A method, device and equipment for identifying hypertension based on pulse cycle characteristics

Abstract

The present invention proposes a method, device and equipment for identifying hypertension based on pulse cycle characteristics. The method is: the pulse data of the sample personnel are pre-processed, period-divided and processed with sliding windows in sequence and labeled, and then four cycle-to-cycle network features with greater contribution to the classification of hypertension diseases are selected as the pulse cycle-to-cycle network feature vector, and then the pulse data of the sample personnel after sliding window processing is input into the multi-scale permutation entropy algorithm, and the multi-scale permutation entropy value is extracted. Finally, the two features are fused to form a hypertension pulse feature vector, which is input into the classification model, and supervised and trained with the labels corresponding to the sample personnel to obtain a trained classification model. The present invention proves the effectiveness of the cycle-to-cycle features of the pulse signal by mining the cycle-to-cycle sensitive features, and also improves the accuracy of hypertension disease classification; the network features are used to make the mined cycle-to-cycle sensitive features interpretable.

Description

Technical Field

[0001] The invention belongs to the technical field of traditional Chinese medicine pulse acquisition systems, and in particular relates to a hypertension identification method, device, and electronic equipment based on pulse cycle characteristics. Background Art

[0002] Traditional Chinese medicine believes that the pulse signal at the human wrist contains a lot of physiological and pathological information, which can reflect the circulation of qi and blood, the function of the lungs and other pathological changes. From the perspective of modern medicine, the pulse signal begins with the periodic contraction and relaxation of the human heart. After the blood is pumped into various organs of the body, it gathers at the wrist artery, so it contains rich cardiovascular information. In recent years, pulse signals have been widely used in the field of disease identification, including hypertension, coronary heart disease, atrial fibrillation, etc.

[0003] In recent years, in the process of using pulse signals for disease identification, feature extraction can be roughly divided into four parts: analysis based on time domain, analysis based on frequency domain, analysis based on time-frequency domain, and analysis based on nonlinear methods. For time domain feature extraction, the amplitude, morphology, cycle, angle, amplitude ratio of various waves such as main wave and dicrotic wave, and other key points of pulse wave can be used for classification, but it can only reflect the local feature changes of pulse signal. For frequency domain feature extraction, STFT, HHT, wavelet transform, power spectrum and other methods are generally used for classification, but the extracted frequency domain information is generally difficult to understand. Time-frequency domain feature extraction combines the characteristics of time domain and frequency domain, and has a good effect for disease classification, but it is also limited to the internal characteristics of single-cycle signals, and it is difficult to pay attention to the relationship between the differences between pulse cycles and disease classification. For nonlinear features, sample entropy, approximate entropy and other entropy features are generally extracted, which can combine the internal differences of the cycle with the trend of the pulse signal as features. Because its features are combined with the differences within the cycle, it has an impact on the interpretability of proving the differences between cycles on the disease classification of pulse signals, which has certain limitations.

[0004] Since the pulse signal is a quasi-periodic signal and there are some small differences between each cycle, this difference is generally considered to contain certain physiological and pathological information. Most of the above feature extraction methods are based on single-cycle methods, and the extracted features cannot express the differences between cycles. Traditional Chinese medicine theory believes that the differences between cycles and the differences within cycles are both important. Therefore, if the intra-cycle features can be extracted, it will be helpful to prove the effectiveness of disease classification based on the inter-cycle differences of pulse signals, and it will be helpful to achieve the interpretability of the inter-cycle difference features, and it will also be helpful to improve the accuracy of disease identification. Summary of the invention

[0005] The technical problem to be solved by the present invention is that in view of the deficiencies of the above-mentioned prior art, a method, device, electronic device and medium for identifying hypertension based on pulse cycle characteristics are provided. The method proves the effectiveness of the pulse signal cycle characteristics by mining the cycle-sensitive features, and also improves the accuracy of hypertension disease classification; the network features are used to make the mined cycle-sensitive features explainable.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is: a method for identifying hypertension based on the characteristics of the pulse cycle, comprising:

[0007] Acquiring pulse data of a sample person, and sequentially performing preprocessing, period division processing, and sliding window processing on the pulse data of the sample person, wherein the pulse data of the sample person is labeled with a label for distinguishing whether the sample person suffers from hypertension;

[0008] A connectivity graph is constructed based on the pulse data of the sample personnel after sliding window processing, a plurality of inter-cycle network features are extracted from the connectivity graph, and four inter-cycle network features with greater contribution to the classification of hypertension diseases are selected from the plurality of inter-cycle network features as the pulse inter-cycle network feature vector, wherein the four inter-cycle network features are respectively the average clustering coefficient, the transitivity coefficient, the betweenness centrality coefficient and the local efficiency;

[0009] Input the pulse data of the sample personnel after sliding window processing into the multi-scale permutation entropy algorithm to extract the multi-scale permutation entropy value;

[0010] The network feature vector and the multi-scale permutation entropy value during the pulse cycle are fused to form a hypertension pulse feature vector, which is input into the classification model and supervised and trained with the labels corresponding to the sample persons to obtain a trained classification model.

[0011] Preferably, a method for identifying hypertension based on pulse cycle characteristics further includes:

[0012] Acquire the pulse data of the person to be diagnosed, and perform preprocessing, period division processing and sliding window processing on the pulse data of the person to be diagnosed in sequence;

[0013] The network feature vector and multi-scale permutation entropy value of the pulse cycle of the person to be diagnosed are obtained from the pulse data of the person to be diagnosed after sliding window processing, and the hypertension pulse feature vector of the person to be diagnosed is fused and input into the trained classification model to determine whether the person to be diagnosed is a hypertension patient.

[0014] Preferably, the pulse data of the sample personnel are sequentially preprocessed, periodically divided and processed by a sliding window, specifically including:

[0015] The pulse data of multiple healthy people and multiple hypertensive patients were collected, and the collected pulse data were respectively subjected to baseline drift removal, smoothing, filtering, and cycle division. Then, a sliding window was used to combine the pulse data of 10 consecutive cycles into a set of hypertension recognition pulse data.

[0016] Preferably, the method of constructing a connectivity graph based on the pulse data of sample personnel after sliding window processing, extracting multiple cycle-to-cycle network features from the connectivity graph, and selecting four cycle-to-cycle network features with greater contribution to the classification of hypertension diseases from the multiple cycle-to-cycle network features as the cycle-to-cycle network feature vector of the pulse specifically includes:

[0017] For the pulse signals of 10 consecutive cycles obtained after sliding window processing, the correlation coefficients between the pulse signals of 10 consecutive cycles are calculated according to the calculation method of Pearson correlation coefficient, and then the appropriate threshold is taken to binarize the matrix, and then a connectivity graph is constructed according to the binarized matrix, and 10 inter-cycle network features of the pulse signal that are effective for the classification of hypertension diseases are extracted from the connectivity graph. The 10 inter-cycle network features are respectively the average clustering coefficient, the transitivity coefficient, the betweenness centrality coefficient, the local efficiency and the number of edges, degree centrality, closeness centrality, global efficiency, the number of connected components and the average node connectivity of the connectivity graph. After experiments, 4 inter-cycle network features with greater contribution to the classification of hypertension diseases are obtained and used as the inter-cycle network feature vectors of the pulse.

[0018] Preferably, the extracted multi-scale permutation entropy values ​​are permutation entropy values ​​at 20 scales.

[0019] Preferably, the classification model is a bagged tree.

[0020] The present invention also provides a hypertension identification device based on pulse cycle characteristics, comprising:

[0021] The acquisition module is used to obtain the pulse data of the sample personnel, and perform preprocessing, period division processing and sliding window processing on the pulse data of the sample personnel in sequence, and the pulse data of the sample personnel is marked with a label for distinguishing whether the sample personnel suffers from hypertension;

[0022] A feature extraction module is used to construct a connectivity graph based on the pulse data of the sample personnel after the sliding window processing, extract multiple cycle network features from the connectivity graph, and select four cycle network features with greater contribution to the classification of hypertension diseases from the multiple cycle network features as the pulse cycle network feature vector, wherein the four cycle network features are respectively an average clustering coefficient, a transitivity coefficient, a betweenness centrality coefficient, and a local efficiency; the pulse data of the sample personnel after the sliding window processing is input into a multi-scale permutation entropy algorithm to extract a multi-scale permutation entropy value;

[0023] The training module is used to fuse the network feature vector and the multi-scale permutation entropy value during the pulse cycle to form a hypertension pulse feature vector, input it into the classification model, and supervise and train it with the labels corresponding to the sample personnel to obtain a trained classification model.

[0024] Preferably, a pulse-based hypertension identification device further comprises:

[0025] The diagnosis module is used to input the hypertension pulse feature vector of the person to be diagnosed into the trained classification model to determine whether the person to be diagnosed is a hypertension patient.

[0026] The present invention further provides an electronic device, characterized in that the electronic device comprises:

[0027] At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the above-mentioned hypertension identification method based on pulse cycle characteristics.

[0028] The present invention also provides a computer-readable storage medium storing computer instructions, wherein the computer instructions are operated to execute the above-mentioned hypertension identification method based on pulse cycle characteristics.

[0029] Compared with the prior art, the present invention has the following advantages:

[0030] 1. The hypertension identification method provided by the present invention is different from the feature extraction method of single cycle and overall signal. By mining the sensitive features between pulse cycles, the effectiveness of the features between pulse signal cycles is proved, and the accuracy of hypertension classification is also improved.

[0031] 2. The present invention extracts multiple inter-cycle network features related to the hypertension pulse cycle through the connectivity graph constructed from the inter-cycle signal. The extracted inter-cycle network features can characterize the characteristics of the connectivity graph, and then four inter-cycle network features with greater contribution to the classification of hypertension diseases are selected from the multiple inter-cycle network features as the inter-cycle network feature vector. The extracted inter-cycle network feature vector is helpful to prove the effectiveness of hypertension classification based on the inter-cycle difference of the pulse signal, which makes the mined inter-cycle sensitive features explainable. At the same time, the pulse data is also input into the multi-scale permutation entropy algorithm, and the multi-scale permutation entropy value is extracted, which is fused with the inter-cycle network feature vector to form a hypertension pulse feature vector for identifying hypertension. The hypertension pulse feature vector constructed by the inventor has a significant effect on improving the accuracy of hypertension identification.

[0032] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 This is a flowchart of a method for identifying hypertension based on pulse cycle characteristics disclosed in Example 1 of the present invention.

[0034] Figure 2 It is a Pearson matrix heat map of healthy people and hypertensive patients in a hypertension identification method based on pulse cycle characteristics disclosed in Example 1 of the present invention.

[0035] Figure 3 It is a Pearson matrix heat map under different thresholds of a hypertension identification method based on pulse cycle characteristics disclosed in Example 1 of the present invention.

[0036] Figure 4 It is a connectivity diagram under different thresholds of a hypertension identification method based on pulse cycle characteristics disclosed in Example 1 of the present invention.

[0037] Figure 5 It is a violin diagram of feature weights based on inter-cycle features of a hypertension identification method based on inter-cycle features of pulse disclosed in Example 1 of the present invention.

[0038] Figure 6 This is a graph of experimental results of a hypertension identification method based on pulse cycle characteristics disclosed in Example 1 of the present invention using multi-scale permutation entropy (MPE) and multi-scale approximate entropy (MSE) at 20 scales.

[0039] Figure 7 This is a result diagram of the effectiveness of the inter-cycle network features of a hypertension identification method based on pulse cycle features disclosed in Example 1 of the present invention. DETAILED DESCRIPTION

[0040] Example 1

[0041] like Figure 1 As shown, this embodiment discloses a method for identifying hypertension based on the characteristics of the pulse cycle, which is:

[0042] S101, obtaining pulse data of a sample person, and sequentially performing preprocessing, period division processing and sliding window processing on the pulse data of the sample person, and labeling the pulse data of the sample person, wherein the label is used to distinguish the pulse data as a pulse signal curve of a healthy person and a hypertensive patient, and the label can be defined as: a label of a healthy person is "1", and a label of a hypertensive patient is "0";

[0043] In this embodiment, there are 33 sample persons, including 26 healthy persons and 7 hypertensive patients. More than 13,000 pulse data of the 33 sample persons are collected. Each pulse data is subjected to baseline drift removal, smoothing, Butterworth filtering, and cycle division operations. Then, a sliding window is used to combine 10 consecutive cycles of pulse data into a set of hypertension recognition pulse data.

[0044] S201, constructing a connectivity graph based on the pulse data of the sample personnel after sliding window processing, extracting multiple cycle-to-cycle network features from the connectivity graph, and selecting four cycle-to-cycle network features with greater contribution to the classification of hypertension diseases from the multiple cycle-to-cycle network features as the pulse cycle-to-cycle network feature vector, wherein the four cycle-to-cycle network features are respectively an average clustering coefficient, a transitivity coefficient, a betweenness centrality coefficient, and a local efficiency;

[0045] Specifically, if any two single-cycle pulse signals are X and Y, the Pearson correlation coefficient between the two signals is:

[0046]

[0047] where ρ X,Y is the Pearson correlation coefficient, cov(X,Y) is the covariance of X and Y, σ X is the standard deviation of signal X, σ Y is the standard deviation of signal Y;

[0048] For the 10 consecutive cycles of pulse signals obtained by the sliding window, the Pearson correlation coefficient between the pulse signals of 10 consecutive cycles is calculated according to the above method of calculating the Pearson correlation coefficient between the two signals. The specific correlation coefficient matrix is: Figure 2 As shown, Figure 2 (a) is the Pearson matrix heat map of healthy people. Figure 2 (b) is the Pearson matrix heat map of patients with hypertension. Then, a suitable threshold is obtained to binarize the matrix, and then a connectivity graph is constructed based on the binarized matrix. Figure 3 It is the Pearson matrix diagram under different thresholds (no threshold 0, 0.70, 0.75, 0.80, 0.85, 0.90, 0.95, 1.00). Figure 4 The network connectivity diagrams are shown in Figure 1 under different thresholds (no threshold 0, 0.70, 0.75, 0.80, 0.85, 0.90, 0.95, 1.00). Through experimental analysis, the present invention can extract the network features between cycles when the threshold is 0.85.

[0049] The inter-cycle network features extracted from the connectivity graph are as follows: the inter-cycle network features obtained by constructing the connectivity graph from the data obtained by period segmentation, and 10 inter-cycle network features of the pulse signal that are effective for the classification of hypertension diseases are extracted from the connectivity graph. The 10 inter-cycle network features are the average clustering coefficient, the transitivity coefficient, the betweenness centrality coefficient, the local efficiency and the number of edges, degree centrality, closeness centrality, global efficiency, number of connected components and average node connectivity of the connectivity graph. The weights of the 10 inter-cycle network features after feature optimization using the relief-f algorithm are plotted into a violin plot, as shown in Figure 5 As shown in the figure (the features of the horizontal axis 1-10 are the number of edges of the connected graph, the average clustering coefficient, the transitivity coefficient, the degree centrality, the closeness centrality, the betweenness centrality coefficient, the local efficiency, the global efficiency, the number of connected components and the average node connectivity), after 20 feature optimizations, it can be observed that the second, third, sixth and seventh features contribute more to the classification of hypertension diseases. The second, third, sixth and seventh features are the average clustering coefficient, the transitivity coefficient, the betweenness centrality coefficient and the local efficiency coefficient, which are used as the network feature vectors between the pulse cycles. The formulas of the network features between these four cycles are:

[0050] Average clustering coefficient (average_clustering):

[0051]

[0052] Where n is the number of nodes in the graph G.

[0053] Transitivity coefficient:

[0054]

[0055] It is calculated by taking the weighted average of the vertex clustering coefficients of all vertices with degree at least 2.

[0056] Betweenness centrality coefficient (betweenness_centrality):

[0057]

[0058] where d st represents the number of shortest paths from s to t, d st (i) represents the number of i nodes passed through by the shortest path from s to t.

[0059] Local efficiency (local_efficiency):

[0060]

[0061] is the shortest path length between adjacent nodes j and h that only contain nodes i.

[0062] S103, inputting the pulse data of the sample personnel after the sliding window processing into the multi-scale permutation entropy algorithm to extract the multi-scale permutation entropy value;

[0063] Specifically, the pulse data of the sample personnel after sliding window processing is used as the input of the multi-scale permutation entropy algorithm, and the permutation entropy values ​​at 20 scales are extracted to construct the hypertension pulse feature vector of the sample personnel. The realization of multi-scale permutation entropy is firstly to obtain signals of multiple scales through coarse-graining, and then to obtain the permutation entropy of the signals of multiple scales respectively.

[0064] The coarse-graining adopted in this embodiment is non-overlapping coarse-graining, which jumps τ data each time and takes the τ data as an average to generate new data.

[0065] The coarse-grained formula is as follows:

[0066]

[0067] Here, τ represents the time scale.

[0068] The permutation entropy formula is as follows:

[0069]

[0070] Among them, P j is the probability of the jth arrangement occurring.

[0071] S104, fusing the pulse cycle network feature vector and the multi-scale permutation entropy value to form a hypertension pulse feature vector, inputting it into a classification model, and supervising and training it with labels corresponding to sample personnel to obtain a trained classification model;

[0072] S105. Obtain the pulse data of the person to be diagnosed, and perform preprocessing, period division processing and sliding window processing on the pulse data of the person to be diagnosed in sequence; obtain the network feature vector and multi-scale permutation entropy value of the pulse cycle of the person to be diagnosed from the pulse data of the person to be diagnosed after sliding window processing according to the methods disclosed in steps S102, S103 and S104, fuse them into a hypertension pulse feature vector of the person to be diagnosed, input them into the trained classification model, and determine whether the person to be diagnosed is a hypertension patient.

[0073] In this embodiment, the classification models involved in the training are ComplexTree, CubicSVM, FineKNN, CosineKNN, WeightedKNN, BoostedTrees, SubspaceKNN, and BaggedTrees. The above 8 classifier models are trained using the pulse data of sample personnel, and the accuracy, precision, area under the ROC curve (AUC) and F1 coefficient are used to evaluate the relevant classification conditions, and then the classification model is preferably selected as BaggedTrees.

[0074] In this embodiment, in addition to inputting the pulse data of the sample personnel after sliding window processing into the multi-scale permutation entropy algorithm to extract the permutation entropy value (MPE) of 20 scales, the multi-scale sample entropy algorithm is also used to extract the sample entropy value (MSE) of 20 scales as a comparative test, and the multi-scale sample entropy value and the pulse network feature vector are fused to form a hypertension pulse feature vector, and the classification model is trained and used to identify hypertension patients (the pulse data of the personnel to be diagnosed for training and identification are the same), and the data in Table 1 are plotted as shown in FIG. Figure 6 The line chart shown, from Figure 6 The results show that MPE has higher classification accuracy, AUC value, and F1 coefficient value than MSE on eight classifiers, which reflects the effectiveness of multi-scale permutation entropy.

[0075] Table 1 Classification performance indicators of MPE and MSE

[0076]

[0077]

[0078] In this embodiment, a number of comparative tests were also conducted, namely:

[0079] In the comparative experiment 1, no optimization was performed from the network features between the 10 cycles, and the hypertension pulse feature vector was directly formed;

[0080] In contrast experiment 2, no optimization was performed from the 10 cycle network features, and they were directly concatenated and fused with the 20 scale permutation entropy values ​​to form a 30-feature hypertension pulse feature vector;

[0081] In comparative experiment 3, 20 scale permutation entropy values ​​were used to directly form the hypertension pulse feature vector;

[0082] After training the ensemble learning classifier with the same data, the classification accuracy was tested and the obtained data are shown in Table 2.

[0083] Table 2 Classification results before and after feature optimization

[0084]

[0085] Plot the data in the table Figure 7 ,Comparative experiments 1, 2, and 3 prove the effectiveness of the inter-cycle network features extracted by the present invention in identifying hypertension diseases.,Comparative experiment 3 and Example 1 prove that the four network features obtained after feature screening have a great contribution to the classification of hypertension.

[0086] Example 2

[0087] This embodiment provides a hypertension identification device based on pulse cycle characteristics, including:

[0088] The acquisition module obtains the pulse data of the sample personnel, and performs preprocessing, period division processing and sliding window processing on the pulse data of the sample personnel in sequence, and labels the pulse data of the sample personnel. The labels are used to distinguish the pulse data as the pulse signal curve of a healthy person and a hypertensive patient. The labels can be defined as: the label of a healthy person is "1", and the label of a hypertensive patient is "0";

[0089] In this embodiment, there are 33 sample persons, including 26 healthy persons and 7 hypertensive patients. More than 13,000 pulse data of the 33 sample persons are collected. Each pulse data is subjected to baseline drift removal, smoothing, Butterworth filtering, and cycle division operations. Then, a sliding window is used to combine 10 consecutive cycles of pulse data into a set of hypertension recognition pulse data.

[0090] A feature extraction module is used to construct a connectivity graph based on the pulse data of the sample personnel after the sliding window processing, extract multiple cycle network features from the connectivity graph, and select four cycle network features with greater contribution to the classification of hypertension diseases from the multiple cycle network features as the pulse cycle network feature vector, wherein the four cycle network features are respectively an average clustering coefficient, a transitivity coefficient, a betweenness centrality coefficient, and a local efficiency; the pulse data of the sample personnel after the sliding window processing is input into a multi-scale permutation entropy algorithm to extract a multi-scale permutation entropy value;

[0091] Specifically, if any two single-cycle pulse signals are X and Y, the Pearson correlation coefficient between the two signals is:

[0092]

[0093] where ρ X,Y is the Pearson correlation coefficient, cov(X,Y) is the covariance of X and Y, σ X is the standard deviation of signal X, σ Y is the standard deviation of signal Y;

[0094] For the 10 consecutive cycles of pulse signals obtained by the sliding window, the Pearson correlation coefficient between the pulse signals of 10 consecutive cycles is calculated according to the above method of calculating the Pearson correlation coefficient between the two signals. The specific correlation coefficient matrix is: Figure 2 As shown, Figure 2 (a) is the Pearson matrix heat map of healthy people. Figure 2 (b) is the Pearson matrix heat map of patients with hypertension. Then, a suitable threshold is obtained to binarize the matrix, and then a connectivity graph is constructed based on the binarized matrix. Figure 3 It is the Pearson matrix diagram under different thresholds (no threshold 0, 0.70, 0.75, 0.80, 0.85, 0.90, 0.95, 1.00). Figure 4 The network connectivity diagrams are shown in Figure 1 under different thresholds (no threshold 0, 0.70, 0.75, 0.80, 0.85, 0.90, 0.95, 1.00). Through experimental analysis, the present invention can extract the network features between cycles when the threshold is 0.85.

[0095] The inter-cycle network features extracted from the connectivity graph are as follows: the inter-cycle network features obtained by constructing the connectivity graph from the data obtained by period segmentation, and 10 inter-cycle network features of the pulse signal that are effective for the classification of hypertension diseases are extracted from the connectivity graph. The 10 inter-cycle network features are the average clustering coefficient, the transitivity coefficient, the betweenness centrality coefficient, the local efficiency and the number of edges, degree centrality, closeness centrality, global efficiency, number of connected components and average node connectivity of the connectivity graph. The weights of the 10 inter-cycle network features after feature optimization using the relief-f algorithm are plotted into a violin plot, as shown in Figure 5 As shown in the figure (the features of the horizontal axis 1-10 are the number of edges of the connected graph, the average clustering coefficient, the transitivity coefficient, the degree centrality, the closeness centrality, the betweenness centrality coefficient, the local efficiency, the global efficiency, the number of connected components and the average node connectivity), after 20 feature optimizations, it can be observed that the second, third, sixth and seventh features contribute more to the classification of hypertension diseases. The second, third, sixth and seventh features are the average clustering coefficient, the transitivity coefficient, the betweenness centrality coefficient and the local efficiency coefficient, which are used as the network feature vectors between the pulse cycles. The formulas of the network features between these four cycles are:

[0096] Average clustering coefficient (average_clustering):

[0097]

[0098] Where n is the number of nodes in the graph G.

[0099] Transitivity coefficient:

[0100]

[0101] It is calculated by taking the weighted average of the vertex clustering coefficients of all vertices with degree at least 2.

[0102] Betweenness centrality coefficient (betweenness_centrality):

[0103]

[0104] where d st represents the number of shortest paths from s to t, d st (i) represents the number of i nodes passed through by the shortest path from s to t.

[0105] Local efficiency (local_efficiency):

[0106]

[0107] is the shortest path length between adjacent nodes j and h that only contain nodes i.

[0108] A training module, for fusing the network feature vector and the multi-scale permutation entropy value during the pulse cycle to form a hypertension pulse feature vector, inputting it into a classification model, and supervising and training it with labels corresponding to sample personnel to obtain a trained classification model;

[0109] Specifically, the pulse data of the sample personnel after sliding window processing is used as the input of the multi-scale permutation entropy algorithm, and the permutation entropy values ​​at 20 scales are extracted to construct the hypertension pulse feature vector of the sample personnel. The realization of multi-scale permutation entropy is firstly to obtain signals of multiple scales through coarse-graining, and then to obtain the permutation entropy of the signals of multiple scales respectively.

[0110] The coarse-graining adopted in this embodiment is non-overlapping coarse-graining, which jumps τ data each time and takes the τ data as an average to generate new data.

[0111] The coarse-grained formula is as follows:

[0112]

[0113] Here, τ represents the time scale.

[0114] The permutation entropy formula is as follows:

[0115]

[0116] Among them, P j is the probability of the jth arrangement occurring.

[0117] In this embodiment, the classification model is a bagged tree (BaggedTrees)

[0118] The diagnosis module and the acquisition module obtain the pulse data of the person to be diagnosed, and perform preprocessing, period division processing and sliding window processing on the pulse data of the person to be diagnosed in turn; the feature extraction module obtains the network feature vector and multi-scale arrangement entropy value of the pulse cycle of the person to be diagnosed from the pulse data of the person to be diagnosed after the sliding window processing, fuses them into the hypertension pulse feature vector of the person to be diagnosed, and inputs them into the classification model trained by the training module to determine whether the person to be diagnosed is a hypertension patient.

[0119] Example 3

[0120] This embodiment further provides an electronic device, characterized in that the electronic device includes:

[0121] At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor so as to enable the at least one processor to execute the hypertension identification method based on pulse cycle characteristics of Example 1.

[0122] Example 4

[0123] This embodiment also provides a computer-readable storage medium storing computer instructions, wherein the computer instructions are operated to execute the hypertension identification method based on pulse cycle characteristics of embodiment 1.

[0124] The above is only a preferred embodiment of the present invention and does not limit the present invention in any way. Any simple modification, change and equivalent change made to the above embodiment according to the technical essence of the invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A method for identifying hypertension based on pulse cycle characteristics, characterized in that: include: Acquiring pulse data of a sample person, and sequentially performing preprocessing, period division processing, and sliding window processing on the pulse data of the sample person, wherein the pulse data of the sample person is labeled with a label for distinguishing whether the sample person suffers from hypertension; A connectivity graph is constructed based on the pulse data of the sample personnel after sliding window processing, a plurality of inter-cycle network features are extracted from the connectivity graph, and four inter-cycle network features with greater contribution to the classification of hypertension diseases are selected from the plurality of inter-cycle network features as the pulse inter-cycle network feature vector, wherein the four inter-cycle network features are respectively the average clustering coefficient, the transitivity coefficient, the betweenness centrality coefficient and the local efficiency; Input the pulse data of the sample personnel after sliding window processing into the multi-scale permutation entropy algorithm to extract the multi-scale permutation entropy value; The network feature vector during the pulse cycle and the multi-scale permutation entropy value are fused to form a hypertension pulse feature vector, which is input into the classification model, and supervised and trained with the labels corresponding to the sample personnel to obtain a trained classification model; The method of constructing a connectivity graph based on the pulse data of sample personnel after sliding window processing specifically includes: calculating the correlation coefficient between the pulse signals of 10 consecutive cycles obtained after sliding window processing according to the calculation method of the Pearson correlation coefficient, then taking a suitable threshold to binarize the matrix, and then constructing a connectivity graph based on the binarized matrix.

2. A method for identifying hypertension based on pulse cycle characteristics according to claim 1, characterized in that: Also includes: Acquire the pulse data of the person to be diagnosed, and perform preprocessing, period division processing and sliding window processing on the pulse data of the person to be diagnosed in sequence; The network feature vector and multi-scale permutation entropy value of the pulse cycle of the person to be diagnosed are obtained from the pulse data of the person to be diagnosed after sliding window processing, and the hypertension pulse feature vector of the person to be diagnosed is fused and input into the trained classification model to determine whether the person to be diagnosed is a hypertension patient.

3. The method for identifying hypertension based on pulse cycle characteristics according to claim 1, characterized in that: The pulse data of the sample personnel are sequentially preprocessed, periodically divided and processed, and the sliding window is processed, specifically including: The pulse data of multiple healthy people and multiple hypertensive patients were collected, and the collected pulse data were respectively subjected to baseline drift removal, smoothing, filtering, and cycle division. Then, a sliding window was used to combine the pulse data of 10 consecutive cycles into a set of hypertension recognition pulse data.

4. A method for identifying hypertension based on pulse cycle characteristics according to claim 3, characterized in that: The method of constructing a connectivity graph based on the pulse data of the sample personnel after the sliding window processing, extracting a plurality of inter-cycle network features from the connectivity graph, and selecting four inter-cycle network features with greater contribution to the classification of hypertension diseases from the plurality of inter-cycle network features as the inter-cycle network feature vector of the pulse specifically includes: For the pulse signals of 10 consecutive cycles obtained after sliding window processing, the correlation coefficients between the pulse signals of 10 consecutive cycles are calculated according to the calculation method of Pearson correlation coefficient, and then the appropriate threshold is taken to binarize the matrix, and then a connectivity graph is constructed according to the binarized matrix, and 10 inter-cycle network features of the pulse signal that are effective for the classification of hypertension diseases are extracted from the connectivity graph. The 10 inter-cycle network features are respectively the average clustering coefficient, the transitivity coefficient, the betweenness centrality coefficient, the local efficiency and the number of edges, degree centrality, closeness centrality, global efficiency, the number of connected components and the average node connectivity of the connectivity graph. After experiments, 4 inter-cycle network features with greater contribution to the classification of hypertension diseases are obtained and used as the inter-cycle network feature vectors of the pulse.

5. A method for identifying hypertension based on pulse cycle characteristics according to claim 1 or 2, characterized in that: The extracted multi-scale permutation entropy values ​​are permutation entropy values ​​at 20 scales.

6. The method for identifying hypertension based on pulse cycle characteristics according to claim 1, characterized in that: The classification model is bagged trees.

7. A hypertension identification device based on pulse cycle characteristics, characterized in that: include: The acquisition module is used to obtain the pulse data of the sample personnel, and perform preprocessing, period division processing and sliding window processing on the pulse data of the sample personnel in sequence, and the pulse data of the sample personnel is marked with a label for distinguishing whether the sample personnel suffers from hypertension; A feature extraction module is used to construct a connectivity graph based on the pulse data of the sample personnel after the sliding window processing, extract multiple cycle network features from the connectivity graph, and select four cycle network features with greater contribution to the classification of hypertension diseases from the multiple cycle network features as the pulse cycle network feature vector, wherein the four cycle network features are respectively an average clustering coefficient, a transitivity coefficient, a betweenness centrality coefficient, and a local efficiency; the pulse data of the sample personnel after the sliding window processing is input into a multi-scale permutation entropy algorithm to extract a multi-scale permutation entropy value; A training module is used to fuse the network feature vector and the multi-scale permutation entropy value during the pulse cycle to form a hypertension pulse feature vector, input it into the classification model, and supervise and train it with the labels corresponding to the sample personnel to obtain a trained classification model; The method of constructing a connectivity graph based on the pulse data of sample personnel after sliding window processing specifically includes: calculating the correlation coefficient between the pulse signals of 10 consecutive cycles obtained after sliding window processing according to the calculation method of the Pearson correlation coefficient, then taking a suitable threshold to binarize the matrix, and then constructing a connectivity graph based on the binarized matrix.

8. The device for identifying hypertension based on pulse cycle characteristics according to claim 7, characterized in that: Also includes: The diagnosis module is used to input the hypertension pulse feature vector of the person to be diagnosed into the trained classification model to determine whether the person to be diagnosed is a hypertension patient.

9. An electronic device, characterized in that: The electronic device comprises: At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor so that the at least one processor can execute the hypertension identification method based on pulse cycle characteristics as described in any one of claims 1-6.

10. A computer-readable storage medium storing computer instructions, wherein the computer instructions are operated to execute the hypertension identification method based on pulse cycle characteristics according to any one of claims 1 to 6.

Images