Health data visualization analysis and decision system based on deep learning

Abstract

The application relates to the technical field of machine learning, and relates to a health data visualization analysis and decision system based on deep learning, which comprises a data standardization module, a multi-dimensional feature analysis module, a feature weighting tensor synthesis module, a feature fusion module, a health state intelligent evaluation module and a decision visualization module, carries out data cleaning on health data of a target user to obtain a standard health data set, carries out multi-dimensional feature analysis on the standard health data set to obtain an initial feature set, assigns weights to the initial feature set and carries out tensor synthesis to obtain a weighted feature vector, carries out multi-level fusion on the weighted feature vector and the initial feature set to obtain a fusion driving signal, intelligently evaluates the fusion driving signal to obtain a health state analysis result and a decision suggestion, carries out graphical rendering on the health state analysis result and the decision suggestion to obtain a dynamic visualization decision interface, and the application can improve the efficiency of health data visualization analysis and decision based on deep learning.

Description

Technical Field

[0001] This invention relates to the field of deep learning technology, and in particular to a health data visualization analysis and decision-making system based on deep learning. Background Technology

[0002] In the current field of health data processing and analysis, traditional methods lack standardized cleaning and feature parsing processes. Raw health data often contains redundancy and anomalies, making it impossible to create a unified, standardized dataset. Multi-dimensional feature extraction also struggles to fully uncover the core attributes of health data, resulting in an unstable data foundation for subsequent analysis and compromising the efficiency and completeness of health feature extraction. Furthermore, existing technologies do not incorporate clinical rule knowledge bases for scientific weighting and tensor synthesis of health features. Feature weight allocation lacks clinical basis, failing to highlight the value of key health indicators and leading to low feature utilization efficiency.

[0003] Existing health data analysis systems mostly employ a single-feature processing model, failing to achieve multi-level fusion of weighted features and initial features. Their feature interaction and deep information mining capabilities are insufficient, making it difficult to generate effective fusion-driven signals, resulting in low accuracy of health status assessment results. Furthermore, health assessment results and decision recommendations lack graphical and dynamic visualization methods, making it difficult to intuitively display complex health data and decision information. Users struggle to quickly understand analysis conclusions and implement decision recommendations, resulting in low efficiency and practicality in generating reference information for overall health data visualization analysis and decision-making. Summary of the Invention

[0004] To achieve the above objectives, the present invention provides a health data visualization analysis and decision-making system based on deep learning, characterized in that the system includes a data standardization module, a multi-dimensional feature parsing module, a feature weighted tensor synthesis module, a feature fusion module, a health status intelligent assessment module, and a decision visualization module, wherein: The data standardization module is used to clean the health data of the target users and obtain a standard health dataset for the target users. The multidimensional feature parsing module is used to perform multidimensional feature parsing on the standard health dataset to obtain the initial feature set of the health data. The feature weighted tensor synthesis module is used to assign weights to the initial feature set based on a predefined clinical rule knowledge base, and then synthesize the weighted feature dataset into a tensor to obtain a weighted feature vector of health data. The feature fusion module is used to perform multi-level feature fusion on the weighted feature vector and the initial feature set to obtain the fusion driving signal of health data. The intelligent health status assessment module is used to intelligently assess fused driving signals based on a clinical rule knowledge base, and obtain the health status analysis results and decision-making suggestions data of the target user. The decision visualization module is used to graphically render the health status analysis results and decision suggestion data, resulting in a dynamic and visualized decision interface for health data.

[0005] In a preferred embodiment, when the data standardization module performs data cleaning on the target user's health data to obtain a standard health dataset for the target user, it is specifically used for: Extract the target user's raw health data; Redundant data is removed from the original health data to obtain the clean health data of the target user; Anomaly detection is performed on cleanliness and health data to obtain compliant and corrected health data for the target user; The compliant corrected health data is normalized and mapped to obtain the standard health dataset for the target user.

[0006] In a preferred embodiment, when the multidimensional feature parsing module performs multidimensional feature parsing on a standard health dataset to obtain an initial feature set of the health data, it is specifically used for: The standard health dataset is divided into analytical dimensions to obtain a multi-dimensional analytical framework for the standard health dataset. Based on a multi-dimensional parsing framework, data extraction is performed on a standard health dataset to obtain multi-dimensional feature data of the standard health dataset. Multidimensional feature data is integrated into an initial feature set for health data.

[0007] In a preferred embodiment, when the feature weighted tensor synthesis module performs weight allocation on the initial feature set based on a predefined clinical rule knowledge base, and then performs tensor synthesis on the weighted feature dataset to obtain a weighted feature vector of the health data, it is specifically used for: Based on a clinical rule knowledge base, the initial feature set is clinically relevant to be assessed, and the weighting factors of the initial feature set are obtained. Based on the weighting factors, the initial feature set is weighted and adjusted to obtain a weighted feature dataset of health data. Tensor quantization is applied to the weighted feature dataset to obtain the weighted feature vector of the health data.

[0008] In a preferred embodiment, when the feature-weighted tensor synthesis module performs a clinical relevance assessment of the initial feature set based on a clinical rule knowledge base to obtain the weighting factors of the initial feature set, it is specifically used for: The clinical rule knowledge base is matched with association rules to obtain the feature rule association table of the initial feature set; Semantic parsing is performed on the feature rule association table to obtain the relevance strength set of the feature rule association table; Based on a predefined weight mapping relationship, the weight values ​​in the correlation intensity set are allocated and transformed to obtain the basic weight set of the correlation intensity set. Consistent fusion of the basic weight set yields the weight factors of the initial feature set.

[0009] In a preferred embodiment, when the feature fusion module performs multi-level feature fusion on the weighted feature vector and the initial feature set to obtain the fusion driving signal for health data, it is specifically used for: Align the weighted feature vectors and the initial feature set by dimensions to obtain aligned feature groups for the health data; Based on the aligned feature groups, the weighted feature vectors and the initial feature set are weighted and concatenated to obtain shallow fusion feature blocks of health data; By performing feature interaction on shallow fusion feature blocks, a deep fusion feature representation of health data is obtained; By performing cross-level feature fusion between shallow fusion feature blocks and deep fusion feature representations, a fusion driving signal for health data is obtained.

[0010] In a preferred embodiment, when the feature fusion module performs feature interaction on the shallow fusion feature blocks to obtain a deep fusion feature representation of the health data, it is specifically used for: Substructure division is performed on the shallow fusion feature blocks to obtain the interaction sequence of health data; Weights are assigned to the interaction sequences to obtain an attention weight map of the health data; Based on the attention weight map, the interaction sequence is weighted to obtain the interaction feature set of health data; The interaction feature set is dimensionally compressed to obtain a refined feature representation of the health data; By performing residual connections between the refined feature representation and the shallow fusion feature block, a deep fusion feature representation of the health data is obtained.

[0011] In a preferred embodiment, when the intelligent health status assessment module performs intelligent assessment of the fused driving signals based on a clinical rule knowledge base to obtain the target user's health status analysis results and decision suggestion data, it is specifically used for: Decision inference is performed on the fused driving signals to obtain a preliminary health status classification of the health data; Based on a clinical rule knowledge base, the clinical indicators for preliminary health status classification are validated for conformity, and the health status determination of the health data is obtained. Based on health status assessment, decision factors are extracted from the clinical rule knowledge base to obtain a multi-dimensional decision factor set for health data. By jointly reasoning with a multi-dimensional set of decision factors and fused driving signals, decision suggestion data for the target user is obtained. The verified health status assessment and personalized decision-making suggestions are structured and encapsulated to obtain the health status analysis results and decision-making suggestions for the target user.

[0012] In a preferred embodiment, when the intelligent health status assessment module performs conformity verification of clinical indicators for preliminary health status classification based on a clinical rule knowledge base to obtain a health status determination from health data, it is specifically used for: Indices were extracted from the preliminary health status classification to obtain a clinical indicator set of health data; The clinical indicators are verified against the clinical rule knowledge base to obtain a conclusion on the conformity assessment of health data. Logical judgment is performed on the compliance assessment conclusions to obtain the conclusion that the health data has been validated. The conclusions of the verification are linked to the preliminary health status classification to obtain the health status determination of the health data.

[0013] In a preferred embodiment, when the decision visualization module performs graphical rendering of the health status analysis results and decision suggestion data to obtain a dynamic visualization decision interface for health data, it is specifically used for: The health status levels in the health status analysis results are visually encoded and reorganized to obtain the initial interface layout of the health data. The suggestion content in the decision suggestion data is associated and bound with the area in the initial interface layout to obtain the mapping relationship between the data and visual elements in the health data; Based on the mapping relationship, the results of health status analysis and decision-making suggestions are visually synthesized to obtain an interactive visualization view of health data; By configuring interactive visualization views to enhance their interactivity, a dynamic visualization decision-making interface for health data can be obtained.

[0014] Compared with the prior art, the present invention has the following beneficial effects: 1. The technology of this invention comprehensively eliminates redundant and abnormal data through a standardized health data process, laying a precise data foundation for subsequent analysis. Relying on a multi-dimensional feature analysis framework, it can deeply mine the core attributes of health data, and combine this with a clinical rule knowledge base for scientific weight allocation and tensor synthesis, highlighting the value of key health indicators. Through multi-level feature fusion, it achieves efficient interaction between shallow and deep features, significantly improving the effectiveness of the fusion-driven signals and substantially enhancing the accuracy of health status analysis results and the efficiency of reference information generation.

[0015] 2. The system's decision visualization module, through intelligent coding and association binding, transforms health status analysis results and decision recommendations into a dynamic interactive interface, intuitively presenting data relationships and evolution trends. This visualization approach not only makes complex health data easier to understand but also accurately matches users' needs for decision-making information, enhancing the practicality and operability of decision recommendations and providing strong support for users to efficiently obtain health decision support. Attached Figure Description

[0016] Figure 1 This is a functional block diagram of a health data visualization analysis and decision-making system based on deep learning, provided in an embodiment of the present invention.

[0017] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments belong to some, but not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “said” and “the” as used in the embodiments of this invention and the appended claims are also intended to include the plural forms, and “multiple” generally includes at least two unless the context clearly indicates otherwise.

[0020] Depending on the context, the word "if" or "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0021] Furthermore, the timing of the steps in the following method embodiments is merely an example and not a strict limitation.

[0022] In practice, the server-side equipment deployed in a deep learning-based health data visualization analysis and decision-making system may consist of one or more devices. This deep learning-based health data visualization analysis and decision-making system can be implemented as a business instance, a virtual machine, or a hardware device. For example, this deep learning-based health data visualization analysis and decision-making system can be implemented as a business instance deployed on one or more devices in a cloud node. Simply put, this deep learning-based health data visualization analysis and decision-making system can be understood as software deployed on a cloud node, used to provide deep learning-based health data visualization analysis and decision-making services to various user terminals. Alternatively, this deep learning-based health data visualization analysis and decision-making system can also be implemented as a virtual machine deployed on one or more devices in a cloud node. This virtual machine contains application software for managing various user terminals. Alternatively, this deep learning-based health data visualization analysis and decision-making system can also be implemented as a server composed of numerous identical or different types of hardware devices, with one or more hardware devices configured to provide deep learning-based health data visualization analysis and decision-making services to various user terminals.

[0023] In terms of implementation, the deep learning-based health data visualization analysis and decision-making system and the user terminal are mutually adaptable. That is, if the deep learning-based health data visualization analysis and decision-making system is implemented as an application installed on a cloud service platform, then the user terminal is implemented as a client that establishes a communication connection with the application; or if the deep learning-based health data visualization analysis and decision-making system is implemented as a website, then the user terminal is implemented as a webpage; or if the deep learning-based health data visualization analysis and decision-making system is implemented as a cloud service platform, then the user terminal is implemented as a mini-program in an instant messaging application.

[0024] like Figure 1 The diagram shown is a functional block diagram of a health data visualization analysis and decision-making system based on deep learning, provided in an embodiment of the present invention.

[0025] The deep learning-based health data visualization analysis and decision-making system 100 described in this invention can be located on a cloud server. In terms of implementation, it can be used as one or more service devices, or as an application installed in the cloud (e.g., a mobile service operator's server, server cluster, etc.), or it can be developed as a website. Depending on the functions implemented, the deep learning-based health data visualization analysis and decision-making system 100 may include a data standardization module 101, a multi-dimensional feature parsing module 102, a feature weighted tensor synthesis module 103, a feature fusion module 104, a health status intelligent assessment module 105, and a decision visualization module 106. The modules described in this invention can also be called units, referring to a series of computer program segments that can be executed by the processor of an electronic device and perform a fixed function, stored in the memory of the electronic device.

[0026] In this embodiment of the invention, in the deep learning-based health data visualization analysis and decision-making system, each of the above modules can be implemented independently and can call other modules. Here, "calling" can be understood as one module connecting to multiple modules of another type and providing corresponding services to those connected modules. In the deep learning-based health data visualization analysis and decision-making system provided by this embodiment of the invention, the applicability of the deep learning-based health data visualization analysis and decision-making system architecture can be adjusted by adding modules and directly calling them without modifying the program code, achieving cluster-based horizontal expansion to quickly and flexibly expand the deep learning-based health data visualization analysis and decision-making system. In practical applications, the above modules can be set in the same device or different devices, or they can be set in virtual devices, such as service instances in a cloud server.

[0027] The following describes the components and specific workflow of a deep learning-based health data visualization analysis and decision-making system, using specific embodiments as examples: The data standardization module 101 is used to clean the health data of the target users and obtain the standard health dataset of the target users. In this embodiment of the invention, when the data standardization module performs data cleaning on the target user's health data to obtain a standard health dataset for the target user, it is specifically used for: Extract the target user's raw health data; Redundant data is removed from the original health data to obtain the clean health data of the target user; Anomaly detection is performed on cleanliness and health data to obtain compliant and corrected health data for the target user; The compliant corrected health data is normalized and mapped to obtain the standard health dataset for the target user.

[0028] From storage media such as local storage of health monitoring devices, user-specific databases of health management platforms, and health data files uploaded by user terminals, all health-related data associated with the unique identifier of the target user are retrieved according to preset health data categories and dimensions such as physiological indicators, lifestyle behaviors, and physical examination reports. During the retrieval process, the original attribute information of each data point, such as collection time, collection device model, and data dimension tags, is fully preserved. All retrieved health-related data are classified and integrated to ultimately form the target user's original health data.

[0029] The original health data of the target user is checked item by item. Multiple data entries with completely identical values ​​under the same collection time, collection device, and data dimension are identified as duplicate and redundant data. Health data without a unique identifier for the target user and without a collection time are identified as identifier redundant data. Irrelevant data that is not related to the user's physiological health or lifestyle health status is identified as dimension redundant data. The above three types of redundant data are deleted one by one. Only the valid data entries that are not duplicated, have valid identifiers, and belong to the health dimension are retained. All valid data entries are re-integrated according to the original data category dimension to finally form the clean health data of the target user.

[0030] For each health data dimension in the clean and healthy data, a corresponding industry-standard reference range is matched. Based on this reference range, fixed upper and lower threshold values ​​are set for each health data dimension. For each data entry in the clean and healthy data, the corresponding upper and lower threshold values ​​are matched according to its data dimension label. The data value is checked one by one to see if it is within the range defined by the upper and lower threshold values. If the data value exceeds the corresponding upper and lower threshold value range, the data value is corrected to the nearest threshold value within the corresponding dimension's upper and lower threshold value range. If the data value is within the corresponding upper and lower threshold value range, the data value remains unchanged. After completing the numerical verification and correction operations for all data entries in the clean and healthy data, all processed data entries are integrated according to the original data category dimension to finally form the compliant corrected health data for the target user.

[0031] To address the issue of numerical range discrepancies across different health data dimensions in compliant health data correction, a corresponding numerical mapping reference interval is defined for each health data dimension. This reference interval is the upper and lower limit threshold range of the corresponding health data dimension determined after outlier detection. All data values ​​under each health data dimension in the compliant health data correction are uniformly mapped to a fixed numerical range of 0 to 1. The minimum data value within each health data dimension's mapping reference interval is converted to 0, and the maximum data value within the same interval is converted to 1. The remaining data values ​​under each health dimension are converted to matching values ​​within the fixed 0 to 1 range based on their actual position within that dimension's mapping reference interval. After completing the mapping conversion of data values ​​under all health data dimensions, the original attribute information of each data entry, such as collection time, collection device model, and data dimension label, is fully preserved. All data entries that have completed the mapping conversion are then reorganized according to their original data category dimensions to ultimately form a standard health dataset for the target user.

[0032] The multidimensional feature parsing module 102 is used to perform multidimensional feature parsing on the standard health dataset to obtain the initial feature set of the health data. In this embodiment of the invention, when the multidimensional feature parsing module performs multidimensional feature parsing on a standard health dataset to obtain an initial feature set of the health data, it is specifically used for: The standard health dataset is divided into analytical dimensions to obtain a multi-dimensional analytical framework for the standard health dataset. Based on a multi-dimensional parsing framework, data extraction is performed on a standard health dataset to obtain multi-dimensional feature data of the standard health dataset. Multidimensional feature data is integrated into an initial feature set for health data.

[0033] Based on the core analytical directions preset for health data application scenarios, the standard health dataset is divided into three fixed analytical dimensions: physiological indicators, lifestyle behaviors, and time series. A multi-dimensional analytical framework is constructed, whereby the physiological indicators dimension is limited to four data items after normalization mapping: blood pressure, blood glucose, heart rate, and blood oxygen saturation; the lifestyle behaviors dimension is limited to three data items after normalization mapping: daily exercise duration, sleep duration, and meal frequency; and the time series dimension is limited to data groupings based on a 24-hour hourly collection cycle. Each analytical dimension corresponds to a unique dimension identifier and clear data item selection rules, with no data overlap between dimensions. This results in a multi-dimensional analytical framework for the standard health dataset with a fixed structure and clear scope.

[0034] Based on the identification and filtering rules of each dimension in the multi-dimensional analysis framework, all data entries in the standard health dataset are scanned one by one. Four types of data entries that match the data dimension labels with the physiological indicator dimension labels are extracted and classified into the physiological indicator dimension feature set. Three types of data entries that match the data dimension labels with the lifestyle behavior dimension labels are extracted and classified into the lifestyle behavior dimension feature set. All data entries that completely correspond to the data collection time and the hourly time period of the time series dimension are extracted and classified into the time series dimension feature set in time series order. During the extraction process, the normalized mapping attribute and the integrity of the original attribute of each data entry are checked simultaneously. Only data entries with complete attributes are retained. The corresponding dimension labels are retained in the three feature sets respectively. Finally, the multi-dimensional feature data of the standard health dataset is obtained.

[0035] Following a fixed order of physiological indicators, lifestyle behaviors, and time series dimensions, multi-dimensional feature data are structurally integrated. Each feature set is prefixed with a dimension name to ensure accurate differentiation between different dimensional feature data. At the same time, all attribute information of each feature data, such as the original collection time, collection device model, normalized value, and dimension label, is retained. During the integration process, data entries with duplicate dimension labels are removed, and only a single complete data entry is retained. Finally, an initial feature set of health data is formed, containing three dimensional feature sets, complete attributes, a regular structure, and no redundant data.

[0036] The feature weighted tensor synthesis module 103 is used to assign weights to the initial feature set based on a predefined clinical rule knowledge base, and to synthesize the feature dataset after weight assignment into a tensor to obtain a weighted feature vector of health data. In this embodiment of the invention, when the feature weighted tensor synthesis module performs weight allocation on the initial feature set based on a predefined clinical rule knowledge base, and then performs tensor synthesis on the weighted feature dataset to obtain a weighted feature vector of the health data, it is specifically used for: Based on a clinical rule knowledge base, the initial feature set is clinically relevant to be assessed, and the weighting factors of the initial feature set are obtained. Based on the weighting factors, the initial feature set is weighted and adjusted to obtain a weighted feature dataset of health data. Tensor quantization is applied to the weighted feature dataset to obtain the weighted feature vector of the health data.

[0037] The feature-weighted tensor synthesis module, when performing a clinical relevance assessment of the initial feature set based on a clinical rule knowledge base to obtain the weighting factors of the initial feature set, is specifically used for: The clinical rule knowledge base is matched with association rules to obtain the feature rule association table of the initial feature set; Semantic parsing is performed on the feature rule association table to obtain the relevance strength set of the feature rule association table; Based on a predefined weight mapping relationship, the weight values ​​in the correlation intensity set are allocated and transformed to obtain the basic weight set of the correlation intensity set. Consistent fusion of the basic weight set yields the weight factors of the initial feature set.

[0038] The predefined clinical rule knowledge base includes guidelines for the diagnosis and treatment of common chronic diseases, clinical correlation standards for physiological indicators, and classification standards for the health impact of lifestyle behaviors. It clearly defines the correlation level and corresponding weight standards between each health characteristic and clinical health assessment: core correlated features are assigned a weight of 0.8, secondary correlated features 0.5, general correlated features 0.3, and features with no clinical correlation 0.0. All feature entries in the three dimensions of the initial feature set of health data are extracted one by one. Each feature entry is precisely compared with the standardized entries in the clinical rule knowledge base to determine the clinical correlation level of each feature. Based on the level, a unique fixed weight value is assigned to each feature entry, and the knowledge base standardized entry number on which the weight assignment is based is recorded simultaneously. This ultimately forms the weight factors of the initial feature set, each corresponding to a specific initial feature set.

[0039] For the feature entries in the initial feature set, which span three dimensions—physiological indicators, lifestyle behaviors, and time series—corresponding weight factors were associated with each feature entry, and a weighted adjustment was performed on each feature entry. During the adjustment process, the original attribute information of each feature entry, such as normalized values, collection time, and device model, was retained. Only a weight factor field was added to each feature entry, clearly indicating the corresponding weight value and the basis number for the assignment, ensuring that the weighted feature entries can be traced back to the clinical rule knowledge base. After weighted adjustment, the consistency of the association between each feature entry and the weight factor was verified one by one. Feature entries with incorrect weight factor assignments or no corresponding assignment basis were removed, and only feature entries with accurate associations and complete attributes were retained. These were then categorized and integrated according to the original three dimensions to finally obtain the weighted feature dataset of health data.

[0040] Following a fixed order of physiological indicators, lifestyle behaviors, and time series dimensions, the weighted feature dataset is structured and tensor-organized to construct a three-dimensional data tensor structure. The first dimension corresponds to the three feature groups: physiological indicators, lifestyle behaviors, and time series data. The second dimension corresponds to the specific feature items within each group. The third dimension corresponds to the normalized value, weight factor, and original attributes of each feature item. Data in each dimension is sorted from highest to lowest weight factor. Each tensor position corresponds to unique feature information and associated data, with no data gaps or redundancy. After constructing the tensor structure, the completeness and accuracy of the data in each dimension are verified to ensure a complete correspondence between the tensor structure and the weighted feature dataset, ultimately forming a weighted feature vector of health data.

[0041] The Clinical Rule Knowledge Base incorporates clinical practice guidelines, disease diagnostic criteria, and feature association specifications, covering the clinical application scenarios, associated disease types, and judgment criteria corresponding to each initial feature. The module first extracts the core attributes of each feature in the initial feature set, including feature name, data source, and clinical representation meaning. Then, it traverses all rule entries in the Clinical Rule Knowledge Base. Each rule entry includes prerequisite feature items, association conditions, and scope of application. When the core attributes of the initial feature completely match the prerequisite feature items of the rule entry, and the feature data source conforms to the scope of application of the rule, the association relationship between the feature and the corresponding rule entry is established. The feature identifier, rule identifier, and matching criteria are recorded in a table in sequence. All successfully matched association relationships and unmatched feature records are summarized to form a feature rule association table. Unmatched features are marked in the table as having no corresponding association rule.

[0042] For each record in the feature rule association table, the module extracts the association between features and rules. It then performs word mapping and semantic decomposition using a pre-defined clinical semantic parsing dictionary. This dictionary includes clinical professional terms, related words, and semantic level definitions. The related words are divided into three categories: necessary association, important association, and general association. During semantic decomposition, the module analyzes the related expressions word by word, identifies the related words, and combines them with the priority attributes of rules in the clinical rule knowledge base to determine the degree of association for each record. Necessary association corresponds to strong association in the semantic parsing results, important association corresponds to medium association, and general association corresponds to weak association. The feature identifier and corresponding association strength level of each record are organized in sequence to form a relevance strength set. Each data in the strength set corresponds one-to-one with a record in the feature rule association table.

[0043] The predefined weight mapping relationship is a fixed correspondence pre-defined based on clinical practice data. It specifies a weight value of 0.8 for strong association, 0.5 for moderate association, 0.2 for weak association, and 0.1 for no association. This mapping relationship has been validated with 500 clinical cases and accurately reflects the influence of features on clinical diagnosis. The module iterates through each strength level record in the correlation strength set, converting the corresponding strength level into a specific weight value according to the above fixed mapping relationship. Simultaneously, it retains the correspondence between the weight value and the feature identifier, as well as the original association rule, ensuring that the weight value can be traced back to the feature rule association table. All converted weight values ​​are categorized and organized according to feature identifiers to form a basic weight set. The same feature may correspond to multiple weight values ​​in the basic weight set.

[0044] The module first groups the weight values ​​in the basic weight set according to feature identifiers, with each group corresponding to a feature in the initial feature set. Then, it performs a consistency check on the weight values ​​within each group, with the check criterion being that the difference between all weight values ​​within the same group does not exceed 0.1. If all weight values ​​within the same group are completely identical, that weight value is directly used as the temporary weight for the corresponding feature. If there are differences within 0.1, the arithmetic mean of all weight values ​​is taken as the temporary weight, rounded to one decimal place. If the difference exceeds 0.1, the clinical priority of the corresponding association rule is used for selection. Priority is divided according to the recommendation level of the treatment guidelines, with the highest priority rule corresponding to the first-level guidelines. The weight value corresponding to the highest priority rule is selected as the temporary weight. After the temporary weights for all features are determined, an integrity check is performed to ensure that each feature in the initial feature set corresponds to a unique temporary weight, with no missing or duplicate weights. The set of temporary weights that passes the check is the weight factor of the initial feature set.

[0045] Feature fusion module 104 is used to perform multi-level feature fusion on weighted feature vectors and initial feature sets to obtain fusion driving signals for health data; In this embodiment of the invention, when the feature fusion module performs multi-level feature fusion on the weighted feature vector and the initial feature set to obtain the fusion driving signal of health data, it is specifically used for: Align the weighted feature vectors and the initial feature set by dimensions to obtain aligned feature groups for the health data; Based on the aligned feature groups, the weighted feature vectors and the initial feature set are weighted and concatenated to obtain shallow fusion feature blocks of health data; By performing feature interaction on shallow fusion feature blocks, a deep fusion feature representation of health data is obtained; By performing cross-level feature fusion between shallow fusion feature blocks and deep fusion feature representations, a fusion driving signal for health data is obtained.

[0046] When the feature fusion module performs feature interaction on the shallow fused feature blocks to obtain a deep fused feature representation of the health data, it is specifically used for: Substructure division is performed on the shallow fusion feature blocks to obtain the interaction sequence of health data; Weights are assigned to the interaction sequences to obtain an attention weight map of the health data; Based on the attention weight map, the interaction sequence is weighted to obtain the interaction feature set of health data; The interaction feature set is dimensionally compressed to obtain a refined feature representation of the health data; By performing residual connections between the refined feature representation and the shallow fusion feature block, a deep fusion feature representation of the health data is obtained.

[0047] The weighted feature vector is obtained by weighting the initial feature set with corresponding weight factors. Its feature dimensions may differ from the initial feature set. Dimension alignment is based on the feature dimensions of the initial feature set, which is fixed at 256 dimensions. The module first extracts the dimensionality information of the weighted feature vector and the initial feature set. If the weighted feature vector has 256 dimensions, it directly determines that the dimensions are consistent. If the weighted feature vector has more than 256 dimensions, it retains the dimensions with the same feature names as the initial feature set and deletes redundant dimensions. If the weighted feature vector has fewer than 256 dimensions, it adds 0 values ​​to the missing dimensions, and the feature names of the added dimensions are consistent with the feature names at the corresponding missing positions in the initial feature set. After dimensionality adjustment, the weighted feature vector and the initial feature set are combined according to their feature names to form aligned feature groups. Each group contains a weighted feature value corresponding to a feature name and an initial feature value.

[0048] Based on aligned feature groups, weighted concatenation employs a method of sequentially aligning and concatenating features according to their positions, with the concatenation order fixed as initial feature values ​​first, followed by corresponding weighted feature values. For each data set in the aligned feature group, initial and weighted feature values ​​are extracted, maintaining the feature identifier association between them. The initial and weighted feature values ​​of all groups are then concatenated sequentially according to their feature order to form a continuous feature sequence. Feature position indices are recorded synchronously during the concatenation process to ensure that the source of each feature value can be traced back to the aligned feature group. The resulting feature sequence after concatenation is the shallow fusion feature block, which has a fixed dimension of 512, strictly corresponding to the number of features in the aligned feature group.

[0049] When performing feature interactions on shallow fusion feature blocks, a neighborhood feature association method is adopted. Each feature element is centered, and two feature elements before and after it are associated to form an interaction window. Each window contains five consecutive feature elements. For each interaction window, it is determined whether there is a synergistic association between the clinical feature types corresponding to the feature elements within the window. If a synergistic association exists, the values ​​of all feature elements within the window are superimposed and integrated, and the integrated value is used as the interaction value of the central feature element. If no synergistic association exists, the original value of the central feature element is retained, and the influence of other elements within the window is discarded. By traversing all feature elements of the shallow fusion feature block in the above manner, the resulting new feature sequence is the deep fusion feature representation, whose dimension is the same as that of the shallow fusion feature block, both being 512-dimensional.

[0050] Cross-level feature fusion employs an element-level overlay fusion method, matching shallow fusion feature blocks with deep fusion feature representations one-to-one according to their corresponding feature elements. For each feature element, the element value of the shallow fusion feature block is added to the corresponding element value of the deep fusion feature representation to obtain the fusion value at that position. After overlay, all fusion values ​​undergo range verification, with the verification standard being that the fusion value is between 0 and 1000. If the fusion value is within this range, it is directly retained; if the fusion value is below 0, it is adjusted to 0; if the fusion value is above 1000, it is adjusted to 1000. After the fusion values ​​at all positions are verified, a continuous feature signal sequence is formed, which serves as the fusion driving signal for the health data.

[0051] The shallow fusion feature block is formed from health data after preprocessing and initial fusion. It contains multiple feature units arranged according to channel dimensions. The substructure is divided according to a fixed number of feature units. Each substructure consists of 8 consecutive feature units. The splitting process traverses all feature units in the shallow fusion feature block and extracts 8 feature units in sequence to form a substructure. If there are fewer than 8 feature units remaining, the remaining feature units are used as the last substructure. All substructures are combined in the original feature unit arrangement order to form an interaction sequence of health data. Each substructure corresponds to one element of the interaction sequence.

[0052] Each substructure in the interaction sequence is traversed, and the feature similarity between each substructure and all substructures in the interaction sequence is calculated. The similarity is judged based on the consistency of the numerical distribution of all feature units within the substructure. By comparing the numerical differences of feature units at corresponding positions in two substructures, the differences at all positions are accumulated and averaged. This average value is converted into a similarity score. The higher the similarity score, the lower the corresponding difference value. Then, the similarity score of each substructure is normalized so that the sum of the similarity scores of all substructures is 1. The normalized score is the weight value of the corresponding substructure. The weight values ​​are arranged in the order of the substructures in the interaction sequence to form a two-dimensional matrix form of the attention weight map of health data. The value at each position in the matrix corresponds to the weight value of the corresponding substructure in the interaction sequence.

[0053] The weight value of each position in the attention weight map is extracted and fused with the weight of each feature unit of the corresponding substructure in the interaction sequence. The fusion method is to directly multiply the original value of each feature unit with the corresponding weight value to obtain the weighted value of the feature unit. All feature units of all substructures in the interaction sequence are traversed one by one to complete the weighted operation of all feature units. The modulated feature units of each substructure are combined in the original order. Then, all substructures are integrated according to the arrangement order of the interaction sequence to form the interaction feature set of health data. Each element in the interaction feature set is a complete substructure feature after weighted modulation.

[0054] A feature selection method is used to compress the dimensionality of the interactive feature set. All feature units of each modulated substructure in the interactive feature set are traversed, and the numerical fluctuation amplitude of each feature unit in all substructures is calculated. The fluctuation amplitude is obtained by calculating the difference between the maximum and minimum values ​​of the feature unit in different substructures. The fluctuation amplitude threshold is set to 0.3. All feature units with fluctuation amplitude greater than 0.3 are retained, and feature units with fluctuation amplitude less than or equal to 0.3 are removed. The retained feature units are re-integrated according to the original arrangement order of the substructures and the position order of the feature units in the interactive feature set to form a refined feature representation of the health data with reduced dimensionality.

[0055] First, the refined feature representation is adapted to match the number of feature units in the refined feature representation with the number of feature units in the shallow fusion feature block. During the adaptation process, if the number of feature units in the refined feature representation is less than that in the shallow fusion feature block, feature units with a value of 0 are added sequentially at the end of the refined feature representation until the number matches. If the number of feature units in the refined feature representation is more than that in the shallow fusion feature block, feature units exceeding the limit at the end are removed according to the arrangement of feature units. Then, the refined feature representation and the feature units at the corresponding positions in the shallow fusion feature block are numerically superimposed. The final value of each feature unit is the sum of the corresponding values ​​of the two. The complete feature set obtained after superposition is the deep fusion feature representation of the health data.

[0056] The intelligent health status assessment module 105 is used to intelligently assess the fusion driving signals based on the clinical rule knowledge base, and obtain the health status analysis results and decision suggestion data of the target user. In this embodiment of the invention, when the intelligent health status assessment module performs intelligent assessment of the fused driving signals based on a clinical rule knowledge base to obtain the target user's health status analysis results and decision suggestion data, it is specifically used for: Decision inference is performed on the fused driving signals to obtain a preliminary health status classification of the health data; Based on a clinical rule knowledge base, the clinical indicators for preliminary health status classification are validated for conformity, and the health status determination of the health data is obtained. Based on health status assessment, decision factors are extracted from the clinical rule knowledge base to obtain a multi-dimensional decision factor set for health data. By jointly reasoning with a multi-dimensional set of decision factors and fused driving signals, decision suggestion data for the target user is obtained. The verified health status assessment and personalized decision-making suggestions are structured and encapsulated to obtain the health status analysis results and decision-making suggestions for the target user.

[0057] The intelligent health status assessment module, when performing conformity verification of clinical indicators for preliminary health status classification based on a clinical rule knowledge base to obtain a health status determination from health data, is specifically used for: Indices were extracted from the preliminary health status classification to obtain a clinical indicator set of health data; The clinical indicators are verified against the clinical rule knowledge base to obtain a conclusion on the conformity assessment of health data. Logical judgment is performed on the compliance assessment conclusions to obtain the conclusion that the health data has been validated. The conclusions of the verification are linked to the preliminary health status classification to obtain the health status determination of the health data.

[0058] The fusion-driven signal is obtained by signal conversion of the deep fusion feature representation output by the feature fusion module. It includes multi-dimensional physiological indicator features and feature association information of the target user. The decision inference process revolves around the feature range corresponding to the preset health category. The preset health category is divided into four categories: healthy, subclinical, mild, and severe. Each category corresponds to a clear physiological indicator feature range. All features in the fusion-driven signal are traversed one by one, and the matching situation of each feature with the feature range of each category is compared. The proportion of matching features to the total number of features is counted. The proportion threshold is set at 90%. If the matching proportion of a single category reaches 90% or above, the category is determined as the preliminary health status classification. If no category matching proportion reaches 90%, the category with the highest matching proportion is taken as the preliminary health status classification and marked as a state to be verified. Finally, the preliminary health status classification of the health data is formed.

[0059] The clinical rules knowledge base includes core clinical indicators corresponding to various health states from authoritative clinical diagnosis and treatment guidelines and industry standards. Each clinical indicator has a clearly defined characteristic judgment standard. Based on the preliminary health state classification, all mandatory clinical indicators of the corresponding category in the knowledge base are retrieved. The driving signal is checked one by one to see if there are any features corresponding to the mandatory indicators. The verification standard is that the feature values ​​fall completely within the indication judgment standard range. If all mandatory clinical indicators pass the verification and there are no conflicting indicators, the preliminary health state classification is maintained as the health state judgment of the health data. If one or more mandatory clinical indicators fail or conflicting indicators appear, the health data is adjusted to the corresponding appropriate health category in the knowledge base based on the characteristic manifestations of the failed indicators, thus forming the final health state judgment of the health data.

[0060] Once the health status is determined, a clinical rule knowledge base is retrieved based on the determination result to extract decision factors directly related to the health status. These decision factors include four categories: physiological indicator monitoring items, intervention measure suitability conditions, risk warning thresholds, and rehabilitation cycle reference values. During the extraction process, only factors directly related to the current health status and without redundancy are retained. Each factor corresponds to a clear clinical basis and characteristic correlation. After extraction, the data is classified and organized according to four dimensions: physiological indicators, intervention measures, risk warnings, and rehabilitation cycles. Factors under each dimension are ranked by importance to form a multi-dimensional decision factor set for health data.

[0061] Each factor in the multi-dimensional decision factor set is associated and matched with the corresponding feature in the fusion driving signal. Each decision factor corresponds to a specific feature item in the fusion driving signal. After matching, joint reasoning is performed based on the factor-feature association reasoning logic preset in the clinical rule knowledge base. The reasoning process combines the factor requirements with the actual values ​​of the corresponding features to generate targeted suggestion items. For example, intervention measure factors are combined with feature values ​​to determine the specific intervention method and execution frequency, and risk warning factors are combined with feature values ​​to clarify the warning level and monitoring frequency. All suggestion items are classified and integrated according to the decision factor dimensions to form decision suggestion data for the target user.

[0062] The structured encapsulation uses a fixed data format and is divided into a health status assessment area and a decision suggestion data area. The health status assessment area is filled with the health status assessment results verified by the clinical rule knowledge base, and the core clinical indicators and verification results on which the assessment is based are marked. The decision suggestion data area is filled with personalized decision suggestion data according to the dimensional order of the multi-dimensional decision factor set. The personalization is reflected in the adjustment of suggestion details based on the target user's basic information. For example, the description of the rehabilitation cycle suggestion is optimized for elderly users. After encapsulation, the result is a health status analysis result and decision suggestion data of the target user with fixed format partitions, complete content and clear logic.

[0063] Based on the preliminary health status classification, the information fields associated with the classification in the health data are traversed to extract the preset clinical indicators. Each indicator must correspond to a clear data dimension, including physiological indicators, symptom manifestations, medical history information, etc. During the extraction process, only indicators with complete data and standardized format are retained, and indicators corresponding to missing values ​​and abnormal format data are removed. Finally, a unified set of clinical indicators is formed. This set must completely cover all the mandatory indicators corresponding to the preliminary health status classification, without omissions or redundancies.

[0064] The clinical rule knowledge base is divided into rule subsets according to health status. Each subset contains the judgment criteria for each clinical indication under the corresponding category. Each indication item in the clinical indication set is compared with the standard of the corresponding rule subset one by one to verify whether the judgment requirements are met. The compliance result of a single indication item is recorded. The proportion of the number of indication items that comply with the rules to the total number of clinical indication items is counted. When the proportion is ≥80%, the compliance assessment conclusion is "high compliance", when the proportion is 40%-79%, it is "medium compliance", and when the proportion is <40%, it is "low compliance", ensuring that the conclusion directly corresponds to the verification process.

[0065] The pre-defined logical judgment rules are as follows: When the compliance assessment conclusion is "high compliance", the verification is directly judged as passed; when the conclusion is "medium compliance", it is checked whether the indicators that do not comply with the rules are non-core indicators. If all non-compliance items are non-core indicators, the verification is judged as passed; if there are core indicators that do not comply, the verification is judged as failed; when the conclusion is "low compliance", the verification is directly judged as failed. Finally, a verification success conclusion containing only two results, "verification passed" or "verification failed", is generated.

[0066] If the verification conclusion is "verification passed", the preliminary health status classification is used as the core content and bound to the "verification passed" label to form a health status judgment result, clarifying that the final health status corresponding to the health data is the preliminary classification result; if the verification conclusion is "verification failed", the preliminary health status classification is bound to the "verification failed" label, and the names of the core indicators that do not conform to the rules and the specific discrepancies are marked to form a complete health status judgment result, ensuring that the result contains both preliminary classification information and verification conclusion, with no missing information.

[0067] The decision visualization module 106 is used to graphically render the health status analysis results and decision suggestion data to obtain a dynamic visualization decision interface for health data.

[0068] In this embodiment of the invention, when the decision visualization module performs graphical rendering of the health status analysis results and decision suggestion data to obtain a dynamic visualization decision interface for health data, it is specifically used for: The health status levels in the health status analysis results are visually encoded and reorganized to obtain the initial interface layout of the health data. The suggestion content in the decision suggestion data is associated and bound with the area in the initial interface layout to obtain the mapping relationship between the data and visual elements in the health data; Based on the mapping relationship, the results of health status analysis and decision-making suggestions are visually synthesized to obtain an interactive visualization view of health data; By configuring interactive visualization views to enhance their interactivity, a dynamic visualization decision-making interface for health data can be obtained.

[0069] The preset health status levels are divided into five fixed levels: Level 1, Level 2, Level 3, Level 4, and Level 5. Each level has its own unique visual coding rules. Color coding is configured according to the RGB color model as follows: Level 1 RGB(0,255,0), Level 2 RGB(153,255,0), Level 3 RGB(255,255,0), Level 4 RGB(255,153,0), and Level 5 RGB(255,0,0). Shape coding is configured as circle, ellipse, square, rectangle, and triangle. Position coding is arranged sequentially on the left side of the interface, from highest to lowest level. In the vertical region, each level corresponds to a fixed vertical coordinate interval: Level 1 occupies the 0-20% interval of the Y-axis, Level 2 occupies the 20%-40% interval, Level 3 occupies the 40%-60% interval, Level 4 occupies the 60%-80% interval, and Level 5 occupies the 80%-100% interval. According to the above coding rules, each health status level in the health status analysis results is coded and converted one by one. Then, according to the fixed layout specification of placing level coding elements on the left, reserving a decision suggestion display area on the right, setting a fixed title bar at the top, and setting an operation prompt bar at the bottom, the layout is reorganized and arranged to obtain the initial interface layout of health data.

[0070] Each suggestion in the decision-making suggestion data is extracted and categorized according to its corresponding health status level. Suggestions at the same level are grouped together, and each group is assigned a unique identifier. This identifier consists of the corresponding health status level number and the group's sequence number; for example, the first suggestion at level one is numbered 1-01. For the display area corresponding to each health status level in the initial interface layout, a unique area identifier is assigned to each area. The area identifier is consistent with the corresponding level number. A mapping table is established to record the correspondence between the identifier of each group of suggestions and the area identifier of the corresponding area. The display position of each group of suggestions within its corresponding area is also clearly defined, arranged sequentially from top to bottom according to the order of suggestions within the group. Each suggestion occupies a fixed height space within the area, and the display height of a single suggestion is the total height of the corresponding area divided by the number of suggestions in that group. The resulting set of correspondences, including suggestion content identifiers, area identifiers, and display position information, represents the mapping relationship between data and visual elements in the health data.

[0071] Based on the established mapping relationship, the coded elements of each level in the health status analysis results are first rendered to the left area of ​​the initial interface layout, ensuring that the color, shape, and position of each level perfectly match the preset coding rules, with a 5-pixel gap between the edges of the coded elements and the boundaries of the corresponding areas, and no overlap or offset between elements. Then, the decision suggestion data is rendered item by item to the right area of ​​the initial interface layout according to the corresponding rules and display positions in the mapping relationship. The suggestion content is displayed in SimSun font, 12-point font, and black font color, with each suggestion separated by a 1-pixel gray dividing line. Simultaneously, a linkage relationship is established between the health status level coded elements and the corresponding decision suggestion content. When the mouse pointer hovers over a certain level coded element, the font color of the suggestion content in the corresponding area switches to the coded color corresponding to that level; after hovering away, the font color reverts to black. Integrating the above rendering results and linkage configuration, an interactive visualization view of the health data is obtained.

[0072] Add click-based interaction to the interactive visualization view. When a health status level code element is clicked, the corresponding suggestion content area automatically expands to 80% of the interface width, while other areas shrink to 5% of the interface width. Clicking on a blank area restores the initial width proportions of all areas, i.e., the level code area on the left occupies 20% of the interface width, and the suggestion display area on the right occupies 80% of the interface width. Add data update synchronization functionality. When the health status analysis results change, the visual presentation of the corresponding level code element remains unchanged, and the associated decision suggestion data is automatically updated according to the content corresponding to the new analysis results. The update process strictly follows the original mapping relationship layout rules. Add view zoom functionality, supporting mouse wheel scrolling to zoom the entire view. The zoom range is limited to 50% to 200% of the original view size. During zooming, the proportions of all elements remain consistent without distortion. After completing all interactive enhancement configurations, a dynamic visualization decision interface for health data is obtained.

[0073] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0074] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0075] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A health data visualization analysis and decision-making system based on deep learning, characterized in that, The system includes a data standardization module, a multi-dimensional feature parsing module, a feature weighted tensor synthesis module, a feature fusion module, a health status intelligent assessment module, and a decision visualization module, wherein: The data standardization module is used to clean the health data of the target users and obtain a standard health dataset for the target users. The multidimensional feature parsing module is used to perform multidimensional feature parsing on the standard health dataset to obtain the initial feature set of the health data. The feature weighted tensor synthesis module is used to assign weights to the initial feature set based on a predefined clinical rule knowledge base, and then synthesize the weighted feature dataset into a tensor to obtain a weighted feature vector of health data. The feature fusion module is used to perform multi-level feature fusion on the weighted feature vector and the initial feature set to obtain the fusion driving signal of health data. The intelligent health status assessment module is used to intelligently assess fused driving signals based on a clinical rule knowledge base, and obtain the health status analysis results and decision-making suggestions data of the target user. The decision visualization module is used to graphically render the health status analysis results and decision suggestion data, resulting in a dynamic and visualized decision interface for health data.

2. The health data visualization analysis and decision-making system based on deep learning as described in claim 1, characterized in that, When the data standardization module performs data cleaning on the target user's health data to obtain a standard health dataset for the target user, it is specifically used for: Extract the target user's raw health data; Redundant data is removed from the original health data to obtain the clean health data of the target user; Anomaly detection is performed on cleanliness and health data to obtain compliant and corrected health data for the target user; The compliant corrected health data is normalized and mapped to obtain the standard health dataset for the target user.

3. The health data visualization analysis and decision-making system based on deep learning as described in claim 1, characterized in that, When performing multi-dimensional feature parsing on a standard health dataset to obtain an initial feature set for the health data, the multi-dimensional feature parsing module is specifically used for: The standard health dataset is divided into analytical dimensions to obtain a multi-dimensional analytical framework for the standard health dataset. Based on a multi-dimensional parsing framework, data extraction is performed on a standard health dataset to obtain multi-dimensional feature data of the standard health dataset. Multidimensional feature data is integrated into an initial feature set for health data.

4. The health data visualization analysis and decision-making system based on deep learning as described in claim 1, characterized in that, The feature weighted tensor synthesis module, when executing the process of assigning weights to the initial feature set based on a predefined clinical rule knowledge base and synthesizing the weighted feature dataset into a tensor to obtain a weighted feature vector of health data, is specifically used for: Based on a clinical rule knowledge base, the initial feature set is clinically relevant to be assessed, and the weighting factors of the initial feature set are obtained. Based on the weighting factors, the initial feature set is weighted and adjusted to obtain a weighted feature dataset of health data. Tensor quantization is applied to the weighted feature dataset to obtain the weighted feature vector of the health data.

5. The deep learning-based health data visualization analysis and decision-making system as described in claim 4, characterized in that, The feature-weighted tensor synthesis module, when performing a clinical relevance assessment of the initial feature set based on a clinical rule knowledge base to obtain the weighting factors of the initial feature set, is specifically used for: The clinical rule knowledge base is matched with association rules to obtain the feature rule association table of the initial feature set; Semantic parsing is performed on the feature rule association table to obtain the relevance strength set of the feature rule association table; Based on a predefined weight mapping relationship, the weight values ​​in the correlation intensity set are allocated and transformed to obtain the basic weight set of the correlation intensity set. Consistent fusion of the basic weight set yields the weight factors of the initial feature set.

6. The health data visualization analysis and decision-making system based on deep learning as described in claim 1, characterized in that, When the feature fusion module performs multi-level feature fusion on the weighted feature vector and the initial feature set to obtain the fusion driving signal for health data, it is specifically used for: Align the weighted feature vectors and the initial feature set by dimensions to obtain aligned feature groups for the health data; Based on the aligned feature groups, the weighted feature vectors and the initial feature set are weighted and concatenated to obtain shallow fusion feature blocks of health data; By performing feature interaction on shallow fusion feature blocks, a deep fusion feature representation of health data is obtained; By performing cross-level feature fusion between shallow fusion feature blocks and deep fusion feature representations, a fusion driving signal for health data is obtained.

7. The health data visualization analysis and decision-making system based on deep learning as described in claim 6, characterized in that, When the feature fusion module performs feature interaction on the shallow fused feature blocks to obtain a deep fused feature representation of the health data, it is specifically used for: Substructure division is performed on the shallow fusion feature blocks to obtain the interaction sequence of health data; Weights are assigned to the interaction sequences to obtain an attention weight map of the health data; Based on the attention weight map, the interaction sequence is weighted to obtain the interaction feature set of health data; The interaction feature set is dimensionally compressed to obtain a refined feature representation of the health data; By performing residual connections between the refined feature representation and the shallow fusion feature block, a deep fusion feature representation of the health data is obtained.

8. The health data visualization analysis and decision-making system based on deep learning as described in claim 1, characterized in that, When the intelligent health status assessment module performs intelligent assessment of fused driving signals based on a clinical rule knowledge base to obtain the target user's health status analysis results and decision-making suggestion data, it is specifically used for: Decision inference is performed on the fused driving signals to obtain a preliminary health status classification of the health data; Based on a clinical rule knowledge base, the clinical indicators for preliminary health status classification are validated for conformity, and the health status determination of the health data is obtained. Based on health status assessment, decision factors are extracted from the clinical rule knowledge base to obtain a multi-dimensional decision factor set for health data. By jointly reasoning with a multi-dimensional set of decision factors and fused driving signals, decision suggestion data for the target user is obtained. The verified health status assessment and personalized decision-making suggestions are structured and encapsulated to obtain the health status analysis results and decision-making suggestions for the target user.

9. The health data visualization analysis and decision-making system based on deep learning as described in claim 8, characterized in that, The intelligent health status assessment module, when performing conformity verification of clinical indicators for preliminary health status classification based on a clinical rule knowledge base to obtain a health status determination from health data, is specifically used for: Indices were extracted from the preliminary health status classification to obtain a clinical indicator set of health data; The clinical indicators are verified against the clinical rule knowledge base to obtain a conclusion on the conformity assessment of health data. Logical judgment is performed on the compliance assessment conclusions to obtain the conclusion that the health data has been validated. The conclusions of the verification are linked to the preliminary health status classification to obtain the health status determination of the health data.

10. The health data visualization analysis and decision-making system based on deep learning as described in claim 1, characterized in that, When the decision visualization module performs graphical rendering of health status analysis results and decision suggestion data to obtain a dynamic visualization decision interface for health data, it is specifically used for: The health status levels in the health status analysis results are visually encoded and reorganized to obtain the initial interface layout of the health data. The suggestion content in the decision suggestion data is associated and bound with the area in the initial interface layout to obtain the mapping relationship between the data and visual elements in the health data; Based on the mapping relationship, the results of health status analysis and decision-making suggestions are visually synthesized to obtain an interactive visualization view of health data; By configuring interactive visualization views to enhance their interactivity, a dynamic visualization decision-making interface for health data can be obtained.

Images