Disease feature extraction and disease identification method based on slack block diagonal sparse embedding

Abstract

The invention provides a disease feature extraction and disease identification method based on slack block diagonal sparse embedding, comprising the following steps: acquiring a disease sample data set, the disease sample comprising category labels and multiple feature indexes; constructing a disease feature extraction model based on slack block diagonal sparse embedding, and performing model training by using a disease sample data set to obtain a projection matrix; the disease feature extraction model comprises sparse representation, a block diagonal matrix and a low-dimensional embedding matrix; iteratively updating the sparse representation, the block diagonal matrix and the low-dimensional embedding matrix by adopting an alternating optimization method to obtain an updated projection matrix; and obtaining target disease data, and performing feature extraction on the target disease data by applying the projection matrix obtained in the previous step to obtain a feature matrix.

Description

Technical Field

[0001] The present invention relates to the technical field of pattern recognition, and in particular to a disease feature extraction and disease recognition method based on relaxed block diagonal sparse embedding. Background Art

[0002] With the rapid development of medical imaging, genomics, and clinical data, data-driven disease diagnosis methods have gained widespread application. Traditional disease diagnosis methods primarily rely on physician experience and traditional medical examinations, but these methods often suffer from low efficiency and accuracy. In recent years, the application of machine learning, particularly deep learning, in medical diagnosis has made significant progress. By analyzing large amounts of medical data, machine learning algorithms can assist physicians in disease prediction and diagnosis. However, in practical applications, medical data often contains a large amount of noise and redundant information, which affects model training and, in turn, the accuracy and reliability of disease diagnosis. Therefore, extracting effective features from complex medical data and reducing redundant information has become a key issue in improving disease diagnosis performance. To address this challenge, researchers have proposed a variety of feature selection and extraction methods in recent years. Among them, sparse embedding and embedding subspace learning techniques have become important research directions.

[0003] In the field of pattern recognition, sparse embedding is an important feature extraction technique with broad application value in improving model efficiency, accuracy, and interpretability. For example, in disease diagnosis, high-dimensional medical data is susceptible to noise. Sparse embedding can be used to select important features and reduce the impact of noise. Sparse representation methods capture the key information of the data by learning sparse representation vectors, effectively finding the most representative features in high-dimensional data. Their core is to represent the data as a sparse combination on a certain basis. By constructing a sparse feature representation, they reduce the data dimensionality while retaining key information and removing redundancy and noise. Embedding subspace learning, on the other hand, reduces the dimensionality of high-dimensional data by learning a projection matrix that captures the key features of the data, transforming the data from the original high-dimensional space into a more compact low-dimensional representation embedding subspace. This method reduces the dimensionality of high-dimensional data while preserving the important information of the original data. However, some traditional embedding subspace learning methods are very sensitive to noise and outliers, which seriously affects the accuracy of disease diagnosis and recognition. Although sparse embedding has been proven to be robust to noise, the sparsification process is uncontrolled and unsupervised. This unsupervised sparsity helps reduce the dimensionality of disease data and the computational complexity, but it also weakens the interpretability of feature extraction and the ability to discriminate disease data to a certain extent. At the same time, traditional sparse embedding methods usually perform sparse encoding on each data sample independently, while ignoring the correlation between samples or the constraints of the overall structure.

[0004] The use of data label information is also crucial in the feature extraction process. Label information, as a supervisory signal, can guide the extraction process of disease features, more accurately locate features that have a key impact on classification, and make the extracted features more conducive to the classification task, thereby improving the overall classification performance. However, traditional methods using label information only focus on preserving discriminant information in the disease data reconstruction process while ignoring projection learning, or explicitly mining label information through regression functions, resulting in completely unsupervised learning of representation coefficients in the reconstruction process. This will lead to insufficient mining of label information, weakening the contribution of discriminant features to the classification task, and thus affecting the model's disease diagnosis and recognition performance.

[0005] In order to solve the above problems, people have been seeking an ideal technical solution. Summary of the Invention

[0006] Based on this, it is necessary to provide a disease feature extraction and disease identification method based on relaxed block diagonal sparse embedding to address the above technical problems.

[0007] In order to achieve the above object, the first aspect of the present invention provides a disease feature extraction method based on relaxed block diagonal sparse embedding, comprising the following steps:

[0008] Obtaining a disease sample dataset, wherein the disease samples include category labels and multiple feature indicators;

[0009] Constructing a disease feature extraction model based on relaxed block diagonal sparse embedding, using a disease sample dataset for model training to obtain a projection matrix; the disease feature extraction model includes a sparse representation, a block diagonal matrix, and a low-dimensional embedding matrix;

[0010] The alternating optimization method is used to iteratively update the sparse representation, block diagonal matrix, and low-dimensional embedding matrix to obtain the updated projection matrix.

[0011] Obtain target disease data, apply the projection matrix obtained in the previous steps to perform feature extraction on the target disease data, and obtain a feature matrix.

[0012] Preferably, a disease feature extraction model based on relaxed block diagonal sparse embedding is constructed, comprising:

[0013] Design the objective function, which is expressed as follows:

[0014]

[0015] Where X=[x1,x2,...,x n ]∈R m×n is the sample matrix, m is the number of samples, n is the number of features; Q∈R m×cis the projection matrix, c< <m;Z∈R n×n is the coefficient matrix; ⊙ is the Hadamard-product operator; M is the dynamic relaxation matrix, B=[b1,b2,...,b n ]∈R n×n is an ideal matrix, where b i =[0...,1,1,1,0,...] T ∈R n ; H=[h1,h2,...,h n ]∈R c ×n is a 0-1 label matrix, where h i =[0,0,…1,…,0,0] T , is x i The label vector, that is, if x i From the kth class, then h i Only the kth item is 1, and the rest are 0.

[0016] Preferably, an alternating optimization method is used to iteratively update the sparse representation, the block diagonal matrix, and the low-dimensional embedding matrix to obtain an updated projection matrix, including:

[0017] Fixing Z and M, equation (1) is transformed into:

[0018]

[0019] Derivative (2) and equal to zero, the final solution is:

[0020]

[0021] in is a diagonal matrix whose diagonal elements are Right now γ=(XZ-X)(XZ-X) T ;

[0022] Fixing Q and M, equation (1) is transformed into:

[0023]

[0024] Solving equation (4), we get:

[0025] Z=(X T QQ T X+I) -1 (X T QQ T X+B⊙M)(5);

[0026] Fixing Q and Z, equation (1) is transformed into:

[0027]

[0028] Let C = B-1, where 1 is a matrix of all 1s, then we get the equivalent formula:

[0029]

[0030] Solving equation (7), we get:

[0031] M=ZC⊙M(8).

[0032] Preferably, after obtaining the disease sample dataset, the disease sample dataset is randomly divided into a training set and a test set;

[0033] Use the training set to train the model and obtain the projection matrix;

[0034] Apply the projection matrix obtained in the previous steps to extract features from the test samples in the test set to obtain a feature matrix;

[0035] Use the KNN classifier to predict and classify the feature matrix obtained in the previous steps to obtain the disease prediction label of the test sample. Compare the disease prediction label with the category label of the test sample to obtain the classification evaluation index.

[0036] Preferably, a data preprocessing step is provided between obtaining the disease sample data set and randomly dividing it into a training set and a test set. The data preprocessing step includes sorting according to category labels and standardizing the data.

[0037] In order to achieve the above object, a second aspect of the present invention provides a disease recognition method based on relaxed block diagonal sparse embedding, comprising the following steps:

[0038] Acquire disease data to be identified, and perform feature extraction on the disease data to be identified based on the disease feature extraction method described in the first aspect to obtain a feature matrix;

[0039] Use the KNN classifier to predict and classify the feature matrix obtained in the previous step to obtain the disease prediction label of the disease data to be identified.

[0040] In order to achieve the above-mentioned object, the third aspect of the present invention provides a disease feature extraction model training device based on relaxed block diagonal sparse embedding, comprising:

[0041] An acquisition unit, configured to acquire a disease sample dataset, wherein the disease sample includes a category label and multiple feature indicators;

[0042] A construction unit is configured to construct a disease feature extraction model based on relaxed block diagonal sparse embedding, train the model using a disease sample dataset, and obtain a projection matrix; the disease feature extraction model includes a sparse representation, a block diagonal matrix, and a low-dimensional embedding matrix; iteratively update the sparse representation, the block diagonal matrix, and the low-dimensional embedding matrix to obtain an updated projection matrix; and apply the projection matrix obtained in the aforementioned steps to extract features from test samples in a test set to obtain a feature matrix;

[0043] The extraction unit is used to obtain target disease data, apply the projection matrix obtained in the above steps to perform feature extraction on the target disease data, and obtain a feature matrix.

[0044] To achieve the above-mentioned object, a fourth aspect of the present invention provides a computer device, comprising a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus;

[0045] Memory for storing computer programs;

[0046] The processor is used to implement the disease feature extraction method based on relaxed block diagonal sparse embedding as described in the first aspect, or implement the disease identification method based on relaxed block diagonal sparse embedding as described in the second aspect when executing the program stored in the memory.

[0047] In order to achieve the above-mentioned purpose, the fifth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, it implements the disease feature extraction method based on relaxed block diagonal sparse embedding as described in the first aspect, or implements the disease identification method based on relaxed block diagonal sparse embedding as described in the second aspect.

[0048] In order to achieve the above-mentioned purpose, the sixth aspect of the present invention provides a computer program product, including a computer program, which, when executed by a processor, implements the disease feature extraction method based on relaxed block diagonal sparse embedding as described in the first aspect, or implements the disease identification method based on relaxed block diagonal sparse embedding as described in the second aspect.

[0049] The beneficial effects of the present invention are:

[0050] Based on traditional sparse embedding, the present invention constructs a relaxed block diagonal matrix to generate a corresponding weight matrix, thereby enriching the details of disease data reconstruction and improving both the intra-class similarity and inter-class discrimination of the reconstructed data.

[0051] To address the problem that traditional methods using label information focus on fitting the label space to obtain discriminative information, this paper proposes a dual path of data reconstruction and projection to fully explore prior knowledge, thereby extracting discriminative features that are more conducive to classification.

[0052] This invention can not only improve the performance of disease diagnosis models, but also reduce the consumption of computing resources while ensuring accuracy, providing a new solution for the efficient processing of large-scale medical data. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is a flow chart of the disease feature extraction method of the present invention.

[0054] Figure 2 It is a structural schematic diagram of the disease feature extraction device of the present invention.

[0055] Figure 3 It is a structural schematic diagram of the computer device of the present invention. DETAILED DESCRIPTION

[0056] The technical solution of the present invention is further described in detail below through specific implementation methods.

[0057] Example 1

[0058] In order to achieve the above objectives, this embodiment provides a disease feature extraction method based on relaxed block diagonal sparse embedding, such as Figure 1 As shown, the following steps are included:

[0059] Step S1: Obtain a disease sample data set, wherein the disease sample includes a category label and multiple feature indicators. The disease sample includes a category label and multiple feature indicators.

[0060] Step S2: construct a disease feature extraction model based on relaxed block diagonal sparse embedding, use a disease sample data set to train the model, and obtain a projection matrix; the disease feature extraction model includes a sparse representation, a block diagonal matrix, and a low-dimensional embedding matrix.

[0061] In one embodiment, a disease feature extraction model based on relaxed block diagonal sparse embedding is constructed, including:

[0062] Design the objective function, which is expressed as follows:

[0063]

[0064] Where X=[x1,x2,...,x n ]∈R m×n is the sample matrix, m is the number of samples, n is the number of features; Q∈R m×c is the projection matrix, c< <m;Z∈R n×n is the coefficient matrix; ⊙ is the Hadamard-product operator; M is the dynamic relaxation matrix, B=[b1,b2,...,b n ]∈Rn×n is an ideal matrix, where b i =[0...,1,1,1,0,...] T ∈R n ; H=[h1,h2,...,h n ]∈R c ×n is a 0-1 label matrix, where h i =[0,0,…1,…,0,0] T , is x i The label vector of x i From the kth category, then h i The kth item is 1 and the rest are 0.

[0065] It is important to note that the first term in formula (1) measures the reconstruction error, i.e., the error between the projection of the original data and the reconstructed data. The second term makes the coefficient matrix Z as close to the relaxed block diagonal structure as possible. The third term promotes the row sparsity of the model, the fourth term focuses on the global information of the features, and the fifth term fits the label information, improving the model's discriminative ability in classification and regression tasks.

[0066] It can be understood that by using the relaxed block diagonal sparse embedding technology, combined with the reconstruction error and sparsity constraints, high-dimensional sample data can be projected into a low-dimensional subspace to obtain the optimal projection matrix.

[0067] It can be understood that in one embodiment, after obtaining the disease sample dataset, the disease sample dataset is further randomly divided into a training set and a test set; then in step S2, the training set is used to perform model training to obtain a projection matrix.

[0068] Furthermore, between obtaining the disease sample data set and randomly dividing it into a training set and a test set, a data preprocessing step is provided, and the data preprocessing step includes sorting according to category labels and standardizing the data.

[0069] In step S3, an alternating optimization method is used to iteratively update the sparse representation, the block diagonal matrix, and the low-dimensional embedding matrix to obtain an updated projection matrix.

[0070] The alternating optimization algorithm is a method that alternates between optimizations of multiple variables or models. It is often used to solve interdependent problems involving multiple variables or models to achieve optimal or suboptimal solutions. The basic steps of this method include initializing the variables or model parameters, then fixing one variable or model parameter while optimizing the others. The optimization is then performed sequentially, with each variable or model parameter fixed, until convergence is achieved.

[0071] This step utilizes the alternating optimization method to solve the three unknown quantities Z, M, and Q in the model of formula (1). Specifically, step S3 includes the following steps:

[0072] Fixing Z and M, equation (1) is transformed into:

[0073]

[0074] Derivative (2) and equal to zero, the final solution is:

[0075]

[0076] in is a diagonal matrix whose diagonal elements are Right now γ=(XZ-X)(XZ-X) T ;

[0077] Fixing Q and M, equation (1) is transformed into:

[0078]

[0079] Solving equation (4), we get:

[0080] Z=(X T QQ T X+I) -1 (X T QQ T X+B⊙M)(5);

[0081] Fixing Q and Z, equation (1) is transformed into:

[0082]

[0083] Let C = B-1, where 1 is a matrix of all 1s, then we get the equivalent formula:

[0084]

[0085] Solving equation (7), we get:

[0086] M=ZC⊙M(8).

[0087] It is understood that if the disease sample dataset is randomly divided into a training set and a test set in step S2, then in step S3, after obtaining the projection matrix, the projection matrix is ​​used to extract features from the test samples in the test set to obtain a feature matrix. The feature matrix obtained in the previous step is then predicted and classified using a KNN classifier to obtain disease prediction labels for the test samples. The disease prediction labels are then compared with the category labels of the test samples to obtain classification evaluation indicators. In specific implementations, the classification evaluation indicators include two evaluation indicators: average recognition accuracy and standard deviation.

[0088] Step S4: Obtain target disease data, apply the projection matrix obtained in the above steps to perform feature extraction on the target disease data, and obtain a feature matrix.

[0089] Data preprocessing steps can bring features to the same scale for subsequent comparison.

[0090] This embodiment combines data reconstruction, subspace projection and classification regression into a learning framework. By data reconstruction, the potential information of the data is mined and the projection matrix is ​​applied. 2,1 Norm constraints and F-norm constraints focus on both global and local information of features. A relaxed block diagonal matrix of reconstruction coefficients is constructed using prior knowledge. Relaxed block diagonal sparse embedding introduces more complex sparsity design and structural modeling, enhancing the correlation between samples or overall structural constraints. The projection matrix is ​​also learned by combining representation coefficients with discriminative information and regression functions. By fully incorporating these techniques, the accuracy of disease recognition is significantly improved.

[0091] This embodiment introduces relaxed block diagonal sparse embedding technology to effectively reduce redundant information in high-dimensional data and optimize the model's feature selection and classification performance. By optimizing the block diagonal structure, the sparsity of the embedded data is ensured, further improving the accuracy and robustness of feature selection.

[0092] This embodiment can not only improve the performance of the disease diagnosis model, but also reduce the consumption of computing resources while ensuring accuracy, providing a new solution for the efficient processing of large-scale medical data.

[0093] Example 2

[0094] This embodiment provides a disease recognition method based on relaxed block diagonal sparse embedding, comprising the following steps:

[0095] Obtaining disease data to be identified, performing feature extraction on the disease data to be identified based on the disease feature extraction method described in Example 1 to obtain a feature matrix;

[0096] Use the KNN classifier to predict and classify the feature matrix obtained in the previous step to obtain the disease prediction label of the disease data to be identified.

[0097] It is understandable that, in specific implementation, other classifiers may be used to perform prediction and classification on the feature matrix obtained in the above steps.

[0098] Furthermore, the disease recognition method proposed in this embodiment is experimented on multiple disease datasets, and the experimental results are compared and analyzed with other sparse embedding methods.

[0099] Specifically, the colon cancer dataset (colon) covers samples from 62 patients, each sample has 2,000 feature indicators and is accurately divided into two category labels based on the condition; the breast cancer diagnosis dataset (wdbc) includes sample data from 569 patients, each sample contains 30 feature indicators and is also divided into two label categories.

[0100] During the experiment, L training samples were randomly selected based on the characteristics of the different datasets, and the remaining patient samples were used as test samples. The L values ​​for the colon cancer dataset (colon) were set to 9, 11, and 13, and for the breast cancer diagnosis dataset (wdbc) were set to 30, 40, and 50. Using the same experimental settings, five widely recognized and representative methods were selected for comparison: LLRSE, RBD-DPL, RSLDA, RSLRE, and SBDCL. Two evaluation metrics were used to measure algorithm performance. The data in the table represents the mean recognition accuracy (± standard deviation).

[0101] The experimental results are as follows:

[0102] Table 1: Experimental results of various methods on the colon cancer dataset (colon)

[0103] 9 11 13 LLRSE 79.091(±6.759) 80.250(±4.923) 81.944(±6.709) RBD-DPL 82.955(±6.797) 84.750(±4.923) 86.111(±3.464) RSLDA 80.227(±5.031) 81.000(±4.743) 82.500(±2.635) RSLRE 76.136(±4.701) 77.250(±6.396) 79.722(±6.935) SBDCL 81.364(±5.750) 82.250(±5.062) 83.611(±4.431) RBDSE 85.000(±2.874) 85.833(±4.802) 86.500(±4.441)

[0104] Table 2: Experimental results of various methods on the breast cancer diagnosis dataset (wdbc)

[0105] 30 40 50 LLRSE 92.534(±1.989) 92.945(±1.596) 93.454(±1.464) RBD-DPL 92.692(±0.883) 93.149(±0.723) 93.667(±0.921) RSLDA 89.234(±1.401) 89.836(±1.472) 90.043(±1.092) RSLRE 89.077(±1.827) 89.387(±1.151) 89.829(±1.371) SBDCL 92.731(±1.060) 93.149(±1.162) 94.009(±0.692) RBDSE 93.556(±1.687) 94.438(±1.024) 94.606(±0.943)

[0106] By learning the projection matrix and constructing the relaxed block diagonal matrix, more complex sparsity design and structural modeling were introduced. After ten rounds of iteration based on the model, the average recognition accuracy and standard deviation were finally obtained, providing a basis for evaluating the performance of the algorithm.

[0107] From the results in Table 1 and Table 2, it can be seen that the method proposed in this embodiment has higher evaluation indicators than the other five methods. Experiments have verified that this embodiment can effectively improve the accuracy of disease diagnosis and recognition.

[0108] It should be understood that, although the steps in the flowcharts of the above embodiments are shown in sequence as indicated by the arrows, these steps are not necessarily performed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be performed in other orders. Moreover, at least a portion of the steps in the flowcharts of the above embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily performed at the same time, but can be performed at different times. The execution order of these steps or stages is not necessarily to be performed in sequence, but can be performed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0109] Example 3

[0110] Based on the same inventive concept, an embodiment of the present application also provides a disease feature extraction device based on relaxed block diagonal sparse embedding for implementing the above-mentioned disease feature extraction method based on relaxed block diagonal sparse embedding. The implementation solution provided by the device is similar to the implementation solution described in the above-mentioned method. Therefore, the specific limitations in one or more embodiments of the disease feature extraction device based on relaxed block diagonal sparse embedding provided below can be referred to the limitations of the disease feature extraction method based on relaxed block diagonal sparse embedding in Example 1, and will not be repeated here.

[0111] Specifically, the disease feature extraction device based on relaxed block diagonal sparse embedding is as follows: Figure 2 Shown, including:

[0112] An acquisition unit is used to acquire a disease sample data set and randomly divide it into a training set and a test set, wherein the disease sample includes a category label and multiple feature indicators;

[0113] A construction unit is configured to construct a disease feature extraction model based on relaxed block diagonal sparse embedding, train the model using a disease sample dataset, and obtain a projection matrix; the disease feature extraction model includes a sparse representation, a block diagonal matrix, and a low-dimensional embedding matrix; iteratively update the sparse representation, the block diagonal matrix, and the low-dimensional embedding matrix to obtain an updated projection matrix; and apply the projection matrix obtained in the aforementioned steps to extract features from test samples in a test set to obtain a feature matrix;

[0114] The extraction unit is used to obtain target disease data, apply the projection matrix obtained in the above steps to perform feature extraction on the target disease data, and obtain a feature matrix.

[0115] In one embodiment, after acquiring the disease sample dataset, the acquisition unit of the disease feature extraction device randomly divides the disease sample dataset into a training set and a test set;

[0116] The construction unit uses the training set to perform model training to obtain a projection matrix.

[0117] The disease feature extraction device also includes a verification unit, which is used to use a KNN classifier to predict and classify the feature matrix obtained in the above steps to obtain a disease prediction label for the test sample; and compare the disease prediction label with the category label of the test sample to obtain a classification evaluation index.

[0118] Example 4

[0119] This embodiment provides a computer device, which may be a terminal, and its internal structure diagram may be as follows: Figure 3 As shown. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. The processor, the memory and the input / output interface are connected via a system bus, and the communication interface, the display unit and the input device are connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, the disease feature extraction method based on relaxed block diagonal sparse embedding as described in Example 1 is implemented, or the disease identification method based on relaxed block diagonal sparse embedding as described in Example 2 is implemented.

[0120] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0121] Example 5

[0122] Based on the above embodiments, this embodiment provides a readable storage medium on which a computer program is stored. When the computer program is executed by a processor, it implements the disease feature extraction method based on relaxed block diagonal sparse embedding as described in Example 1, or implements the disease identification method based on relaxed block diagonal sparse embedding as described in Example 2.

[0123] Example 6

[0124] Based on the above embodiments, this embodiment provides a computer program product, including a computer program, which, when executed by a processor, implements the disease feature extraction method based on relaxed block diagonal sparse embedding as described in Example 1, or implements the disease identification method based on relaxed block diagonal sparse embedding as described in Example 2.

[0125] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiment methods can be implemented 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 memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may 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 may 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).

[0126] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to preferred embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or some technical features can be replaced by equivalents without departing from the spirit of the technical solutions of the present invention. They should all be included in the scope of the technical solutions claimed for protection by the present invention.

Claims

1. A disease feature extraction method based on relaxed block diagonal sparse embedding, characterized in that: The following steps are involved: Obtaining a disease sample dataset, wherein the disease samples include category labels and multiple feature indicators; Constructing a disease feature extraction model based on relaxed block diagonal sparse embedding, using a disease sample dataset for model training to obtain a projection matrix; the disease feature extraction model includes a sparse representation, a block diagonal matrix, and a low-dimensional embedding matrix; The alternating optimization method is used to iteratively update the sparse representation, block diagonal matrix, and low-dimensional embedding matrix to obtain the updated projection matrix. Obtain target disease data, apply the projection matrix obtained in the previous steps to perform feature extraction on the target disease data, and obtain a feature matrix.

2. The disease feature extraction method based on relaxed block diagonal sparse embedding according to claim 1, characterized in that: Build a disease feature extraction model based on relaxed block diagonal sparse embedding, including: Design the objective function, which is expressed as follows: Where X=[x1,x2,...,x n ]∈R m×n is the sample matrix, m is the number of samples, n is the number of features; Q∈R m×c is the projection matrix, c< <m;Z∈R n×n is the coefficient matrix; ⊙ is the Hadamard-product operator; M is the dynamic relaxation matrix, B=[b1,b2,...,b n ]∈R n×n is an ideal matrix, where b i =[0...,1,1,1,0,...] T ∈R n ; H=[h1,h2,...,h n ]∈R c ×n is a 0-1 label matrix, where h i =[0,0,…1,…,0,0] T , is x i The label vector, that is, if x i From the kth class, then h i Only the kth item is 1, and the rest are 0.

3. The disease feature extraction method based on relaxed block diagonal sparse embedding according to claim 2, characterized in that: The alternating optimization method is used to iteratively update the sparse representation, block diagonal matrix, and low-dimensional embedding matrix to obtain the updated projection matrix, including: Fixing Z and M, equation (1) is transformed into: Derivative (2) and equal to zero, the final solution is: in is a diagonal matrix whose diagonal elements are Right now γ=(XZ-X)(XZ-X) T ; Fixing Q and M, equation (1) is transformed into: Solving equation (4), we get: Z=(X T QQ T X+I) -1 (X T QQ T X+B⊙M) (5); Fixing Q and Z, equation (1) is transformed into: Let C = B-1, where 1 is a matrix of all 1s, then we get the equivalent formula: Solving equation (7), we get: M=ZC⊙M (8).

4. A disease feature extraction method based on relaxed block diagonal sparse embedding according to any one of claims 1 to 3, characterized in that: After obtaining the disease sample data set, the disease sample data set is randomly divided into a training set and a test set; Use the training set to train the model and obtain the projection matrix; Apply the projection matrix obtained in the previous steps to extract features from the test samples in the test set to obtain a feature matrix; Use the KNN classifier to predict and classify the feature matrix obtained in the previous steps to obtain the disease prediction label of the test sample. Compare the disease prediction label with the category label of the test sample to obtain the classification evaluation index.

5. The disease feature extraction method based on relaxed block diagonal sparse embedding according to claim 4, characterized in that: Between obtaining a disease sample data set and randomly dividing it into a training set and a test set, a data preprocessing step is also provided. The data preprocessing step includes sorting according to category labels and standardizing the data.

6. A disease recognition method based on relaxed block diagonal sparse embedding, characterized in that The following steps are involved: Acquire disease data to be identified, perform feature extraction on the disease data to be identified based on the disease feature extraction method according to claim 5, and obtain a feature matrix; Use the KNN classifier to predict and classify the feature matrix obtained in the previous step to obtain the disease prediction label of the disease data to be identified.

7. A disease feature extraction device based on relaxed block diagonal sparse embedding, characterized in that: include: An acquisition unit, used to acquire a disease sample data set; A construction unit is used to construct a disease feature extraction model based on relaxed block diagonal sparse embedding, use a disease sample data set to train the model, and obtain a projection matrix; the disease feature extraction model includes a sparse representation, a block diagonal matrix, and a low-dimensional embedding matrix; Iteratively update the sparse representation, block diagonal matrix, and low-dimensional embedding matrix to obtain an updated projection matrix; apply the projection matrix obtained in the above steps to extract features from the test samples in the test set to obtain a feature matrix; The extraction unit is used to obtain target disease data, apply the projection matrix obtained in the above steps to perform feature extraction on the target disease data, and obtain a feature matrix.

8. A computer device, characterized in that: The processor, the communication interface, the memory and the communication bus are connected to each other via the communication bus. Memory for storing computer programs; The processor is configured to implement the disease feature extraction method based on relaxed block diagonal sparse embedding as described in any one of claims 1 to 5, or implement the disease identification method based on relaxed block diagonal sparse embedding as described in claim 6 when executing the program stored in the memory.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements the disease feature extraction method based on relaxed block diagonal sparse embedding as described in any one of claims 1 to 5, or implements the disease identification method based on relaxed block diagonal sparse embedding as described in claim 6.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, it implements the disease feature extraction method based on relaxed block diagonal sparse embedding as described in any one of claims 1 to 5, or implements the disease identification method based on relaxed block diagonal sparse embedding as described in claim 6.

Images