0153] Dataset division: 80% training set, 20% test set, 5-fold cross validation.

[0154] Comparison methods: support vector machine (SVM), logistic regression (LR), and single random forest (RF).

[0155] Evaluation indicators: Accuracy, sensitivity, specificity, and AUC value.

[0156] 2. Experimental results

[0157] (1) The performance comparison results of various methods are shown in Table 1 and Figure 1 shown.

[0158] Table 1 Performance comparison results of various methods

[0159]

[0160]

[0161] Through Table 1 and Figure 1 The results show that the proposed recognition method significantly outperforms the comparison method in all indicators, with an AUC value of 0.98, demonstrating its strong ability to distinguish atrial fibrillation from normal heart rhythm. The proposed method uses a two-stage classifier design (basic features + meta-features) to improve accuracy by 4.1% compared to a single random forest method, validating the effectiveness of multi-feature fusion.

[0162] (2) Comparison with 12-lead electrocardiogram

[0163] The clinical information of the patients is shown in Table 2 below.

[0164] Table 2 Clinical information of patients

[0165]

[0166] A total of 50 patients were tested (25 with atrial fibrillation and 25 with normal heart rhythm). Single-lead electrocardiogram (using the method of the present invention) and traditional 12-lead electrocardiogram data were collected from each patient.

[0167] Evaluation indicators:

[0168] Accuracy, sensitivity, specificity, and AUC value.

[0169] The performance comparison results of the method proposed in the present invention and the traditional 12-lead electrocardiogram are shown in Table 3 below.

[0170] Table 3 Performance comparison results of the method proposed in this invention and the traditional 12-lead electrocardiogram

[0171]

[0172] The results in Table 3 above show that the method proposed in the present invention has basically the same performance as the traditional 12-lead ECG. However, compared with the traditional 12-lead ECG, the present invention only requires 3 patches to complete the acquisition of electrical signals and the identification of atrial fibrillation. It eliminates the need for limb leads and reduces the number of precordial leads, significantly reducing the complexity of the device.

[0173] (3) Evaluate performance on different feature sets

[0174] We used cross-validation to select optimal parameters and evaluate the performance of our method on different feature sets. We used metrics such as confusion matrix, accuracy, and F1 score to evaluate performance, ensuring that our method achieved high recognition accuracy.

[0175] 1. Dataset Settings

[0176] The MIT-BIH Arrhythmia Database dataset is used, which contains 48 records, each of which contains two channels of electrocardiogram signals, and each channel is about 30 minutes. The present invention selects 40 records as the training set and 8 records as the test set.

[0177] 2. Use cross-validation to select the best parameters

[0178] Assuming that the present invention selects the random forest classifier as the basic classifier, the parameters that need to be adjusted include:

[0179] Number of trees: 10, 50, 100;

[0180] Feature selection method: random selection, information gain;

[0181] The present invention uses 10-fold cross validation, dividing the training set into 10 subsets, and performing training and validation in turn. For each parameter combination, the average accuracy of the 10 cross validations is calculated, and the parameter combination with the highest average accuracy is selected as the optimal parameter combination.

[0182] 3. Specific use of evaluation indicators

[0183] Assume that after cross-validation, the number of trees is selected as 50, and the feature selection method is information gain as the optimal parameter. The test set is input into the model, and the prediction results are as follows:

[0184] Table 4 Prediction results

[0185]

[0186]

[0187] The confusion matrix obtained from Table 4 is shown in Table 5 below.

[0188] Table 5 Confusion Matrix

[0189]

[0190] The results of each evaluation index calculated by the results in Table 4 and Table 5 are as follows:

[0191] Accuracy: (40 + 35) / 80 = 87.5%

[0192] Recall rate: 40 / 42 = 95.2%

[0193] F1 value: 2*87.5%*95.2% / (87.5%+95.2%)=91.3%

[0194] According to the above results, the accuracy of the recognition method proposed in the present invention is 87.5%, the recall rate is 95.2%, and the F1 value is 91.3%, which shows that the recognition method proposed in the present invention has high recognition accuracy and can effectively identify atrial fibrillation.

[0195] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to the embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the claims and their equivalents.

Claims

1. A novel single-lead atrial fibrillation classification and recognition method, characterized in that: The following steps are involved: S1. Acquire single-lead electrocardiogram signal; S2. preprocessing the acquired single-lead electrocardiogram signal to obtain a processed signal; S3. Extracting basic time-domain features and frequency-domain features from the processed signal to obtain extracted features, and constructing a feature matrix using seven features from the extracted features; S4. A dataset is constructed by constructing a feature matrix and a corresponding label vector, and a base-level classifier is trained using the dataset to output meta-features. S5, combining the remaining features in the extracted features with the meta-features to obtain a mixed feature vector; S6. The model is trained and predicted by combining the mixed feature vector with the random forest algorithm to obtain the prediction results, and the prediction results are classified to obtain the atrial fibrillation classification and recognition results.

2. A novel single-lead atrial fibrillation classification and identification method according to claim 1, characterized in that: In step S2, preprocessing includes performing baseline correction and denoising on the single-lead electrocardiogram signal.

3. A novel single-lead atrial fibrillation classification and identification method according to claim 1, characterized in that: In step S3, the basic time domain features include RR interval, PR interval, RR interval mean, RR interval standard deviation, RR interval variation coefficient, P wave peak value, P wave mean value, P wave amplitude variance, T wave peak value, T wave mean value, P wave correlation coefficient and Rényi entropy feature; The basic frequency domain features include heartbeat feature 1 and heartbeat feature 2.

4. A novel single-lead atrial fibrillation classification and identification method according to claim 1, characterized in that: In step S3, the seven features are RR interval, RR interval variation coefficient, PR interval, P wave correlation coefficient, Renyi entropy feature, heartbeat feature 1 and heartbeat feature 2.

5. A novel single-lead atrial fibrillation classification and identification method according to claim 1, characterized in that: In step S4, the base-level classifier consists of a linear discriminant analysis classifier and a random forest classifier; The expression of the posterior probability of the base-level classifier is: Where C LDA (C k |X) for each category C k Below, the posterior probability of the linear discriminant analysis classifier; C RF (C k |X) for each category C k Below, the posterior probability of the random forest classifier.

6. A novel single-lead atrial fibrillation classification and identification method according to claim 5, characterized in that: In step S4, the meta-features are the posterior probability mean and the posterior probability standard deviation of the base-level classifier.

7. A novel single-lead atrial fibrillation classification and identification method according to claim 5, characterized in that: In step S6, the expression of the prediction result is: y pred，final =RF[X mixed ,Y initial ] In the formula, RF means training with random forest algorithm, X mixed is the mixed eigenvector, Y initial is the training result, y pred，final To predict the results; The expression of atrial fibrillation classification and recognition results is: Where Abnormal Heart Rate is the atrial fibrillation classification result, y pred，final For the prediction results.

8. A novel single-lead atrial fibrillation classification and recognition system, characterized by: The method according to any one of claims 1 to 7 is implemented, comprising: A signal acquisition module, used for acquiring a single-lead electrocardiogram signal; A preprocessing module, used for preprocessing the acquired single-lead electrocardiogram signal to obtain a processed signal; The feature extraction module is used to extract the basic time domain features and frequency domain features from the processed signal to obtain extracted features, and construct a feature matrix using the seven features in the extracted features; The model training module is used to construct a dataset using a feature matrix and corresponding label vectors, and use the dataset to train a base-level classifier to output meta-features. A hybrid feature vector construction module is used to combine the remaining basic features and meta-features in the extracted features to obtain a hybrid feature vector; The classification and recognition module is used to train and predict the model by combining mixed feature vectors with the random forest algorithm to obtain prediction results, classify the prediction results, and obtain atrial fibrillation classification and recognition results.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium includes a stored computer program; wherein, when the computer program is run, it controls the device where the computer-readable storage medium is located to execute the steps of the novel single-lead atrial fibrillation classification and identification method according to any one of claims 1 to 7.

10. A terminal device, characterized in that: The invention comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor implements the novel single-lead atrial fibrillation classification and identification method according to any one of claims 1 to 7 when executing the computer program.

Images