Metabolic chronic disease nursing data supervision system based on big data

Abstract

The application discloses a metabolic chronic disease nursing data supervision system based on big data and belongs to the technical field of big data processing. The system comprises a nursing data acquisition module, a nursing feature tensorization construction module, a block self-adaptive feature noise reduction module, a bidirectional associated feature enhancement module, a disease risk grade classification module and a metabolic chronic disease nursing data supervision module. The application specifically performs singular value decomposition on a block diagonal matrix, constructs an enhanced optimization objective function, obtains a low-rank matrix corresponding to each business slice, and improves the overall accuracy of metabolic chronic disease nursing data supervision. The nursing associated adjacency matrix is used as a mask to obtain an intermediate associated weight matrix, a joint optimization objective function is constructed, a Lagrange multiplier is introduced to derive a closed expression, and a dichotomy method is used to solve the problem, a bidirectional aggregation enhanced feature matrix is constructed, and the credibility and effectiveness of metabolic chronic disease nursing data whole-process risk supervision are effectively improved.

Description

Technical Field

[0001] This invention belongs to the field of big data processing technology, specifically referring to a big data-based monitoring system for the care of metabolic chronic diseases. Background Technology

[0002] The metabolic chronic disease nursing data monitoring system utilizes big data analytics and intelligent algorithms to integrate multi-dimensional patient management data, assess disease risk levels in real time, provide decision support for healthcare professionals, and help optimize personalized care plans, thereby improving the efficiency of metabolic chronic disease management. However, existing metabolic chronic disease nursing data monitoring systems suffer from several problems. Firstly, redundant noise is easily introduced during data collection. Secondly, the contribution of different feature indicators varies significantly, and traditional noise reduction methods struggle to distinguish the importance of information from different dimensions, leading to the weakening of key features and residual noise interfering with the accuracy of subsequent nursing data monitoring. Furthermore, existing systems exhibit complex patient group relationships, making it difficult to effectively reflect the true disease correlations. Additionally, they separate the correlations between patient samples from the correlations between nursing indicators, resulting in insufficient expression of disease risk classification features and reduced credibility of nursing data monitoring. Summary of the Invention

[0003] To address the aforementioned issues and overcome the shortcomings of existing technologies, this invention provides a big data-based monitoring system for metabolic chronic disease nursing care. Addressing the problems of existing systems where redundant noise easily gets mixed in during data collection, and where the contribution of different feature indicators varies significantly, traditional noise reduction methods struggle to distinguish the importance of information from different dimensions, leading to weakened key features and residual noise interfering with the accuracy of subsequent nursing data monitoring, this solution obtains a block-based diagonal matrix by dividing the data into three business segments. Singular value sequences are obtained through singular value decomposition, adaptive weight coefficients are calculated, and a weighted nuclear norm regularization term is constructed. This term, together with the approximation error term, forms an enhanced optimization objective function. The proximal gradient descent method is used iteratively to obtain the low-rank matrix corresponding to each business segment. These are then integrated to obtain a two-dimensional patient nursing low-rank feature matrix, providing accurate and reliable data support for subsequent risk level classification, intelligent early warning, and other full-process monitoring work, comprehensively improving... To improve the overall accuracy of monitoring and supervision of nursing data for metabolic chronic diseases, this solution addresses the problems in existing systems. These include complex patient group relationships that fail to effectively reflect the true disease progression, and the separation of relationships between patient samples from those between nursing indicators, leading to insufficient expression of disease risk classification features and reduced credibility of nursing data monitoring. This solution uses the nursing association adjacency matrix as a mask to obtain the intermediate association weight matrix, calculates the second-order neighborhood association matrix, constructs a joint optimization objective function under non-negativity and row sum constraints, introduces Lagrange multipliers to derive a closed-form expression, and uses a bisection method to solve for the standardized patient association matrix. Through neighborhood information aggregation, it obtains the patient aggregated feature matrix and the indicator aggregated feature matrix, constructing a bidirectional aggregation enhanced feature matrix. This significantly strengthens the deep feature expression capability of chronic disease nursing data, effectively improving the credibility and effectiveness of risk monitoring throughout the entire process of metabolic chronic disease nursing data.

[0004] The present invention provides a big data-based monitoring system for metabolic chronic disease nursing care, which includes a nursing data acquisition module, a nursing feature tensor quantification construction module, a block adaptive feature denoising module, a bidirectional correlation feature enhancement module, a disease risk level classification module, and a metabolic chronic disease nursing data monitoring module.

[0005] The nursing data acquisition module collects historical nursing data of patients with metabolic chronic diseases and constructs a nursing dataset for patients with metabolic chronic diseases.

[0006] The nursing feature tensor construction module constructs a two-dimensional patient nursing feature matrix and a nursing association adjacency matrix, which are then extended to obtain a three-dimensional patient nursing tensor.

[0007] The block-adaptive feature denoising module obtains a block-diagonal matrix according to three types of business slices, obtains a singular value sequence through singular value decomposition, calculates adaptive weight coefficients, constructs a weighted nuclear norm regularization term, and together with the approximation error term, forms an enhanced optimization objective function. Iteratively solves the problem using the proximal gradient descent method to obtain the low-rank matrix corresponding to each business slice, and integrates them to obtain a two-dimensional patient care low-rank feature matrix.

[0008] The bidirectional association feature enhancement module uses the nursing association adjacency matrix as a mask to obtain the intermediate association weight matrix, calculates the second-order neighborhood association matrix, constructs a joint optimization objective function, introduces Lagrange multipliers to derive a closed expression, and uses the bisection method to solve it to obtain the standardized patient association matrix. Through neighborhood information aggregation, it obtains the patient aggregation feature matrix and the indicator aggregation feature matrix, and constructs a bidirectional aggregation enhancement feature matrix.

[0009] The disease risk level classification module calculates a low-dimensional latent feature matrix and generates a predicted risk label vector.

[0010] The metabolic chronic disease nursing data monitoring module generates risk warning messages and pushes them to the medical management terminal.

[0011] Furthermore, the nursing data acquisition module collects historical nursing data of patients with metabolic chronic diseases, including vital signs data, laboratory test data, and medication and lifestyle data; and performs data encoding, data normalization, and disease risk level labeling on the collected historical nursing data of patients with metabolic chronic diseases to construct a nursing dataset for patients with metabolic chronic diseases.

[0012] Furthermore, the nursing feature tensor construction module constructs a two-dimensional patient nursing feature matrix based on the nursing dataset of patients with metabolic chronic diseases, and constructs a nursing association adjacency matrix according to the cosine similarity of the features. According to the division of three types of business, namely physical signs, test indicators, medication and lifestyle behaviors, the two-dimensional patient nursing feature matrix is ​​extended into a three-dimensional patient nursing tensor.

[0013] Furthermore, the block-based adaptive feature denoising module specifically includes the following:

[0014] Block Singular Value Decomposition Unit: The three-dimensional patient care tensor is sliced ​​into three types of business segments and converted into a block diagonal matrix. Singular value decomposition is performed on the block diagonal matrix to obtain the singular value sequence and its corresponding left and right singular vectors for each business segment.

[0015] An adaptive weighted noise reduction unit is used. Based on the singular value sequence corresponding to each business slice, the adaptive weight coefficient of each singular value is calculated. A weighted nuclear norm regularization term is constructed on each business slice, which together with the approximation error term forms an enhanced optimization objective function. The proximal gradient descent method is used to iteratively solve the function to obtain the low-rank matrix corresponding to each business slice. The low-rank matrices obtained from solving each business slice are spliced ​​and integrated according to the original business order to obtain a two-dimensional patient care low-rank feature matrix.

[0016] Furthermore, the bidirectional correlation feature enhancement module specifically includes the following:

[0017] Radial basis function mask association unit: Based on the two-dimensional patient care low-rank feature matrix, the Euclidean distance between features of patients is calculated, the distance is converted into initial similarity using radial basis function, and the nursing association adjacency matrix is ​​used as a mask to obtain the intermediate association weight matrix;

[0018] Neighborhood joint objective construction unit: Based on the intermediate correlation weight matrix, the second-order neighborhood correlation matrix is ​​calculated by matrix multiplication. Under the conditions of non-negativity constraints and row sum constraints, a joint optimization objective function that integrates first-order distance information and second-order neighborhood information is constructed.

[0019] Lagrange closed-form solver; for the joint optimization objective function, Lagrange multipliers are introduced to handle row and sum constraints, a Lagrange function is constructed, the partial derivative of the intermediate association weight value is calculated and set to zero, the closed expression of the optimal association weight with respect to the Lagrange multipliers is derived, the optimal Lagrange multipliers are solved by the bisection method, the solution satisfies the row and sum constraints, the converged optimal association weights are integrated into the optimal patient-risk association matrix, a self-loop identity matrix is ​​added, the corresponding degree matrix is ​​calculated and row normalization is performed to obtain the standardized patient association matrix;

[0020] A bidirectional aggregation and fusion unit is used. Based on a standardized patient association matrix, neighborhood information of the two-dimensional patient care low-rank feature matrix is ​​aggregated in the patient dimension to obtain a patient aggregated feature matrix. The nursing association adjacency matrix corresponding to the two-dimensional patient care low-rank feature matrix is ​​constructed according to the cosine similarity of the features, and self-loop addition and row normalization are completed. Neighborhood information of the feature dimension is aggregated in the two-dimensional patient care low-rank feature matrix to obtain an indicator aggregated feature matrix. The patient aggregated feature matrix and the indicator aggregated feature matrix are weighted and fused to obtain a bidirectional aggregation enhanced feature matrix.

[0021] Furthermore, the disease risk level classification module inputs the bidirectional aggregation enhancement feature matrix into a two-layer fully connected neural network, obtains a low-dimensional latent feature matrix through nonlinear transformation, and then maps it to the original prediction score matrix of the three disease risk levels through the output layer. The Softmax function is used to convert the original prediction scores into normalized prediction probabilities corresponding to each level to obtain a probability matrix. The category index corresponding to the maximum probability value is taken for each row of the probability matrix to generate a prediction risk label vector.

[0022] Furthermore, the metabolic chronic disease nursing data monitoring module collects nursing data of patients with metabolic chronic diseases to be tested, and performs data encoding and normalization processing. The processed nursing data of patients with metabolic chronic diseases to be tested is then sequentially input into the nursing feature tensor quantification construction module, the block adaptive feature denoising module, the bidirectional correlation feature enhancement module, and the disease risk level classification module. The system outputs the corresponding disease risk level prediction results. For patients determined to be at medium or high risk, the system automatically generates risk warning messages and pushes them to the medical and nursing management terminal.

[0023] The beneficial effects achieved by the present invention using the above solution are as follows:

[0024] (1) In response to the problems in the existing metabolic chronic disease nursing data monitoring system, such as the easy mixing of redundant noise during the collection process and the significant differences in the contribution of different feature indicators, traditional noise reduction methods are difficult to distinguish the importance of information in each dimension, resulting in the weakening of key features and the interference of noise residue on the accuracy of subsequent nursing data monitoring, this solution obtains a block diagonal matrix according to three types of business slices, obtains the singular value sequence through singular value decomposition, quantifies the information content of each feature indicator, and provides a clear and quantifiable judgment basis for distinguishing effective information from noise; calculates the adaptive weight coefficient, constructs the weighted nuclear norm regularization term, and forms an enhanced optimization objective function together with the approximate error term, and uses the proximal gradient descent method to iteratively solve the problem to obtain the low-rank matrix corresponding to each business slice, and integrates it to obtain a two-dimensional patient nursing low-rank feature matrix, so that the key features of each business dimension can be preserved to the greatest extent and the noise can be suppressed in a targeted manner, providing accurate and reliable data support for subsequent risk level classification, intelligent early warning and other full-process monitoring work, and comprehensively improving the overall accuracy of metabolic chronic disease nursing data monitoring.

[0025] (2) In response to the problems in the existing metabolic chronic disease nursing data supervision system, such as the complex relationships among patient groups, which make it difficult to effectively reflect the true disease relationships, and the separation of the relationships between patient samples and the relationships between nursing indicators, resulting in insufficient expression of disease risk classification features and reduced credibility of nursing data supervision, this solution uses the nursing association adjacency matrix as a mask to obtain the intermediate association weight matrix, thereby eliminating invalid association interference in the data; calculates the second-order neighborhood association matrix, and constructs a joint optimization objective function under non-negativity and row sum constraints, so that the patient relationship reflects both direct disease similarity and indirect transmission association; introduces Lagrange multipliers to derive closed expressions and uses the bisection method to solve them, thereby obtaining a standardized patient association matrix, enabling the supervision system to have the ability to quickly model the relationships of large-scale patient groups; and obtains the patient aggregation feature matrix and the indicator aggregation feature matrix through neighborhood information aggregation, constructs a bidirectional aggregation enhanced feature matrix, realizes the complementary fusion of patient association and indicator association features, greatly enhances the deep feature expression ability of chronic disease nursing data, and effectively improves the credibility and effectiveness of the whole-process risk supervision of metabolic chronic disease nursing data. Attached Figure Description

[0026] Figure 1 A schematic diagram of the big data-based metabolic chronic disease nursing data monitoring system provided by the present invention;

[0027] Figure 2 This is a schematic diagram of the bidirectional correlation feature enhancement module.

[0028] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0029] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0030] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0031] Example 1, see Figure 1The present invention provides a big data-based nursing data monitoring system for metabolic chronic diseases, including a nursing data acquisition module, a nursing feature tensor quantification construction module, a block adaptive feature denoising module, a bidirectional correlation feature enhancement module, a disease risk level classification module, and a metabolic chronic disease nursing data monitoring module.

[0032] The nursing data acquisition module collects historical nursing data of patients with metabolic chronic diseases and constructs a nursing dataset for patients with metabolic chronic diseases.

[0033] The nursing feature tensor construction module constructs a two-dimensional patient nursing feature matrix and a nursing association adjacency matrix, which are then extended to obtain a three-dimensional patient nursing tensor.

[0034] The block-adaptive feature denoising module obtains a block-diagonal matrix according to three types of business slices, obtains a singular value sequence through singular value decomposition, calculates adaptive weight coefficients, constructs a weighted nuclear norm regularization term, and together with the approximation error term, forms an enhanced optimization objective function. Iteratively solves the problem using the proximal gradient descent method to obtain the low-rank matrix corresponding to each business slice, and integrates them to obtain a two-dimensional patient care low-rank feature matrix.

[0035] The bidirectional association feature enhancement module uses the nursing association adjacency matrix as a mask to obtain the intermediate association weight matrix, calculates the second-order neighborhood association matrix, constructs a joint optimization objective function, introduces Lagrange multipliers to derive a closed expression, and uses the bisection method to solve it to obtain the standardized patient association matrix. Through neighborhood information aggregation, it obtains the patient aggregation feature matrix and the indicator aggregation feature matrix, and constructs a bidirectional aggregation enhancement feature matrix.

[0036] The disease risk level classification module calculates a low-dimensional latent feature matrix and generates a predicted risk label vector.

[0037] The metabolic chronic disease nursing data monitoring module generates risk warning messages and pushes them to the medical management terminal.

[0038] Example 2, see Figure 1 This embodiment is based on the above embodiment. In the nursing data acquisition module, historical nursing data of patients with metabolic chronic diseases are collected, including vital sign data, laboratory indicator data, and medication and lifestyle behavior data. The collected historical nursing data of patients with metabolic chronic diseases is encoded, normalized, and labeled with disease risk level to construct a nursing dataset of patients with metabolic chronic diseases.

[0039] The vital signs data include systolic blood pressure, diastolic blood pressure, resting heart rate, height, weight, body mass index, waist circumference, and hip circumference;

[0040] The test data include fasting blood glucose, 2-hour postprandial blood glucose, glycated hemoglobin, total cholesterol, triglycerides, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, blood uric acid, and liver and kidney function indicators.

[0041] The medication and lifestyle data include the types of medications used for chronic diseases, daily frequency of medication, medication adherence, daily exercise duration, type of exercise, dietary structure score, average daily fat intake, average daily sugar intake, smoking status, and alcohol consumption status.

[0042] The data encoding uses One-Hot encoding to convert categorical data into numerical data.

[0043] The data normalization method uses the max-min scaling method to uniformly map numerical data to the [0, 1] interval, eliminating dimensional differences.

[0044] The disease risk level labeling is based on a comprehensive assessment of vital signs, laboratory indicators, medication use, and lifestyle behaviors. The labeling is used to indicate the disease risk level of the data as a data tag, including low risk (indicators are basically normal, medication use and lifestyle are good), medium risk (some indicators are slightly abnormal, medication use or lifestyle has defects), and high risk (multiple indicators are significantly abnormal, medication use is not standardized and lifestyle is poor).

[0045] Example 3, see Figure 1 This embodiment, based on the above embodiment, divides the nursing feature tensor construction module into three independent business dimensions for the nursing data of patients with metabolic chronic diseases. A two-dimensional matrix cannot distinguish the business attributes and distribution characteristics of each type of data, which is not conducive to subsequent targeted data supervision and risk analysis. Based on the nursing dataset of patients with metabolic chronic diseases, a two-dimensional patient nursing feature matrix is ​​constructed. Based on the cosine similarity of features, a nursing association adjacency matrix A is constructed. According to the classification of three categories of services—physical signs, laboratory indicators, and medication and lifestyle behaviors—the two-dimensional patient nursing feature matrix is ​​expanded into a three-dimensional patient nursing tensor. Feature quantification was performed according to business categories to clearly distinguish different types of nursing data, which not only meets the requirements of subsequent data processing but also facilitates multi-dimensional monitoring of chronic disease conditions and risk tracing; the formulas used are as follows:

[0046] ;

[0047] In the formula, It is the i-th patient sample x in the metabolic chronic disease patient care dataset D. i Compared with the j-th patient sample x j The nursing relationship status between them, where i and j are patient sample indices. It is x i With x jThe cosine similarity between the patient care characteristics of the two; θ is the similarity threshold. The cosine similarity of all patients in the data set D is statistically analyzed and a distribution histogram is plotted. The similarity value corresponding to the first frequency mutation inflection point on the histogram is set as the similarity threshold. is a real number field, n is the total number of patient samples in dataset D, m is the total dimension of data features; t is the number of business categories, t=3, corresponding to vital signs, test indicators, medication and lifestyle behaviors respectively; q is the number of features under each business category.

[0048] Example 4, see Figure 1 This embodiment, based on the above embodiment, specifically includes the following in the block-based adaptive feature denoising module:

[0049] Block Singular Value Decomposition Unit: Nursing data for patients with metabolic chronic diseases is prone to redundant noise generated during measurement and data entry, affecting the accuracy of subsequent disease analysis. The three-dimensional patient nursing tensor is sliced ​​into three business segments and converted into a block diagonal matrix. Singular value decomposition is then performed on the block diagonal matrix to obtain the singular value sequence and its corresponding left and right singular vectors for each business segment. The magnitude of the singular value directly reflects the importance of the information contained in the corresponding feature component: the larger the value, the more effective disease information it contains; the smaller the value, the more noise and redundant information it contains. This decomposition process provides the foundation for subsequent differentiated noise reduction. The formula used is as follows:

[0050] ;

[0051] ;

[0052] In the formula, It is a block diagonal matrix. It is a function for constructing a block diagonal matrix. , and These are two-dimensional tensor slice matrices corresponding to the three types of business. It is the slice matrix corresponding to the z-th class; z is the business category index. and These are the left singular vector matrix and the right singular vector matrix, respectively. It is a singular value diagonal matrix, and T is the matrix transpose; , and These are the 1st, 2nd, and qth business items in the z-th category. z q singular values, sorted in descending order; z It is the number of features under the z-th type of business, that is, the upper limit of the number of singular values ​​in this slice;

[0053] An adaptive weighted noise reduction unit is used. The impact of various nursing indicators for chronic diseases on disease assessment varies, requiring differentiated processing strategies to meet the needs of clinical nursing data monitoring and assessment. Based on the singular value sequences corresponding to each business segment, an adaptive weight coefficient is calculated for each singular value, ensuring that components with larger singular values ​​and more effective information receive higher retention weights. Weighted nuclear norm regularization terms are constructed for each business segment, forming an enhanced optimization objective function together with the approximation error term. The proximal gradient descent method is used for iterative solution to obtain the low-rank matrix corresponding to each business segment. In this optimization process, important features corresponding to large singular values ​​are given high retention weights, while noise components corresponding to small singular values ​​are further suppressed, thus achieving differentiated noise reduction across different business dimensions. The low-rank matrices obtained from each business segment are concatenated and integrated according to the original business order to obtain a two-dimensional patient nursing low-rank feature matrix. By using differentiated weights to strengthen the role of key indicators and weaken noise interference, data is integrated after noise reduction by business segment, and big data processing effectively improves the overall quality of chronic disease care data and the reliability of risk analysis; the formulas used are as follows:

[0054] ;

[0055] ;

[0056] In the formula, It is the adaptive weighting coefficient of the b-th singular value in the z-th business category. and These are the b-th and k-th singular values ​​in the z-th business category, respectively, where b and k are singular value indices. It is the low-rank matrix to be solved corresponding to the z-th type of business slice; It is a smoothing term. ; It is an approximation error term, used to measure the deviation between the denoised matrix and the original matrix; It is a weighted nuclear norm regularization term used to apply differential penalties to the singular values ​​of the target matrix during the optimization process.

[0057] By performing the above operations, this solution addresses the problems in existing metabolic chronic disease nursing data monitoring systems, such as the easy incorporation of redundant noise during data collection and significant differences in the contribution of different feature indicators. Traditional noise reduction methods struggle to distinguish the importance of information across dimensions, leading to the weakening of key features and residual noise interfering with the accuracy of subsequent nursing data monitoring. This solution obtains a block-based diagonal matrix by dividing the data into three business segments. Singular value decomposition is used to obtain singular value sequences, quantifying the information content of each feature indicator and providing a clear and quantifiable basis for distinguishing effective information from noise. Adaptive weighting coefficients are calculated, and a weighted nuclear norm regularization term is constructed. This term, together with the approximation error term, forms an enhanced optimization objective function. The proximal gradient descent method is used for iterative solution to obtain the low-rank matrix corresponding to each business segment. This is then integrated to obtain a two-dimensional patient nursing low-rank feature matrix, maximizing the preservation of key features across each business dimension and effectively suppressing noise. This provides accurate and reliable data support for subsequent risk level classification, intelligent early warning, and other full-process monitoring work, comprehensively improving the overall accuracy of metabolic chronic disease nursing data monitoring.

[0058] Example 5, see Figure 1 and Figure 2 This embodiment, based on the above embodiment, specifically includes the following in the bidirectional correlation feature enhancement module:

[0059] Radial basis function mask association unit; The association data of patients with metabolic chronic diseases is complex and contains a large amount of invalid association information, requiring the screening of valid patient associations to meet the risk monitoring needs of the chronic disease group; Based on the two-dimensional patient care low-rank feature matrix, the Euclidean distance between patient features is calculated, and the radial basis function is used to convert the distance into an initial similarity. The nursing association adjacency matrix A is used as a mask to hide unrelated patient pairs in the original adjacency, resulting in an intermediate association weight matrix S; invalid data interference is accurately removed through big data processing, strengthening the data foundation for risk association monitoring of the chronic disease patient group; The formulas used are as follows:

[0060] ;

[0061] In the formula, It is the intermediate association weight value between the i-th patient and the j-th patient. σ is the Euclidean distance between the features of the i-th patient and the j-th patient; σ is the bandwidth parameter, which controls the rate at which similarity decays with distance, and its value is set to the median of the Euclidean distances between all patients.

[0062] Neighborhood joint target construction unit; the disease association of patients with metabolic chronic diseases has multi-dimensional characteristics, and a single association information cannot cover the complete disease association pattern; based on the intermediate association weight matrix, matrix multiplication is used to calculate the second-order neighborhood association matrix. Under the constraints of nonnegativity and row sum, a joint optimization objective function integrating first-order distance information and second-order neighborhood information is constructed. ;Comprehensively depict the disease distribution structure of the patient population with metabolic chronic diseases; In the formula, S i It is the intermediate association weight vector corresponding to the i-th patient. It is the second-order neighborhood association value between the i-th patient and the j-th patient;

[0063] Lagrange closed-form solver; In large-scale chronic disease care data, patient relationships are complex, and traditional iterative solutions are inefficient and struggle to guarantee row and normalization constraints; For the joint optimization objective function, Lagrange multipliers are introduced to handle row and normalization constraints, and a Lagrange function is constructed to assign weights to intermediate relationships. By taking the partial derivatives and setting them to zero, a closed-form expression for the optimal association weights with respect to the Lagrange multipliers is derived. The optimal Lagrange multipliers are then solved using a bisection method, ensuring the solution satisfies row sum constraints. The converged optimal association weights are integrated into the optimal patient-risk association matrix. A self-loop identity matrix is ​​added, and the corresponding degree matrix is ​​calculated and row normalization is performed to obtain the standardized patient association matrix. The closed-form solution is obtained quickly, improving computational efficiency and numerical stability, effectively ensuring the accuracy and standardization of risk monitoring for chronic disease groups. The formulas used are as follows:

[0064] ;

[0065] ;

[0066] ;

[0067] In the formula, β is the optimal association weight between the i-th patient and the j-th patient; β is the second-order neighborhood matching coefficient, which controls the degree of matching between the optimized association weight and the second-order neighborhood information. ;λ i It is the Lagrange multiplier corresponding to the weight vector in the i-th row. It is the optimal correlation weight function with Lagrange multipliers as independent variables; These are the weight row and loss function, used to measure the deviation between the sum of the current temporary optimal weights and the constraint value 1; It is the optimal patient-risk association matrix. I is an n-order identity matrix. It is a matrix The inverse of the degree matrix, It is a standardized patient association matrix;

[0068] A bidirectional aggregation and fusion unit is used; the patient associations and indicator associations in metabolic chronic disease nursing data are independent and fragmented, which cannot support comprehensive disease risk monitoring. Based on a standardized patient association matrix, neighborhood information of the two-dimensional patient nursing low-rank feature matrix is ​​aggregated at the patient dimension to mine disease association features between patient samples, resulting in a patient aggregated feature matrix. A nursing association adjacency matrix corresponding to the two-dimensional patient nursing low-rank feature matrix is ​​constructed based on the cosine similarity of the features. The process involves self-loop addition and row normalization, followed by neighborhood information aggregation of the two-dimensional patient care low-rank feature matrix. This process uncovers the correlation features between nursing indicators, yielding an indicator aggregation feature matrix. The patient aggregation feature matrix and the indicator aggregation feature matrix are then weighted and fused to obtain a bidirectional aggregation enhanced feature matrix. Big data analysis is used to enhance feature representation capabilities, thereby improving the accuracy and interpretability of chronic disease risk level classification. The formulas used are as follows:

[0069] ;

[0070] ;

[0071] ;

[0072] In the formula, H p It is the patient aggregate feature matrix, H f It is an indicator aggregation feature matrix. yes The inverse of the degree matrix, W p and W f It is a learnable weight matrix, X e It is a bidirectional aggregation enhancement feature matrix, where δ is the feature fusion balance coefficient, used to adjust the proportion of patient aggregation features and indicator aggregation features in the fusion. .

[0073] By performing the above operations, this solution addresses the problems in existing metabolic chronic disease nursing data monitoring systems, such as complex patient group relationships that fail to effectively reflect true disease correlations, and the separation of patient sample correlations from nursing indicator correlations, leading to insufficient disease risk classification feature expression and reduced credibility of nursing data monitoring. This solution uses the nursing correlation adjacency matrix as a mask to obtain an intermediate correlation weight matrix, eliminating invalid correlation interference in the data. A second-order neighborhood correlation matrix is ​​calculated, and a joint optimization objective function is constructed under non-negativity and row sum constraints, ensuring that patient correlations simultaneously reflect direct disease similarity and indirect transmission correlations. A closed-form expression is derived using Lagrange multipliers and solved using a bisection method to obtain a standardized patient correlation matrix, enabling the monitoring system to quickly model correlations for large-scale patient groups. By aggregating neighborhood information, patient aggregated feature matrices and indicator aggregated feature matrices are obtained, constructing a bidirectional aggregated enhanced feature matrix to achieve complementary fusion of patient correlation and indicator correlation features. This significantly enhances the deep feature expression capability of chronic disease nursing data, effectively improving the credibility and effectiveness of full-process risk monitoring of metabolic chronic disease nursing data.

[0074] Example 6, see Figure 1 This embodiment is based on the above embodiment. In the disease risk level classification module, the bidirectional aggregation enhancement feature matrix is ​​input into a two-layer fully connected neural network. After nonlinear transformation, a low-dimensional latent feature matrix is ​​obtained. Then, it is mapped to the original prediction score matrix of three disease risk levels through the output layer. The Softmax function is used to convert the original prediction scores into normalized prediction probabilities corresponding to each level, resulting in a probability matrix. The category index corresponding to the maximum probability value of each row of the probability matrix is ​​taken to generate a prediction risk label vector. Cross-entropy loss is used as the objective function for optimization. All learnable parameters from the nursing feature tensor quantization construction module to the disease risk level classification module are updated through the backpropagation algorithm until convergence, thus completing the training.

[0075] Example 7, see Figure 1 This embodiment, based on the above embodiment, collects nursing data of patients with metabolic chronic diseases under test in the metabolic chronic disease nursing data monitoring module. This data includes vital signs, laboratory test results, and medication and lifestyle behavior data. The data is then encoded and normalized. The processed nursing data of patients with metabolic chronic diseases under test is sequentially input into the nursing feature tensor construction module, the block adaptive feature denoising module, the bidirectional correlation feature enhancement module, and the disease risk level classification module. The system outputs the corresponding disease risk level prediction results. For patients determined to be at medium or high risk, the system automatically generates risk warning messages and pushes them to the medical and nursing management terminal for timely review and intervention by clinical nursing staff. This achieves fully automated monitoring and risk warning of metabolic chronic disease nursing data throughout the entire process.

[0076] It should be noted that, in this document, the terms “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0077] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

[0078] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A big data-based monitoring system for the care of metabolic chronic diseases, characterized by: It includes a nursing data acquisition module, a nursing feature tensor quantification and construction module, a block adaptive feature denoising module, a bidirectional correlation feature enhancement module, a disease risk level classification module, and a metabolic chronic disease nursing data monitoring module; The nursing data acquisition module collects historical nursing data of patients with metabolic chronic diseases and constructs a nursing dataset for patients with metabolic chronic diseases. The nursing feature tensor construction module constructs a two-dimensional patient nursing feature matrix and a nursing association adjacency matrix, which are then extended to obtain a three-dimensional patient nursing tensor. The block-adaptive feature denoising module obtains a block-diagonal matrix according to three types of business slices, obtains a singular value sequence through singular value decomposition, calculates adaptive weight coefficients, constructs a weighted nuclear norm regularization term, and together with the approximation error term, forms an enhanced optimization objective function. Iteratively solves the problem using the proximal gradient descent method to obtain the low-rank matrix corresponding to each business slice, and integrates them to obtain a two-dimensional patient care low-rank feature matrix. The bidirectional association feature enhancement module uses the nursing association adjacency matrix as a mask to obtain the intermediate association weight matrix, calculates the second-order neighborhood association matrix, constructs a joint optimization objective function, introduces Lagrange multipliers to derive a closed expression, and uses the bisection method to solve it to obtain the standardized patient association matrix. Through neighborhood information aggregation, it obtains the patient aggregation feature matrix and the indicator aggregation feature matrix, and constructs a bidirectional aggregation enhancement feature matrix. The disease risk level classification module calculates a low-dimensional latent feature matrix and generates a predicted risk label vector. The metabolic chronic disease nursing data monitoring module generates risk warning messages and pushes them to the medical management terminal.

2. The big data-based metabolic chronic disease nursing data monitoring system according to claim 1, characterized in that: The nursing feature tensor construction module is based on the nursing dataset of patients with metabolic chronic diseases. It constructs a two-dimensional patient nursing feature matrix and a nursing association adjacency matrix based on the cosine similarity of the features. According to the division of three types of business, the two-dimensional patient nursing feature matrix is ​​extended into a three-dimensional patient nursing tensor.

3. The big data-based metabolic chronic disease nursing data monitoring system according to claim 1, characterized in that: The block-based adaptive feature denoising module specifically includes the following: Block Singular Value Decomposition Unit: The three-dimensional patient care tensor is sliced ​​into three types of business segments and converted into a block diagonal matrix. Singular value decomposition is performed on the block diagonal matrix to obtain the singular value sequence and its corresponding left and right singular vectors for each business segment. Adaptive weighted noise reduction unit.

4. The big data-based metabolic chronic disease nursing data monitoring system according to claim 3, characterized in that: The adaptive weighted noise reduction unit calculates the adaptive weight coefficient for each singular value based on the singular value sequence corresponding to each business slice. A weighted nuclear norm regularization term is constructed on each business slice, which together with the approximation error term forms an enhanced optimization objective function. The proximal gradient descent method is used to iteratively solve the function to obtain the low-rank matrix corresponding to each business slice. The low-rank matrices obtained from solving each business slice are then concatenated and integrated according to the original business order to obtain a two-dimensional patient care low-rank feature matrix.

5. The big data-based metabolic chronic disease nursing data monitoring system according to claim 1, characterized in that: The bidirectional correlation feature enhancement module specifically includes the following: Radial basis function mask association unit: Based on the two-dimensional patient care low-rank feature matrix, the Euclidean distance between features of patients is calculated, the distance is converted into initial similarity using radial basis function, and the nursing association adjacency matrix is ​​used as a mask to obtain the intermediate association weight matrix; Neighborhood joint target construction unit; Based on the intermediate association weight matrix, the second-order neighborhood association matrix is ​​calculated using matrix multiplication. Under the conditions of non-negativity constraints and row sum constraints, a joint optimization objective function that integrates first-order distance information and second-order neighborhood information is constructed. Lagrange closed-form solution element; Two-way aggregation and fusion unit.

6. The big data-based metabolic chronic disease nursing data monitoring system according to claim 5, characterized in that: The Lagrange closed-form solution unit is designed for the joint optimization objective function. It introduces Lagrange multipliers to handle row and sum constraints, constructs a Lagrange function, calculates the partial derivative of the intermediate association weights and sets it to zero, derives the closed expression of the optimal association weights with respect to the Lagrange multipliers, uses the bisection method to solve for the optimal Lagrange multipliers, ensures that the solution satisfies the row and sum constraints, integrates the converged optimal association weights into the optimal patient-risk association matrix, adds a self-loop identity matrix, calculates the corresponding degree matrix and performs row normalization to obtain the standardized patient association matrix.

7. The big data-based metabolic chronic disease nursing data monitoring system according to claim 6, characterized in that: The bidirectional aggregation and fusion unit is based on a standardized patient association matrix. It aggregates neighborhood information of the two-dimensional patient care low-rank feature matrix in the patient dimension to obtain a patient aggregated feature matrix. It constructs a nursing association adjacency matrix corresponding to the two-dimensional patient care low-rank feature matrix based on the cosine similarity of the features, and completes self-loop addition and row normalization processing. It then aggregates neighborhood information of the two-dimensional patient care low-rank feature matrix in the feature dimension to obtain an index aggregated feature matrix. Finally, it performs weighted fusion of the patient aggregated feature matrix and the index aggregated feature matrix to obtain a bidirectional aggregation enhanced feature matrix.

8. The big data-based metabolic chronic disease nursing data monitoring system according to claim 1, characterized in that: The disease risk level classification module inputs the bidirectional aggregated enhanced feature matrix into a two-layer fully connected neural network, obtains a low-dimensional latent feature matrix through nonlinear transformation, and then maps it to the original prediction score matrix of three disease risk levels through the output layer. The Softmax function is used to convert the original prediction scores into normalized prediction probabilities corresponding to each level, resulting in a probability matrix. The category index corresponding to the maximum probability value is taken for each row of the probability matrix to generate a prediction risk label vector.

9. The big data-based metabolic chronic disease nursing data monitoring system according to claim 1, characterized in that: The metabolic chronic disease nursing data monitoring module collects nursing data of patients with metabolic chronic diseases to be tested, performs data encoding and normalization, and then sequentially inputs the processed nursing data of patients with metabolic chronic diseases to the nursing feature tensor quantification construction module, the block adaptive feature denoising module, the bidirectional correlation feature enhancement module, and the disease risk level classification module. The module outputs the corresponding disease risk level prediction results. For patients who are determined to be of medium or high risk, the system automatically generates risk warning messages and pushes them to the medical and nursing management terminal to complete the monitoring.

10. The big data-based metabolic chronic disease nursing data monitoring system according to claim 1, characterized in that: The nursing data acquisition module collects historical nursing data of patients with metabolic chronic diseases, including vital signs data, laboratory test data, and medication and lifestyle data; and encodes, normalizes, and labels the disease risk level of the collected historical nursing data of patients with metabolic chronic diseases to construct a nursing dataset for patients with metabolic chronic diseases.

Images