Medical big data analysis and prediction system and method
By constructing graph structure and feature tree, using dynamic feature mapping functions and association weight calculation methods, combined with adaptive recursive optimization algorithm, the problems of data fusion, robustness and dynamic adjustment in medical big data analysis are solved, and the analysis model is improved Predictive performance and interpretability.
Patent Information
- Application Number
- CN202411517066.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-29
- Publication Date
- 2025-02-14
AI Technical Summary
Existing medical big data analysis technology is difficult to effectively integrate multiple types of medical data, and its robustness is insufficient, resulting in overfitting or underfitting of the model, and lacks dynamic adjustment and real-time update mechanisms, making it difficult to adapt to the rapidly changing medical data environment. The simultaneously generated analytical and predictive results lack interpretability.
By building graph structure and feature tree, using dynamic feature mapping functions and correlation weight calculation methods, dynamically adjust the model parameters of medical analysis models, improve feature expression ability and model robustness, and optimize model parameters through adaptive recursive optimization algorithm to adapt Data changes.
Improves the feature expression ability and model prediction performance of medical data analysis, enhances the robustness of noise and irrelevant features, provides more intuitive and actionable risk assessment information, and helps medical professionals develop personalized prevention strategies and treatment options.
Smart Images
Figure CN119446489A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of medical big data analysis, and in particular to a system and method for medical big data analysis and prediction. Background Art
[0002] In recent years, the rapid accumulation of medical data and the development of informatization have led to the continuous improvement of medical analysis models. In the field of medical big data analysis, existing technologies mainly rely on traditional machine learning and statistical analysis methods, such as linear regression, decision trees, and support vector machines. Although these methods perform well in processing structured data, they are still limited in capturing complex data patterns and identifying nonlinear relationships when facing the growing multi-source heterogeneous medical data (such as electronic medical records, genomic data, and medical images). In recent years, the development of neural networks and deep learning technologies has improved the ability to extract features and recognize patterns from complex data, but neural networks and deep learning technologies often require a large amount of labeled data for training and are susceptible to interference when processing unbalanced and noisy data.
[0003] In summary, the main problems faced by existing technologies in medical data analysis include: it is difficult to effectively integrate and utilize multiple types of medical data for comprehensive analysis; the lack of robustness to noise and irrelevant features during the modeling process easily leads to overfitting or underfitting of the model; there is a lack of effective mechanisms for dynamic adjustment and real-time updating of prediction models, making it difficult to adapt to the rapidly changing medical data environment. In addition, many existing medical data analysis and prediction methods lack sufficient interpretability in the generated analysis and prediction results, which limits the application of medical professionals in clinical decision-making. In order to overcome the above problems, there is an urgent need for an algorithm that can dynamically adjust and optimize feature influences and model parameters to enhance the expressiveness of the data and the predictive performance of the model. Summary of the invention
[0004] The present invention provides a system and method for medical big data analysis and prediction to solve the technical problems that it is difficult to effectively integrate and utilize various types of medical data for comprehensive analysis; the lack of robustness to noise and irrelevant features in the modeling process, which easily leads to overfitting or underfitting of the model; the lack of effective mechanism in the dynamic adjustment and real-time updating of the prediction model, which makes it difficult to adapt to the rapidly changing medical data environment; and the lack of sufficient interpretability in the generated analysis and prediction results, which limits the application of medical professionals in clinical decision-making.
[0005] The present invention provides a system and method for analyzing and predicting medical big data, which specifically includes the following technical solutions:
[0006] A method for analyzing and predicting medical big data, comprising the following steps:
[0007] S1: Collect and clean the original medical data to obtain medical data; build a medical analysis model based on the graph structure and feature tree, analyze the medical data, and generate risk prediction values;
[0008] S2: Based on the risk prediction value, use the adaptive recursive optimization algorithm to dynamically adjust the model parameters of the medical analysis model, verify and evaluate the medical analysis model, deploy the verified and evaluated medical analysis model into the actual environment, and continuously monitor the performance of the medical analysis model.
[0009] Preferably, the S1 specifically includes:
[0010] Medical data is used to construct a graph structure and define the node feature vector of each node. The node feature vector of the node is linearly transformed and nonlinearly activated using a dynamic feature mapping function to obtain the mapping feature vector of the node. The mapping feature vector formula is as follows:
[0011]
[0012] Among them, z i is the mapped feature vector of node i; exp is the exponential operation; W1 is the first-layer weight matrix, which is used to linearly adjust the dimension and direction after feature mapping; W2 is the second-layer weight matrix, which is used to adjust the node feature vector x of node i i Perform linear transformation; x i is the node feature vector of node i; b1 is the first-layer bias vector, which is used for bias adjustment after the first-layer linear transformation; b2 is the second-layer bias vector, which is used for bias adjustment after the second-layer linear transformation; tanh is the hyperbolic tangent function; ||·|| represents the Euclidean norm of the vector.
[0013] Preferably, the S1 specifically includes:
[0014] The association weights between the mapping feature vectors of the calculation nodes are calculated as follows:
[0015]
[0016] Among them, α ik is the association weight between the mapping feature vector of node i and the mapping feature vector of node k; T is the transposition operation; A is the adaptive weight matrix; z k is the mapping feature vector of node k; k is the index variable used to traverse the neighbor node set of node i; is the set of neighbor nodes of node i;
[0017] The mapping feature vector of the node is weighted by using the associated weight to obtain the enhanced feature vector of the node;
[0018]
[0019] Among them, z′ i is the enhanced feature vector of node i.
[0020] Preferably, the S1 specifically includes:
[0021] Construct a feature tree, the nodes of which include leaf nodes, intermediate nodes, and root nodes; use the enhanced feature vectors of the nodes in the graph structure as the enhanced feature vectors of the leaf nodes of the feature tree for hierarchical modeling, and calculate the feature influence of each node in the feature tree;
[0022] The feature influence of the leaf node is obtained using the feature contribution function. The feature contribution function formula is as follows:
[0023]
[0024] Among them, β i represents the feature influence of leaf node i; f(z′ i ) is the feature contribution function of leaf node i, which is used as the feature influence of leaf node i when building the medical analysis model; u is the weight vector; b is the bias term;
[0025] The formula for the characteristic influence of the root node and the intermediate node is as follows:
[0026]
[0027] Among them, w pq is the connection weight between node p and node q; node q represents the child node of node p; σ is the parameter controlling the width of the Gaussian kernel function; z′ p is the enhanced feature vector of node p; z′ q is the enhanced feature vector of node q; β p is the characteristic influence of node p; γ is the adjustment coefficient; child(p) is the set of child nodes of node p; β q is the characteristic influence of node q.
[0028] Preferably, the S1 specifically includes:
[0029] The risk prediction value is calculated based on the feature influence of each node in the feature tree. The specific formula is:
[0030]
[0031] Among them, R i is the risk prediction value; nodes is the set of all nodes in the feature tree; ζ is the node index variable in the feature tree; β ζ is the characteristic influence of node ζ.
[0032] Preferably, the S2 specifically includes:
[0033] In the implementation process of the adaptive recursive optimization algorithm, an adaptive loss function is defined to measure the difference between the predicted results of the medical analysis model and the actual labels.
[0034] Preferably, the S2 specifically includes:
[0035] The adaptive loss function is used to dynamically adjust the model parameters of the medical analysis model, including feature influence, connection weights, weight vectors, and bias terms.
[0036] A system for analyzing and predicting medical big data, comprising the following parts:
[0037] Data preprocessing module, feature extraction module, feature tree modeling module, prediction and optimization module, verification and deployment module;
[0038] A data preprocessing module collects raw medical data and performs data cleaning on the raw medical data to obtain medical data. The data preprocessing module is connected to the feature extraction module through data transmission;
[0039] Use medical data to build a medical analysis model. The medical analysis model generates risk prediction values by analyzing and calculating medical data. The specific process is implemented through feature extraction modules and feature tree modeling modules.
[0040] The feature extraction module constructs a graph structure based on medical data and generates mapping feature vectors of nodes of the graph structure using a dynamic feature mapping function; according to the association weights between the mapping feature vectors of nodes of the graph structure and the mapping feature vectors of neighboring nodes, the enhanced feature vectors of nodes of the graph structure are calculated and generated. The feature extraction module is connected to the feature tree modeling module by means of data transmission;
[0041] The feature tree modeling module constructs a feature tree, uses the enhanced feature vectors of the nodes of the graph structure as the enhanced feature vectors of the leaf nodes of the feature tree for hierarchical modeling, and calculates the feature influence of each node in the feature tree; calculates the risk prediction value based on the feature influence of each node in the feature tree; the feature tree modeling module is connected to the prediction and optimization module through data transmission;
[0042] The prediction and optimization module uses an adaptive recursive optimization algorithm to dynamically adjust feature influence, connection weights, weight vectors, and bias terms based on an adaptive loss function. The prediction and optimization module is connected to the verification and deployment module through data transmission.
[0043] The verification and deployment module is used to verify and evaluate the medical analysis model, deploy the verified and evaluated medical analysis model in the actual environment, monitor the performance of the medical analysis model, and continuously optimize the medical analysis model using an adaptive recursive optimization algorithm based on actual usage feedback.
[0044] The beneficial effects of the technical solution of the present invention are:
[0045] 1. By introducing dynamic feature mapping functions and associated weight calculation methods, the expressive power of features and the ability to capture complex data patterns are improved; by constructing a graph structure and performing linear transformation and nonlinear activation on node feature vectors, complex patterns in the data can be accurately identified and enhanced; the mapped feature vectors are weighted using associated weights to obtain enhanced feature vectors, so that the influence of important features is amplified, the interference of noise and irrelevant features on the medical analysis model is suppressed, and the accuracy and robustness of the medical analysis model are improved.
[0046] 2. Use hierarchical modeling to build a feature tree, and calculate the feature influence of each node in the feature tree to accurately reflect the importance of the feature in risk prediction; calculate the risk prediction value based on the feature influence of each node in the feature tree to quantify the risk level of an individual's illness; convert complex medical data analysis results into intuitive and actionable risk assessment information to help medical professionals develop personalized prevention strategies and treatment plans, and improve the accuracy and timeliness of diagnosis.
[0047] 3. Introduce an adaptive recursive optimization algorithm to dynamically adjust the model parameters of the medical analysis model so that the medical analysis model can adapt to new data distribution and feature changes. The adaptive recursive optimization algorithm optimizes the model parameters of the medical analysis model through recursive optimization strategies and iterative updates of adaptive loss functions, thereby improving the fitting ability and prediction performance of the medical analysis model. In practical applications, the adaptive recursive optimization algorithm is used to optimize the medical analysis model based on actual usage feedback to adapt to the ever-changing medical data and needs, thereby improving the overall effectiveness and practicality of the medical analysis model. BRIEF DESCRIPTION OF THE DRAWINGS
[0048] Figure 1 This is a system structure diagram of a medical big data analysis and prediction system according to the present invention;
[0049] Figure 2 This is a flow chart of a method for medical big data analysis and prediction described in the present invention. DETAILED DESCRIPTION
[0050] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined purpose of the invention, the technical scheme in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.
[0051] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
[0052] The specific scheme of a system and method for medical big data analysis and prediction provided by the present invention is described in detail below with reference to the accompanying drawings.
[0053] See attached Figure 1 , which shows a system structure diagram of a medical big data analysis and prediction system provided by an embodiment of the present invention, the system includes the following parts:
[0054] Data preprocessing module, feature extraction module, feature tree modeling module, prediction and optimization module, verification and deployment module;
[0055] The data preprocessing module collects raw medical data from the hospital's information management system, laboratory database, and medical imaging system, cleans the raw medical data to obtain medical data, and the data preprocessing module is connected to the feature extraction module by means of data transmission;
[0056] Use medical data to build a medical analysis model. The medical analysis model generates risk prediction values by analyzing and calculating medical data. The specific process is implemented through feature extraction modules and feature tree modeling modules.
[0057] The feature extraction module constructs a graph structure based on medical data, generates mapping feature vectors of nodes of the graph structure using a dynamic feature mapping function, and then calculates and generates enhanced feature vectors of nodes of the graph structure based on the association weights between the mapping feature vectors of nodes of the graph structure and mapping feature vectors of neighboring nodes. The feature extraction module is connected to the feature tree modeling module by means of data transmission;
[0058] The feature tree modeling module constructs a feature tree, the nodes of the feature tree include root nodes, leaf nodes and intermediate nodes; uses the enhanced feature vectors of the nodes of the graph structure as the enhanced feature vectors of the leaf nodes of the feature tree for hierarchical modeling, and calculates the feature influence of each node in the feature tree; calculates the risk prediction value based on the feature influence of each node in the feature tree; the feature tree modeling module is connected to the prediction and optimization module by means of data transmission;
[0059] The prediction and optimization module dynamically adjusts the feature influence, connection weight, weight vector and bias term based on the adaptive loss function through an adaptive recursive optimization algorithm to improve the fitting ability of the medical analysis model. The prediction and optimization module is connected to the verification and deployment module through data transmission;
[0060] The verification and deployment module is used to verify and evaluate the medical analysis model, deploy the verified and evaluated medical analysis model in an actual environment, continuously monitor the performance of the medical analysis model, and continuously optimize the medical analysis model using an adaptive recursive optimization algorithm based on actual usage feedback.
[0061] See attached Figure 2 , which shows a flow chart of a method for medical big data analysis and prediction provided by an embodiment of the present invention, the method comprising the following steps:
[0062] S1: Collect and clean the original medical data to obtain medical data; build a medical analysis model based on the graph structure and feature tree, analyze the medical data, and generate risk prediction values;
[0063] Collect original medical data from the hospital's information management system, laboratory database, and medical imaging system, including electronic medical records (EMR), laboratory test results, medical imaging data (such as X-rays, MRI), genomic data, etc.
[0064] The original medical data is cleaned, including missing value processing, duplicate data processing, formatting and standardization, outlier detection and processing, etc., to obtain medical data. Data cleaning is an existing technology and will not be described in detail here.
[0065] Medical analysis models are built using medical data. Medical analysis models generate risk prediction values by analyzing and calculating medical data. Risk prediction values reflect the probability or risk level of an individual suffering from a certain disease. By converting complex medical data into intuitive risk prediction values, it can help medical professionals develop more accurate prevention strategies and treatment plans, and improve the scientificity and accuracy of medical decision-making.
[0066] The process of building a medical analysis model is as follows:
[0067] First, we use medical data to construct a graph structure. The nodes of the graph structure represent data entities, such as specific patients or specific medical events (such as diagnosis results and prescriptions), and the edges represent the relationship between nodes, such as the association between patients and medical events. The node feature vector represents the relevant attributes of the node, including age, gender, BMI, and medical history. The weight of the edge is calculated by comparing the angle cosine value of the node feature vector.
[0068] The node feature vector expression is as follows:
[0069] x i =[x i1 , x i2 , ..., x ia , ..., x im ]
[0070] Among them, x i is the node feature vector of node i; x ia is the node feature vector x i The ath component of represents the specific eigenvalue, a∈[1,m], where m is the number of features in the node feature vector of node i.
[0071] Then, the dynamic feature mapping function is used to perform linear transformation and nonlinear activation on the node feature vector to obtain the mapping feature vector. By extracting node features and strengthening the expressive power of node features, the complex patterns in medical data can be effectively captured. Linear transformation helps to align and combine different features, while nonlinear activation enhances the medical analysis model's ability to recognize complex relationships and nonlinear patterns, and improves the discrimination of node features.
[0072] The mapping feature vector formula is as follows:
[0073]
[0074] Among them, z i is the mapping feature vector of node i; exp is the exponential operation; W1 is the first-layer weight matrix, which is used to linearly adjust the dimension and direction after feature mapping; W2 is the second-layer weight matrix, which is used to adjust x i Perform linear transformation; b1 is the first layer bias vector, which is used to adjust the bias after the first layer linear transformation; b2 is the second layer bias vector, which is used to adjust the bias after the second layer linear transformation; tanh is the hyperbolic tangent function; ||·|| represents the Euclidean norm of the vector, which is used to measure the size or length of the vector.
[0075] Furthermore, the association weights between the mapped feature vectors are calculated, and the mapped feature vectors are weighted using the association weights to obtain enhanced feature vectors. By calculating the association weights between the mapped feature vectors, the medical analysis model can measure the relationship and importance between different feature vectors, and use the association weights to weight the mapped feature vectors, which not only strengthens the influence of important features, but also suppresses the interference of noise and irrelevant features on the medical analysis model.
[0076] The formula for enhancing feature vectors and associated weights is as follows:
[0077]
[0078] Among them, z′ iis the enhanced feature vector of node i; α ik is the association weight between the mapping feature vector of node i and the mapping feature vector of node k; z k is the mapping feature vector of node k; is the set of neighbor nodes of node i, indicating other nodes directly associated with node i; Used to calculate the similarity between node i and node k, and enhance its influence through square and exponential operations; is a transposition operation; k is an index variable used to traverse the set of neighbor nodes of node i; exp is an exponential operation; A is an adaptive weight matrix used to adjust the association between mapping feature vectors, which is obtained by calculating the covariance matrix of the mapping feature vectors using the principal component analysis (PCA) technology. PCA is an existing technology and will not be described in detail here.
[0079] Furthermore, a feature tree is constructed, and the enhanced feature vectors of the nodes in the graph structure are used as the enhanced feature vectors of the leaf nodes of the feature tree for hierarchical modeling. The nodes of the feature tree include root nodes, leaf nodes, and intermediate nodes; the leaf nodes represent specific features, and the root nodes and intermediate nodes represent the aggregation results of the features; the enhanced feature vectors of the intermediate nodes are based on the enhanced feature vectors of the child nodes of the intermediate nodes, and are calculated by existing aggregation methods (such as maximum pooling or weighted summation). The feature influence of each node in the feature tree is calculated.
[0080] The feature influence of the leaf node is obtained using the feature contribution function. The feature contribution function formula is as follows:
[0081]
[0082] Among them, β i represents the feature influence of leaf node i; f(z′ i ) is the feature contribution function of leaf node i, which is used as the feature influence of leaf node i when building the medical analysis model and is used to calculate z′ i The contribution of feature contribution function in risk prediction; u is the weight vector, which is used for the linear combination of feature contribution function; b is the bias term, which is used to adjust the output range of feature contribution function. The initial values of weight vector and bias term are set by random initialization and initialization to zero respectively, and then calculated by adaptive recursive optimization algorithm in the training process of medical analysis model.
[0083] The formula for the characteristic influence of the root node and the intermediate node is as follows:
[0084]
[0085] Among them, w pqis the connection weight between node p and node q, indicating the degree of influence between a node in the feature tree and its child nodes; node q represents the child node of node p; σ is a parameter that controls the width of the Gaussian kernel function, determines the similarity decay speed, and is set according to the specific implementation scenario; is the Gaussian similarity of the distances between nodes; β p is the feature influence of node p; γ is the adjustment coefficient, which controls the sensitivity of feature influence to input changes; child(p) is the set of child nodes of node p; z′ p is the enhanced feature vector of node p; z′ q is the enhanced feature vector of node q; β q is the characteristic influence of node q.
[0086] Finally, the risk prediction value is calculated based on the feature influence of each node in the feature tree. The risk prediction value formula is as follows:
[0087]
[0088] Among them, R i is the risk prediction value, which is the i-th sample (i.e. z′ i )’s medical analysis model prediction results; nodes is the set of all nodes in the feature tree; ζ is the node index variable in the feature tree; β ζ is the characteristic influence of node ζ.
[0089] S2: Based on the risk prediction value, use the adaptive recursive optimization algorithm to dynamically adjust the model parameters of the medical analysis model, verify and evaluate the medical analysis model, deploy the verified and evaluated medical analysis model into the actual environment, and continuously monitor the performance of the medical analysis model.
[0090] First, the adaptive recursive optimization algorithm is used to dynamically adjust the model parameters of the medical analysis model to improve the prediction performance of the medical analysis model. Specifically, the adaptive recursive optimization algorithm can dynamically adjust the feature influence, connection weight, weight vector and bias term based on the adaptive loss function to improve the prediction performance of the medical analysis model and enable the medical analysis model to adapt to new data distribution and feature changes; the recursive optimization strategy is adopted to gradually minimize the adaptive loss function by continuously iteratively updating the model parameters of the medical analysis model, thereby improving the fitting ability of the medical analysis model.
[0091] The adaptive recursive optimization algorithm process is as follows:
[0092] Using adaptive loss function Measures the difference between the predicted results of the medical analysis model and the actual labels. The adaptive loss function formula is as follows:
[0093]
[0094] in, is an adaptive loss function used to measure the difference between the prediction results of the medical analysis model and the actual label; n is the number of data samples, which indicates the total number of data samples used in the training process of the medical analysis model; y i is the actual label, indicating the true disease status of the i-th sample; R i is the risk prediction value, which is the prediction result of the medical analysis model for the i-th sample, reflecting the output of the medical analysis model for predicting the disease risk of a specific patient under the current parameter settings.
[0095] The adaptive loss function is used to dynamically adjust the model parameters of the medical analysis model, including feature influence, connection weights, weight vectors, and bias terms.
[0096] The parameter iterative update formula of the weight vector and bias term is as follows:
[0097]
[0098] Among them, u is the iterated weight vector, which is used for the linear combination of feature contribution functions; is the weight vector before iteration; b is the bias term after iteration, which is used to adjust the output range of the feature contribution function; is the bias term before iteration; sigmoid is the activation function; η is the learning rate, which is used to control the size of the step size during parameter update; λ is the regularization coefficient of the weight vector; λ b is the regularization coefficient of the bias term; is the adaptive loss function right The partial derivative of is the adaptive loss function right The learning rate and regularization coefficient are obtained through experiments and parameter adjustment. The initial values of the weight vector and the bias term are calculated by random initialization (such as Gaussian distribution, uniform distribution) method. Random initialization is a prior art and will not be described in detail here.
[0099] The iterative update formula for the feature influence and connection weight of the root node and intermediate nodes is as follows:
[0100]
[0101] Among them, β p is the feature influence of node p after iteration; β′ p is the feature influence of node p before iteration; β′ q is the feature influence of node q before iteration; child(p) is the set of child nodes of node p; w pqis the connection weight between node p and node q after iteration; w′ pq is the connection weight between node p and node q before iteration; is the adaptive loss function β′ p The partial derivative of is the adaptive loss function For w′ pq The partial derivative of p is the enhanced feature vector of node p; z′ q is the enhanced feature vector of node q; ∈ is a constant used to prevent the denominator from being zero and is set according to the specific implementation scenario.
[0102] An adaptive recursive optimization algorithm is used to optimize the model parameters of the iterative medical analysis model until the adaptive loss function converges to a preset value, or the optimization iteration is ended when a preset number of iterations is reached. The preset value and the preset number of iterations are set according to the specific implementation scenario.
[0103] Next, the medical analysis model is verified and evaluated to ensure its effectiveness and reliability in practical applications. The verification and evaluation process is as follows:
[0104] The cross-validation method is used to train and test the medical analysis model to evaluate the generalization ability of the medical analysis model. Then, the performance of the medical analysis model is quantified through indicators such as accuracy, precision, recall, and F1 score. In addition, the ROC curve and AUC value are used to evaluate the classification ability of the medical analysis model. The verification and evaluation process uses existing technologies and will not be described in detail here.
[0105] Finally, the verified and evaluated medical analysis model is deployed in the actual environment, and the performance of the medical analysis model is continuously monitored. The medical analysis model is optimized using an adaptive recursive optimization algorithm based on actual usage feedback to adapt to the ever-changing medical data and medical analysis needs, thereby improving the overall effectiveness and practicality of the medical analysis model.
[0106] In summary, a system and method for medical big data analysis and prediction have been completed.
[0107] The order of the embodiments of the invention is for description only and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0108] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.
[0109] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention.
Claims
1. A method for analyzing and predicting medical big data, characterized in that: The following steps are involved: S1: Collect and clean the original medical data to obtain medical data; build a medical analysis model based on the graph structure and feature tree, analyze the medical data, and generate risk prediction values; S2: Based on the risk prediction value, use the adaptive recursive optimization algorithm to dynamically adjust the model parameters of the medical analysis model, verify and evaluate the medical analysis model, deploy the verified and evaluated medical analysis model into the actual environment, and continuously monitor the performance of the medical analysis model.
2. The method for analyzing and predicting medical big data according to claim 1, characterized in that: The S1 specifically includes: Medical data is used to construct a graph structure and define the node feature vector of each node. The node feature vector of the node is linearly transformed and nonlinearly activated using a dynamic feature mapping function to obtain the mapping feature vector of the node. The mapping feature vector formula is as follows: Among them, z i is the mapped feature vector of node i; exp is the exponential operation; W1 is the first-layer weight matrix, which is used to linearly adjust the dimension and direction after feature mapping; W2 is the second-layer weight matrix, which is used to adjust the node feature vector x of node i i Perform linear transformation; x i is the node feature vector of node i; b1 is the first-layer bias vector, which is used for bias adjustment after the first-layer linear transformation; b2 is the second-layer bias vector, which is used for bias adjustment after the second-layer linear transformation; tanh is the hyperbolic tangent function; ||·|‖ represents the Euclidean norm of the vector.
3. The method for analyzing and predicting medical big data according to claim 2, characterized in that: The S1 specifically includes: The association weights between the mapping feature vectors of the calculation nodes are calculated as follows: Among them, α ik is the association weight between the mapping feature vector of node i and the mapping feature vector of node k; is the transpose operation; A is the adaptive weight matrix; z k is the mapping feature vector of node k; k is the index variable used to traverse the neighbor node set of node i; is the set of neighbor nodes of node i; The mapped feature vector of the node is weighted by using the associated weight to obtain the enhanced feature vector of the node; Among them, z′ i is the enhanced feature vector of node i.
4. The method for analyzing and predicting medical big data according to claim 3, characterized in that: The S1 specifically includes: Construct a feature tree, the nodes of which include leaf nodes, intermediate nodes, and root nodes; use the enhanced feature vectors of the nodes in the graph structure as the enhanced feature vectors of the leaf nodes of the feature tree for hierarchical modeling, and calculate the feature influence of each node in the feature tree; The feature influence of the leaf node is obtained using the feature contribution function. The feature contribution function formula is as follows: Among them, β i represents the feature influence of leaf node i; f(z′ i ) is the feature contribution function of leaf node i, which is used as the feature influence of leaf node i when building the medical analysis model; u is the weight vector; b is the bias term; The formula for the characteristic influence of the root node and the intermediate node is as follows: Among them, w pq is the connection weight between node p and node q; node q represents the child node of node p; σ is the parameter controlling the width of the Gaussian kernel function; z′ p is the enhanced feature vector of node p; z′ q is the enhanced feature vector of node q; β p is the characteristic influence of node p; γ is the adjustment coefficient; child(p) is the set of child nodes of node p; β q is the characteristic influence of node q.
5. The method for analyzing and predicting medical big data according to claim 4, characterized in that: The S1 specifically includes: The risk prediction value is calculated based on the feature influence of each node in the feature tree. The specific formula is: Among them, R i is the risk prediction value; nodes is the set of all nodes in the feature tree; ζ is the node index variable in the feature tree; β ζ is the characteristic influence of node ζ.
6. The method for analyzing and predicting medical big data according to claim 1, characterized in that: The S2 specifically includes: In the implementation process of the adaptive recursive optimization algorithm, an adaptive loss function is defined to measure the difference between the predicted results of the medical analysis model and the actual labels.
7. The method for analyzing and predicting medical big data according to claim 6, characterized in that: The S2 specifically includes: The adaptive loss function is used to dynamically adjust the model parameters of the medical analysis model, including feature influence, connection weights, weight vectors, and bias terms.
8. A system for analyzing and predicting medical big data, applied to the method for analyzing and predicting medical big data according to claim 1, characterized in that: Includes the following sections: Data preprocessing module, feature extraction module, feature tree modeling module, prediction and optimization module, verification and deployment module; A data preprocessing module collects raw medical data and performs data cleaning on the raw medical data to obtain medical data. The data preprocessing module is connected to the feature extraction module through data transmission; Use medical data to build a medical analysis model. The medical analysis model generates risk prediction values by analyzing and calculating medical data. The specific process is implemented through feature extraction modules and feature tree modeling modules. The feature extraction module constructs a graph structure based on medical data and generates mapping feature vectors of nodes of the graph structure using a dynamic feature mapping function; according to the association weights between the mapping feature vectors of nodes of the graph structure and the mapping feature vectors of neighboring nodes, the enhanced feature vectors of nodes of the graph structure are calculated and generated. The feature extraction module is connected to the feature tree modeling module by means of data transmission; The feature tree modeling module constructs a feature tree, uses the enhanced feature vectors of the nodes of the graph structure as the enhanced feature vectors of the leaf nodes of the feature tree for hierarchical modeling, and calculates the feature influence of each node in the feature tree; The risk prediction value is calculated based on the feature influence of each node in the feature tree; The tree modeling module is connected to the prediction and optimization module by means of data transmission; The prediction and optimization module uses an adaptive recursive optimization algorithm to dynamically adjust feature influence, connection weights, weight vectors, and bias terms based on an adaptive loss function. The prediction and optimization module is connected to the verification and deployment module through data transmission. The verification and deployment module is used to verify and evaluate the medical analysis model, deploy the verified and evaluated medical analysis model in the actual environment, monitor the performance of the medical analysis model, and continuously optimize the medical analysis model using an adaptive recursive optimization algorithm based on actual usage feedback.