Fatty liver health management method and system

Abstract

The invention provides a fatty liver health management method and system, and relates to the technical field of medical health management, and the method comprises the steps: obtaining a liver ultrasonic image; performing Radon transformation on the liver ultrasonic image, and constructing a projection matrix; performing two-dimensional discrete cosine transform on the projection matrix, and constructing a frequency characteristic matrix; performing feature extraction on the frequency feature matrix to obtain a one-dimensional feature vector; performing feature dimension reduction on the one-dimensional feature vector to obtain a plurality of local sensitive discriminant analysis components; performing feature selection on each local sensitivity discriminant analysis component through a maximum correlation minimum redundancy algorithm, and determining a target local sensitivity discriminant analysis component; constructing a fatty liver health management model based on a support vector machine; and inputting the target local sensitivity discriminant analysis component into a fatty liver health management model based on a support vector machine for analysis to obtain a fatty liver diagnosis prediction result. The method does not depend on clinical experience of doctors any more, and is easy to operate and high in diagnosis accuracy.

Description

Technical Field

[0001] The present invention relates to the field of medical health management technology, and in particular to a fatty liver health management method and system. Background Art

[0002] With the changes in modern lifestyles, people's dietary structure has also undergone a significant change. The popularity of high-fat, high-calorie diets, coupled with the lack of sufficient physical exercise, has made fatty liver a common liver disease. Fatty liver is caused by excessive fat deposition in the liver, which in turn leads to impaired liver function. If fatty liver is not discovered and effectively managed in a long-term manner, it may further develop into hepatitis, cirrhosis, and even liver cancer, seriously threatening people's lives and health. Therefore, early diagnosis and timely intervention of fatty liver are particularly important.

[0003] In the existing technology, traditional methods for diagnosing fatty liver are mainly through blood biochemical tests and liver imaging examinations. Although these methods can provide certain information on liver lesions, they often rely on the doctor's clinical experience, are cumbersome to operate, and have poor diagnostic accuracy.

[0004] In addition, when faced with large-scale imaging data, the existing automated diagnosis methods for fatty liver often fail to meet actual needs due to the huge amount of data, image noise and loss of detail information. Summary of the invention

[0005] In order to solve the technical problems that the traditional fatty liver diagnosis method relies on the clinical experience of doctors, is cumbersome to operate and has poor diagnostic accuracy, and the existing automated fatty liver diagnosis method is often difficult to meet actual needs in the face of large-scale imaging data due to the huge amount of data, image noise and loss of detail information, the present invention provides a fatty liver health management method and system.

[0006] The technical solution provided by the embodiment of the present invention is as follows:

[0007] First aspect:

[0008] An embodiment of the present invention provides a method for managing fatty liver health, comprising:

[0009] S1: Acquire ultrasound images of the liver;

[0010] S2: performing Radon transformation on the liver ultrasound image to construct a projection matrix;

[0011] S3: Perform a two-dimensional discrete cosine transform on the projection matrix to construct a frequency feature matrix;

[0012] S4: extracting features from the frequency feature matrix to obtain a one-dimensional feature vector;

[0013] S5: performing feature dimension reduction on the one-dimensional feature vector to obtain multiple local sensitive discriminant analysis components;

[0014] S6: performing feature selection on each of the local sensitive discriminant analysis components through a maximum relevance minimum redundancy algorithm to determine a target local sensitive discriminant analysis component;

[0015] S7: Construct a fatty liver health management model based on support vector machine;

[0016] S8: Inputting the target local sensitive discriminant analysis component into the fatty liver health management model based on support vector machine for analysis to obtain the fatty liver diagnosis prediction result.

[0017] Second aspect:

[0018] A fatty liver health management system provided by an embodiment of the present invention includes: a memory and one or more processors;

[0019] One or more application programs are stored in the memory, and the one or more application programs are suitable for being executed by the one or more processors to implement the above-mentioned fatty liver health management method.

[0020] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0021] In the present invention, a Radon transform is performed on a liver ultrasonic image to construct a projection matrix, a two-dimensional discrete cosine transform is performed on the projection matrix to construct a frequency feature matrix, feature extraction is performed on the frequency feature matrix to obtain a one-dimensional feature vector, feature dimension reduction is performed on the one-dimensional feature vector to obtain multiple local sensitive discriminant analysis components, and feature selection is performed on each local sensitive discriminant analysis component through a maximum relevance minimum redundancy algorithm to determine a target local sensitive discriminant analysis component. When faced with large-scale imaging data, the invention can effectively cope with the phenomena of huge data volume, image noise and loss of detail information, thereby ensuring that the diagnostic performance can meet actual needs. By inputting the target local sensitive discriminant analysis component into a fatty liver health management model based on a support vector machine for analysis, the invention no longer relies on the clinical experience of doctors, and the operation is simple and the diagnostic accuracy is high. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0023] Figure 1A schematic diagram of a process for fatty liver health management method provided by an embodiment of the present invention;

[0024] Figure 2 A schematic diagram of the structure of a fatty liver health management system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0025] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0026] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "example" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of the word "example" is intended to present the concept in a specific way. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or it can be either of the two.

[0027] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0028] Reference Manual Attached Figure 1 , showing a flow chart of a fatty liver health management method provided by an embodiment of the present invention.

[0029] The embodiment of the present invention provides a fatty liver health management method, which can be implemented by a fatty liver health management device, which can be a terminal or a server. The processing flow of the fatty liver health management method can include the following steps:

[0030] S1: Acquire ultrasound images of the liver.

[0031] S2: Perform Radon transform on the liver ultrasound image and construct a projection matrix.

[0032] Specifically, the liver ultrasound image is projected at different angles (0° to 180°) through Radon transform, different areas of the image are mapped to corresponding projection distances, and the projection results at various angles are accumulated to finally form a two-dimensional projection matrix.

[0033] It should be noted that Radon transform is a mathematical method that converts images from the spatial domain to the projection domain, and is widely used in image reconstruction and analysis. It projects a two-dimensional image along different angles and projects each part of the image information onto a straight line, thereby generating a set of one-dimensional projection data. Each projection data reflects the cumulative value of pixels in different directions in the image. Through this transformation, different viewing angle information of the image can be obtained from multiple angles. It is often used in CT scanning, medical imaging and other fields to help reconstruct and analyze the internal structure in the image.

[0034] In a possible implementation manner, S2 is specifically:

[0035] According to the following formula, the Radon transform of the liver ultrasound image is performed to construct the projection matrix:

[0036]

[0037] Among them, R() represents the projection matrix, r represents the projection distance, θ represents the projection angle, f() represents the pixel value of the pixel point in the liver ultrasound image, x represents the horizontal coordinate of the pixel point in the liver ultrasound image, y represents the vertical coordinate of the pixel point in the liver ultrasound image, δ represents the Dirac Delta function, cos represents the cosine operation, and sin represents the sine operation.

[0038] In the present invention, the main advantage of Radon transform is that it can obtain multi-angle information of the image by projecting the image at different angles (0° to 180°). This multi-angle projection can help fully capture the important structures and details in the image. Radon transform converts a two-dimensional image into one-dimensional projection data, which is essentially compressing the image. By accumulating projection results at different angles, redundant information can be reduced and the core structural features of the image can be retained. The projection of Radon transform at different angles can effectively suppress noise and some local interference signals.

[0039] S3: Perform a two-dimensional discrete cosine transform on the projection matrix to construct a frequency feature matrix.

[0040] It should be noted that the two-dimensional discrete cosine transform (2D-DCT) is a mathematical transform used to convert two-dimensional signals (such as images) from the spatial domain to the frequency domain. It represents the different frequency components of the image in the frequency domain by representing the pixel values ​​of the image as the weighted sum of a set of cosine basis functions. DCT can concentrate the low-frequency information in the image in the first few coefficients, while the high-frequency information (usually representing details or noise) is scattered in the following coefficients. Therefore, DCT is very useful in image compression because high-frequency components can be discarded to reduce the amount of data while retaining most of the image information. Two-dimensional discrete cosine transform is widely used in image compression, image processing, feature extraction and other fields.

[0041] In a possible implementation, S3 is specifically:

[0042] According to the following formula, the projection matrix is ​​subjected to a two-dimensional discrete cosine transform to construct a frequency feature matrix:

[0043]

[0044]

[0045] Among them, F() represents the frequency feature matrix, u represents the horizontal direction in the frequency feature matrix, v represents the vertical direction in the frequency feature matrix, a(u) represents the scaling factor of the two-dimensional discrete cosine transform in the horizontal direction, a(v) represents the scaling factor of the two-dimensional discrete cosine transform in the vertical direction, M represents the total number of projection distances, and N represents the total number of projection angles.

[0046] In the present invention, the two-dimensional discrete cosine transform (2D-DCT) can convert image data from the spatial domain to the frequency domain, so that the information in the image is represented by the frequency components. The spatial domain represents the pixel values ​​of the image, while the frequency domain represents the frequency components in the image (such as the low-frequency and high-frequency parts of the image). Through this conversion, the features in the image can be extracted and analyzed more efficiently. Compared with the original spatial domain image, the calculation of the frequency domain image is often more efficient. In the frequency domain, a lot of unimportant information (such as high-frequency noise) can be easily removed, and the remaining low-frequency information usually contains the core features of the image. This not only reduces the amount of calculation, but also speeds up the image analysis process.

[0047] S4: Extract features from the frequency feature matrix to obtain a one-dimensional feature vector.

[0048] In the present invention, by extracting features from the frequency feature matrix and converting them into a one-dimensional feature vector, the representation of the image can be simplified. The one-dimensional feature vector has a lower dimension, and compared with the original two-dimensional matrix, the one-dimensional vector can be processed and calculated faster and requires less storage.

[0049] In a possible implementation, S4 specifically includes:

[0050] S401: Starting from the upper left corner of the frequency feature matrix, scanning along the diagonal in a zigzag order in a zigzag scanning manner, extracting features from the frequency feature matrix to extract part of the frequency features in the frequency feature matrix.

[0051] Specifically, starting from the upper left corner of the frequency feature matrix, extract the elements on the first diagonal line in sequence, then scan to the upper right to extract the elements on the second diagonal line, and then scan to the lower left to extract the elements on the third diagonal line, and so on, extract the elements in the frequency feature matrix along the diagonal line in sequence to complete the scanning process of the entire matrix.

[0052] It should be noted that zigzag scanning is a method of traversing matrix elements in a specific order. Its basic idea is to start from a corner of the matrix, scan along the diagonal direction, and access each element in the matrix in turn along the zigzag path. Specifically, starting from the upper left corner, scan the first diagonal, then turn to the next diagonal, change the scanning direction, and continue scanning along the diagonal until the entire matrix is ​​traversed.

[0053] S402: Arrange the extracted partial frequency features in a scanning order to obtain a one-dimensional feature vector.

[0054] In the present invention, features are extracted by zigzag scanning, which can more effectively obtain the most representative frequency features from the matrix compared to other conventional row-by-row or column-by-column scanning methods. Through this feature extraction method, the scanning process helps to filter out redundant information in the image. By scanning the matrix in a zigzag order, it can be ensured that low-frequency features (usually containing the main structural information of the image) are preferentially extracted, while high-frequency features (usually related to details or noise) are processed in subsequent steps.

[0055] S5: Perform feature dimensionality reduction on the one-dimensional feature vector to obtain multiple local sensitive discriminant analysis components.

[0056] Among them, the local sensitive discriminant analysis component refers to the low-dimensional feature representation extracted from the feature vector through the local sensitive discriminant analysis (LSDA) algorithm, which can effectively distinguish different categories.

[0057] In the present invention, feature dimensionality reduction effectively reduces the computational complexity by reducing the dimension of the feature space. The reduced-dimensional features usually contain the key information that best reflects the category distinction, making the model's discrimination process clearer and more explainable.

[0058] In a possible implementation, S5 specifically includes sub-steps S501 to S505:

[0059] S501: constructing the intra-class graph and the inter-class graph of the one-dimensional feature vector respectively through the local sensitive discriminant analysis algorithm.

[0060] Among them, the intra-class graph refers to a graph that describes the similarities between samples of the same category, where the edge weights of the graph represent the similarities between samples, and the edge connections reflect the compactness between samples of the same category.

[0061] Among them, the between-class graph refers to a graph that describes the differences between samples of different categories, the edge weight represents the dissimilarity between samples of different categories, and the edge connection reflects the separability between categories.

[0062] It should be noted that Local Sensitivity Discriminant Analysis (LSDA) is a dimensionality reduction algorithm based on graph theory and discriminant analysis, which aims to improve the classification performance of data in low-dimensional space. It combines information on intra-class compactness and inter-class separability to improve the distinction between different categories while retaining the local structure of the data. LSDA captures the local relationship between data points by constructing a local graph structure, and uses discriminant analysis to optimize the distribution within and between classes, so that in the space after dimensionality reduction, data points of the same type are closer, while data points of different types are more separated.

[0063] S502: Establishing a first objective function for measuring the intra-class compactness of the intra-class graph and a second objective function for measuring the inter-class separability of the inter-class graph.

[0064] Among them, intra-class compactness refers to the similarity or closeness between samples in the same category. The higher the intra-class compactness, the more similar and closer the samples in the same category are to each other, that is, the samples in the same category are more closely clustered in the feature space.

[0065] Among them, inter-class separability refers to the difference or separation degree between samples of different categories. The higher the inter-class separability, the farther the distance between samples of different categories in the feature space, and the easier it is to distinguish between categories.

[0066] Optionally, the first objective function is specifically:

[0067]

[0068] Among them, J1 represents the first objective function, min represents the minimum value, and n i represents the i-th frequency feature in the one-dimensional feature vector after feature dimensionality reduction, n jIt represents the jth frequency feature in the one-dimensional feature vector after feature dimensionality reduction, and W1 represents the weight matrix of the intra-class graph.

[0069] Among them, each frequency feature in the one-dimensional feature vector after feature dimension reduction is specifically:

[0070] n i =α T p i

[0071] n j =α T p j

[0072] Among them, α represents the transformation matrix and T represents the transposition operation.

[0073] Among them, the weight matrix of the intra-class graph is specifically:

[0074]

[0075] Among them, p i represents the i-th frequency feature in the one-dimensional feature vector, p j Represents the jth frequency feature in the one-dimensional feature vector.

[0076] The second objective function is specifically:

[0077]

[0078] Among them, J2 represents the second objective function, max represents the minimum value, and W2 represents the weight matrix of the inter-class graph.

[0079] Among them, the weight matrix of the inter-class graph is specifically:

[0080]

[0081] S503: Optimizing the first objective function and the second objective function respectively by using the Laplace embedding algorithm.

[0082] It should be noted that the Laplacian Eigenmaps algorithm is an unsupervised learning method for data dimensionality reduction, which aims to maintain the local structure of the data while mapping high-dimensional data to a low-dimensional space. The algorithm captures the local geometric structure of the data by constructing a similarity graph between data points using the Laplacian Matrix. In this method, a graph is first constructed based on the similarity between data points, where the similarity can be defined by distance or neighbor relationship. Then, the Laplacian matrix is ​​used to describe the structural characteristics of the graph. By performing eigenvalue decomposition on the Laplacian matrix, a low-dimensional embedding space can be obtained, in which the local structure of the original data is effectively preserved, and the arrangement of data points in the low-dimensional space reflects their relationship in the high-dimensional space.

[0083] Specifically, according to the weight matrices of the intra-class graph and the inter-class graph, the Laplacian matrices of the intra-class graph and the inter-class graph are constructed respectively:

[0084] l1=d1-W1

[0085] l2=d2-W2

[0086] Among them, l1 represents the Laplacian matrix of the intra-class graph, d1 represents the degree matrix of the intra-class graph (the number of connections between each sample and samples of the same class), l2 represents the Laplacian matrix of the inter-class graph, and d2 represents the degree matrix of the inter-class graph (the number of connections between each sample and samples of different classes).

[0087] Furthermore, the Laplacian matrices of the constructed intra-class graph and inter-class graph are subjected to eigenvalue decomposition to obtain the optimized intra-class compactness and inter-class separability.

[0088] S504: Reconstruct the corresponding intra-class graph and inter-class graph respectively through the first objective function and the second objective function obtained by optimization.

[0089] S505: According to the reconstructed intra-class graph and inter-class graph, the one-dimensional feature vector is rearranged to complete the feature dimension reduction of the one-dimensional feature vector, and a plurality of local sensitive discriminant analysis components are obtained.

[0090] In the present invention, by constructing intra-class graphs and inter-class graphs and optimizing their objective functions, the classification performance of data in low-dimensional space is effectively improved, and the differences between samples can be more clearly reflected. The process of feature dimensionality reduction helps to remove redundant features, reduce noise interference, and improve the accuracy of the classification model. Feature dimensionality reduction is performed by local sensitive discriminant analysis (LSDA), combined with the Laplace embedding algorithm, while maintaining the local structure of the data, the intra-class compactness and inter-class separability can be optimized, thereby improving the classification performance, reducing redundancy, reducing noise interference, enhancing the model stability, and improving the computational efficiency.

[0091] S6: Perform feature selection on each local sensitive discriminant analysis component through the maximum relevance minimum redundancy algorithm to determine the target local sensitive discriminant analysis component.

[0092] It should be noted that the Maximum Relevance Minimum Redundancy (mRMR) algorithm is a feature selection method that aims to select the most discriminative features from high-dimensional data. The core idea is to maximize the correlation between the selected features and the target variable while minimizing the redundancy between features. Specifically, the mRMR algorithm first evaluates the correlation between each feature and the target category, and then calculates the redundancy between features (i.e., the similarity between features) to ensure that the features in the selected feature set can provide rich information without containing too much repeated information.

[0093] In the present invention, the maximum relevance minimum redundancy algorithm can effectively select the most discriminative and least redundant features from multiple local sensitive discriminant analysis (LSDA) components. Through this process, the accuracy, computational efficiency and generalization ability of the model can be significantly improved, while reducing the impact of overfitting and noise.

[0094] In a possible implementation, S6 specifically includes sub-steps S601 to S603:

[0095] S601: Calculate the correlation index between each local sensitive discriminant analysis component and the target category, and sort them in ascending order:

[0096] I1(m i ,c)=H(m i )+H(c)-H(m i ,c)

[0097] Among them, I1(m i ,c) represents the correlation index between the i-th local sensitive discriminant analysis component and the target category, H(m i ) represents the entropy of the i-th local sensitive discriminant analysis component, H(c) represents the entropy of the target category, H(m i ,c) represents the joint entropy of the i-th local sensitive discriminant analysis component and the target category.

[0098] Among them, the correlation index is a measure of the relationship between the feature and the target category, reflecting the importance of the feature in distinguishing different categories. The higher the correlation, the stronger the feature's ability to distinguish the target category.

[0099] S602: Calculate the redundancy index between each local sensitive discriminant analysis component and sort them in ascending order:

[0100] I2(m i ,m j )=H(m i )+H(m j )-H(m i ,m j )

[0101] Among them, I2(m i ,m j ) represents the redundancy index between the i-th local sensitive discriminant analysis component and the j-th local sensitive discriminant analysis component, H(m j ) represents the entropy of the jth local sensitive discriminant analysis component, H(m i ,m j ) represents the joint entropy of the i-th local sensitive discriminant analysis component and the j-th local sensitive discriminant analysis component.

[0102] Among them, the redundancy index measures the similarity between different features, indicating whether the information they provide is repeated. The lower the redundancy, the stronger the information complementarity between the features, which helps to improve the efficiency and accuracy of the model.

[0103] S603: Selecting the local sensitive discriminant analysis component with the maximum correlation index and the minimum redundancy index as the target local sensitive discriminant analysis component.

[0104] In the present invention, by calculating the correlation index between each local sensitive discriminant analysis component and the target category, the strength of the relationship between each feature (component) and the category can be evaluated, and the component with the largest correlation index can be selected, which can ensure that the features that best reflect the category differences are selected, and ultimately improve the discriminative ability of the classifier. By calculating the redundancy index between each local sensitive discriminant analysis component, the degree of information duplication between the components can be measured, and components with lower redundancy can be selected, which can effectively remove features that provide duplicate information and reduce the complexity of the model.

[0105] S7: Construct a fatty liver health management model based on support vector machine.

[0106] It should be noted that Support Vector Machine (SVM) is a machine learning algorithm commonly used for classification and regression analysis, and is particularly good at processing high-dimensional data. The basic idea is to achieve classification by finding an optimal hyperplane to separate samples of different categories in the feature space. The goal of SVM is to maximize the margin (that is, the distance between the hyperplane and the nearest sample point) to ensure the accuracy and robustness of the classification. This optimal hyperplane is determined by key sample points called support vectors, which are the training data points closest to the hyperplane.

[0107] In the present invention, the support vector machine performs well in the binary classification task, especially when the sample distribution is complex, SVM can find the best decision boundary in the high-dimensional space by selecting the appropriate kernel function. By constructing a fatty liver health management model based on support vector machine (SVM), not only can the classification accuracy be improved, making the health management of fatty liver more accurate, but also it can show strong ability in processing high-dimensional data, ensuring that the model has robustness and generalization ability.

[0108] S8: Input the target local sensitive discriminant analysis component into the fatty liver health management model based on support vector machine for analysis to obtain the fatty liver diagnosis prediction result.

[0109] In a possible implementation manner, S8 specifically includes:

[0110] According to the following formula, the target local sensitive discriminant analysis component is input into the fatty liver health management model based on support vector machine for analysis to obtain the fatty liver diagnosis prediction result:

[0111]

[0112] Among them, C() represents the prediction result of fatty liver diagnosis, ω0 represents the weight vector of the support vector machine, T represents the transpose operation, m t represents the target local sensitive discriminant analysis component, and b represents the bias term of the support vector machine.

[0113] In the present invention, the SVM model distinguishes samples of different categories by finding the optimal hyperplane. When the target components processed by LSDA are input, SVM can use these efficient feature representations for classification, and can more accurately define the category boundaries, thereby improving the diagnostic ability of fatty liver. Inputting the target local sensitive discriminant analysis (LSDA) component into the fatty liver health management model based on support vector machine (SVM) can give full play to the advantages of LSDA in dimensionality reduction and feature optimization, ensuring that the SVM model can use the most discriminative features for accurate classification.

[0114] In a possible implementation, the fatty liver health management method further includes:

[0115] S9: With the goal of minimizing the mean square error loss function, the support vector machine is trained through the krill swarm optimization algorithm to determine the optimal parameters of the support vector machine.

[0116] It should be noted that the mean squared error (MSE) loss function is a commonly used loss function in regression problems, which measures the difference between the predicted value and the true value. It evaluates the prediction performance of the model by calculating the square of the difference between the predicted value and the true value. By minimizing the mean squared error loss function, the predicted value of the model can be as close to the true value as possible, thereby optimizing the performance of the model. The mean squared error loss function has the characteristic of penalizing large errors, so it can effectively drive the model to learn more accurate predictions.

[0117] It should be noted that Shrimp Swarm Optimization (SSO) is an intelligent optimization algorithm that simulates the foraging behavior of krill groups. It is inspired by the collective behavior of krill in finding food in nature. Krill groups find the optimal foraging area through collaboration and interaction. The algorithm imitates this process and uses multiple "krill individuals" to explore and develop in the search space to find the optimal solution to the problem. In the optimization process, each krill individual represents a potential solution, and through mutual attraction and information sharing, the collective moves towards the global optimal solution. The algorithm combines the local search of individuals with the global search of the group, which can effectively avoid falling into the local optimal solution and has a strong global search capability.

[0118] In the present invention, the mean square error loss function (MSE) is used as a common loss function in regression problems. By minimizing the difference between the predicted value and the true value, the prediction ability of the model can be effectively optimized. During the training process of the support vector machine (SVM), the selection of hyperparameters is very critical to whether the model is overfitting. By optimizing the parameters of the SVM during the training phase using the krill swarm optimization algorithm, the overfitting problem can be effectively avoided.

[0119] In a possible implementation, S9 specifically includes sub-steps S901 and S902:

[0120] S901: With the goal of minimizing the mean square error loss function, construct the objective function:

[0121]

[0122] Where min means taking the minimum value, MSE means the objective function, θ' means the model parameter set of the support vector machine, S means the total number of liver ultrasound images, and Y s represents the true diagnosis result of the sth liver ultrasound image, represents the actual diagnosis result of the sth liver ultrasound image.

[0123] S902: According to the objective function, the support vector machine is trained by using a krill swarm optimization algorithm to determine optimal parameters of the support vector machine.

[0124] In a possible implementation, S902 specifically includes sub-steps S9021 to S9022:

[0125] S9021: Initialize parameters and set the maximum number of iterations of the krill swarm optimization algorithm.

[0126] S9022: Generate an initial population through Logistic chaos mapping. The initial population includes multiple krill individuals, each of which represents a feasible set of model parameters of a support vector machine:

[0127] L i (t+1) = γL i (t)·(1-L i (t))

[0128] Among them, L i (t+1) represents the position of the i-th krill individual at the t+1 iteration, γ represents the branching coefficient with a value of 3.95, and L i (t) represents the position of the i-th krill individual at the tth iteration.

[0129] It should be noted that Logistic Chaotic Map is a nonlinear dynamic system based on Logistic function, which is used to generate a series with chaotic characteristics. Given the initial values ​​and parameters, the output of Logistic Map will produce complex and unpredictable results, and is very sensitive to the initial conditions, with chaotic characteristics. This characteristic makes it possible to generate pseudo-random numbers, which are widely used in cryptography, random process modeling, optimization algorithms (such as initial population generation in krill swarm optimization algorithm) and other fields.

[0130] S9023: Taking the objective function as the fitness function, calculating the fitness value of each krill individual, and taking the krill individual with the maximum fitness value as the current krill individual.

[0131] S9024: Adaptively adjust the inertia weight of foraging motion and induced motion through cosine control factors:

[0132]

[0133] Among them, ω f represents the inertia weight of the foraging motion after adaptive adjustment, ω n represents the inertia weight of the induced motion after adaptive adjustment, ω max represents the maximum value of the inertia weight, ω min represents the minimum value of the inertia weight, T represents the maximum number of iterations, and P represents a random number in the range of (0,1).

[0134] It should be noted that the cosine control factor is a factor used to adjust the search strategy in the optimization algorithm, usually used to dynamically adjust the step size or inertia weight of the search. During the optimization process, the cosine control factor gradually reduces the step size or adjusts the exploration ability of the algorithm according to the changes in the cosine function. Specifically, the fluctuation characteristics of the cosine function allow the optimization algorithm to gradually reduce the search range as the iteration progresses, so that the algorithm can conduct extensive exploration in the early stage and focus on fine local search in the later stage.

[0135] S9025: Adaptively adjust the foraging motion and the induced motion according to the inertia weights of the adaptively adjusted foraging motion and the induced motion:

[0136]

[0137] F i =V f β i +w f F i old

[0138]

[0139] Among them, F′ i represents the foraging movement of the i-th krill individual after adaptive adjustment, F i represents the foraging movement of the i-th krill individual, V f represents the foraging speed, β i represents the foraging direction of the i-th krill individual, F i old represents the last foraging movement of the i-th krill individual, represents the current best foraging direction to attract the i-th krill individual, β i best represents the historical best foraging direction to attract the i-th krill individual, K′ i represents the induced motion of the i-th krill individual after adaptive adjustment, K i represents the induced motion of the i-th krill individual, K max represents the maximum induced velocity, a i represents the induction factor of the i-th krill individual, represents the last induced movement of the i-th krill individual.

[0140] S9026: Update the current position of the krill individual according to the adaptively adjusted foraging movement and induced movement:

[0141]

[0142] Among them, L i (t+Δt) represents the position of the i-th krill individual at the t+Δt-th iteration, Δt represents the step length when the krill individual is updated, and D i represents the random diffusion rate of the i-th krill individual, D max represents the maximum diffusion speed, and δ represents the random disturbance factor with a value range of [-1,1].

[0143] In a possible implementation manner, after S9026 and before S9027, the process further includes:

[0144] S9028: Update the position of the current krill individual through the Cauchy mutation operation:

[0145]

[0146] in, represents the position of krill individuals after Cauchy mutation, represents the position of the i-th krill individual after Cauchy mutation at the t-th iteration, and C(0,1) represents a Cauchy distribution random variable.

[0147] It should be noted that the Cauchy Mutation operation is a random mutation method that is often used in optimization algorithms to introduce randomness and enhance the diversity of the search. In this operation, the mutation process is based on the Cauchy distribution, which is a probability distribution with a thick tail, meaning that it allows for large random fluctuations or jumps. This property enables the Cauchy mutation to jump out of the local optimal solution during the optimization process, increase the scope of exploration, and avoid falling into the local optimum. Specifically, the Cauchy mutation operation updates the individual's position by combining the current individual's position with a random number sampled from the Cauchy distribution, thereby introducing a larger amplitude of variation.

[0148] S9027: Repeat sub-steps S9024 to S9026 until the maximum number of iterations is reached.

[0149] In the present invention, the parameters of the SVM model can be efficiently optimized by the krill swarm optimization algorithm (SSO). SSO simulates the foraging behavior of krill swarms to perform global search, which can avoid local optimal solutions and ensure that the global optimal model parameters are finally obtained. In the optimization process, mechanisms such as Logistic chaos mapping and Cauchy mutation operation introduce additional randomness, which enables the algorithm to effectively avoid local optimal solutions in the optimization process. The optimization process of SSO can effectively control the exploration and development balance of the search by adaptively adjusting the inertia weight and the variation amplitude. This dynamic adjustment mechanism helps to avoid overfitting.

[0150] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0151] In the present invention, a Radon transform is performed on a liver ultrasonic image to construct a projection matrix, a two-dimensional discrete cosine transform is performed on the projection matrix to construct a frequency feature matrix, feature extraction is performed on the frequency feature matrix to obtain a one-dimensional feature vector, feature dimension reduction is performed on the one-dimensional feature vector to obtain multiple local sensitive discriminant analysis components, and feature selection is performed on each local sensitive discriminant analysis component through a maximum relevance minimum redundancy algorithm to determine a target local sensitive discriminant analysis component. When faced with large-scale imaging data, the invention can effectively cope with the phenomena of huge data volume, image noise and loss of detail information, thereby ensuring that the diagnostic performance can meet actual needs. By inputting the target local sensitive discriminant analysis component into a fatty liver health management model based on a support vector machine for analysis, the invention no longer relies on the clinical experience of doctors, and the operation is simple and the diagnostic accuracy is high.

[0152] Reference Manual Attached Figure 2 , showing a structural schematic diagram of a fatty liver health management system provided by the present invention.

[0153] The present invention further provides a fatty liver health management system 30 , comprising: a memory 303 and one or more processors 301 .

[0154] One or more application programs are stored in the memory 303 , and the one or more application programs are suitable for being executed by the one or more processors 301 to implement the fatty liver health management method described in the method embodiment.

[0155] The fatty liver health management system 30 includes: a processor 301 and a memory 303. The processor 301 and the memory 303 are connected, for example, via a bus 302.

[0156] The structure of the fatty liver health management system 30 does not constitute a limitation on the embodiment of the present invention.

[0157] Processor 301 may be a CPU, a general purpose processor, a DSP, an ASIC, an FPGA or other programmable logic device, a transistor logic device, a hardware component or any combination thereof. It may implement or execute various exemplary logic blocks, modules and circuits described in conjunction with the disclosure of the present invention. Processor 301 may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0158] The bus 302 may include a path to transmit information between the above components. The bus 302 may be a PCI bus or an EISA bus, etc. The bus 302 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0159] The memory 303 can be a ROM or other type of static storage device that can store static information and instructions, a RAM or other type of dynamic storage device that can store information and instructions, or an EEPROM, a CD-ROM or other optical disk storage, an optical disk storage (including a compressed optical disk, a laser disk, an optical disk, a digital versatile disk, a Blu-ray disk, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited to these.

[0160] It should be noted that the fatty liver health management system 30 can implement the above-mentioned fatty liver health management method and can achieve the same or similar technical effects. To avoid repetition, the present invention will not go into details.

[0161] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0162] In the present invention, a Radon transform is performed on a liver ultrasonic image to construct a projection matrix, a two-dimensional discrete cosine transform is performed on the projection matrix to construct a frequency feature matrix, feature extraction is performed on the frequency feature matrix to obtain a one-dimensional feature vector, feature dimension reduction is performed on the one-dimensional feature vector to obtain multiple local sensitive discriminant analysis components, and feature selection is performed on each local sensitive discriminant analysis component through a maximum relevance minimum redundancy algorithm to determine a target local sensitive discriminant analysis component. When faced with large-scale imaging data, the invention can effectively cope with the phenomena of huge data volume, image noise and loss of detail information, thereby ensuring that the diagnostic performance can meet actual needs. By inputting the target local sensitive discriminant analysis component into a fatty liver health management model based on a support vector machine for analysis, the invention no longer relies on the clinical experience of doctors, and the operation is simple and the diagnostic accuracy is high.

[0163] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

[0164] There are a few points to note:

[0165] (1) The drawings of the embodiments of the present invention only relate to the structures related to the embodiments of the present invention, and other structures may refer to the general design.

[0166] (2) For the sake of clarity, in the drawings used to describe the embodiments of the present invention, the thickness of the layers or regions is exaggerated or reduced, that is, these drawings are not drawn according to the actual scale. It is understood that when an element such as a layer, film, region or substrate is referred to as being "on" or "under" another element, the element may be "directly" "on" or "under" the other element or there may be intermediate elements.

[0167] (3) In the absence of conflict, the embodiments of the present invention and the features therein may be combined with each other to obtain new embodiments.

[0168] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. The protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for healthy management of fatty liver, characterized in that: include: S1: Acquire ultrasound images of the liver; S2: performing Radon transformation on the liver ultrasound image to construct a projection matrix; S3: Perform a two-dimensional discrete cosine transform on the projection matrix to construct a frequency feature matrix; S4: extracting features from the frequency feature matrix to obtain a one-dimensional feature vector; S5: performing feature dimension reduction on the one-dimensional feature vector to obtain multiple local sensitive discriminant analysis components; S6: performing feature selection on each of the local sensitive discriminant analysis components through a maximum relevance minimum redundancy algorithm to determine a target local sensitive discriminant analysis component; S7: Construct a fatty liver health management model based on support vector machine; S8: Inputting the target local sensitive discriminant analysis component into the fatty liver health management model based on support vector machine for analysis to obtain the fatty liver diagnosis prediction result.

2. The fatty liver health management method according to claim 1, characterized in that: The S2 is specifically: According to the following formula, the liver ultrasound image is subjected to Radon transformation to construct a projection matrix: Among them, R() represents the projection matrix, r represents the projection distance, θ represents the projection angle, f() represents the pixel value of the pixel point in the liver ultrasound image, x represents the horizontal coordinate of the pixel point in the liver ultrasound image, y represents the vertical coordinate of the pixel point in the liver ultrasound image, δ represents the Dirac Delta function, cos represents the cosine operation, and sin represents the sine operation.

3. The fatty liver health management method according to claim 1, characterized in that: The S3 is specifically: According to the following formula, the projection matrix is ​​subjected to a two-dimensional discrete cosine transform to construct a frequency feature matrix: Among them, F() represents the frequency feature matrix, u represents the horizontal direction in the frequency feature matrix, v represents the vertical direction in the frequency feature matrix, a(u) represents the scaling factor of the two-dimensional discrete cosine transform in the horizontal direction, a(v) represents the scaling factor of the two-dimensional discrete cosine transform in the vertical direction, M represents the total number of projection distances, and N represents the total number of projection angles.

4. The fatty liver health management method according to claim 1, characterized in that: The S4 specifically includes: S401: starting from the upper left corner of the frequency feature matrix, scanning along the diagonal in a zigzag order in a zigzag scanning manner, performing feature extraction on the frequency feature matrix to extract part of the frequency features in the frequency feature matrix; S402: Arrange the extracted partial frequency features in a scanning order to obtain the one-dimensional feature vector.

5. The fatty liver health management method according to claim 1, characterized in that: The S5 specifically includes: S501: constructing an intra-class graph and an inter-class graph of the one-dimensional feature vector respectively by using a local sensitive discriminant analysis algorithm; S502: establishing a first objective function for measuring the intra-class compactness of the intra-class graph and a second objective function for measuring the inter-class separability of the inter-class graph; S503: optimizing the first objective function and the second objective function respectively by using a Laplace embedding algorithm; S504: Reconstructing the corresponding intra-class graph and inter-class graph respectively through the first objective function and the second objective function obtained by optimization; S505: According to the reconstructed intra-class graph and inter-class graph, the one-dimensional feature vector is rearranged to complete the feature dimension reduction of the one-dimensional feature vector, and a plurality of local sensitive discriminant analysis components are obtained.

6. The fatty liver health management method according to claim 5, characterized in that: The first objective function is specifically: Among them, J1 represents the first objective function, min represents the minimum value, and n i represents the i-th frequency feature in the one-dimensional feature vector after feature dimensionality reduction, n j represents the jth frequency feature in the one-dimensional feature vector after feature dimensionality reduction, and W1 represents the weight matrix of the intra-class graph; The second objective function is specifically: Among them, J2 represents the second objective function, max represents the minimum value, and W2 represents the weight matrix of the inter-class graph.

7. The fatty liver health management method according to claim 1, characterized in that: The S6 specifically includes: S601: Calculate the correlation index between each of the local sensitive discriminant analysis components and the target category, and sort them in ascending order: I1(m i ,c)=H(m i )+H(c)-H(m i ,c) Among them, I1(m i ,c) represents the correlation index between the i-th local sensitive discriminant analysis component and the target category, H(m i ) represents the entropy of the i-th local sensitive discriminant analysis component, H(c) represents the entropy of the target category, H(m i ,c) represents the joint entropy of the i-th local sensitive discriminant analysis component and the target category; S602: Calculate the redundancy index between the local sensitive discriminant analysis components and sort them in ascending order: I2(m i ,m j )=H(m i )+H(m j )-H(m i ,m j ) Among them, I2(m i ,m j ) represents the redundancy index between the i-th local sensitive discriminant analysis component and the j-th local sensitive discriminant analysis component, H(m j ) represents the entropy of the jth local sensitive discriminant analysis component, H(m i ,m j ) represents the joint entropy of the i-th local sensitive discriminant analysis component and the j-th local sensitive discriminant analysis component; S603: Selecting the local sensitive discriminant analysis component with the maximum correlation index and the minimum redundancy index as the target local sensitive discriminant analysis component.

8. The fatty liver health management method according to claim 1, characterized in that: The S8 is specifically: According to the following formula, the target local sensitive discriminant analysis component is input into the fatty liver health management model based on support vector machine for analysis to obtain the fatty liver diagnosis prediction result: Among them, C() represents the prediction result of fatty liver diagnosis, ω0 represents the weight vector of the support vector machine, T represents the transpose operation, m t represents the target local sensitive discriminant analysis component, and b represents the bias term of the support vector machine.

9. The fatty liver health management method according to claim 1, characterized in that: Also includes: S9: With the goal of minimizing the mean square error loss function, the support vector machine is trained by a krill swarm optimization algorithm to determine the optimal parameters of the support vector machine.

10. A fatty liver health management system, characterized in that: include: memory and one or more processors; One or more application programs are stored in the memory, and the one or more application programs are suitable for being executed by the one or more processors to implement the fatty liver health management method according to any one of claims 1 to 9.

Images