Internet-based health examination report data analysis management system

Abstract

This invention relates to the field of health data management technology, specifically to an internet-based health check-up report data analysis and management system. The system includes: receiving the original document of a user's health check-up report; parsing and extracting the user's health check-up data records; matching the report with a pre-built medical knowledge ontology database and performing semantic annotation to generate enhanced health check-up data records; comparing the detected values ​​with reference ranges, identifying abnormal detection items and generating preliminary abnormality markers; and performing correlation analysis on the abnormal detection items based on physiological associations in the medical knowledge ontology database to find combinations of abnormal items associated with physiological pathways. This system solves the problems of existing technologies where health check-up data lacks medical semantic associations and abnormality identification is isolated, improving the systematicness and completeness of health check-up data analysis, and is suitable for health check-up data management in an internet environment.

Description

Technical Field

[0001] This invention relates to the field of health data management technology, and in particular to an internet-based health checkup report data analysis and management system. Background Technology

[0002] Currently, among the technologies related to internet-based health check-up report management, there are already systems that can receive original check-up report documents and extract basic data. These systems can obtain basic information such as user identification, test item names, test values, units, and reference ranges from the check-up report, complete the preliminary sorting and storage of check-up data, and provide basic data query functions for users and medical personnel.

[0003] In existing technical solutions, the extracted physical examination data consists only of isolated numerical values ​​and item information, lacking corresponding medical semantic associations. It is impossible to clarify the medical meaning of the test items and the physiological relationship between the items, which makes subsequent data interpretation difficult. At the same time, anomaly identification can only simply compare the test values ​​with the reference range and determine whether a single item is abnormal. It cannot explore the physiological relationship between different abnormal items, and it is difficult to find abnormal combinations formed by multiple related abnormal items, thus failing to comprehensively reflect the user's health status.

[0004] Existing technologies cannot achieve semantic enhancement processing of physical examination data, resulting in a lack of interpretable medical correlations in the data. Furthermore, anomaly identification is limited to a single item and cannot uncover correlational anomalies based on physiological mechanisms. This makes it difficult to meet users' needs for in-depth analysis and accurate interpretation of physical examination data. A technical solution that can solve the above problems is needed. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and propose an Internet-based health checkup report data analysis and management system.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: an Internet-based health checkup report data analysis and management system, comprising:

[0007] The report parsing and extraction module receives the original document of the user's physical examination report, performs format parsing and content extraction on the original document of the user's physical examination report, and obtains user physical examination data records containing user identifier, test item items, test item values, test units and reference value ranges;

[0008] The semantic annotation module matches the user's physical examination data record with a pre-built medical knowledge ontology. The medical knowledge ontology defines the medical meaning of all standard test items, the physiological relationship between items, and the logical rules for judging outliers. Based on the medical knowledge ontology, the module performs semantic annotation on each user's physical examination data record to generate an enhanced physical examination data record with semantic tags.

[0009] The anomaly identification module analyzes the values ​​of the test items in the enhanced physical examination data record, compares the values ​​of the test items with their corresponding reference value ranges, identifies abnormal test items that deviate from the reference value ranges according to the anomaly judgment logic rules defined in the medical knowledge ontology, and generates a preliminary anomaly label for each identified abnormal test item.

[0010] The association analysis module, based on the physiological associations between items defined in the medical knowledge ontology, performs association analysis on the abnormal detection items carrying the preliminary abnormality identifier, and finds abnormal item combinations formed by multiple abnormal detection items that are related on the physiological path.

[0011] As a further aspect of the present invention, the original document of the user's physical examination report is parsed and its content extracted to obtain a user's physical examination data record containing a user identifier, test item entries, test item values, test units, and reference value ranges, including:

[0012] Identify the file format of the user's physical examination report original document, and call the document parser corresponding to the file format to read the document content;

[0013] Using a pre-trained named entity recognition model, text fragments belonging to the detection item name, detection value, and detection unit are identified from the read document content;

[0014] The identified text fragments of the detection item name are fuzzily matched with a pre-set standard detection item name dictionary, and the text fragments of the detection item name are normalized to standard detection item names.

[0015] Establish the data structure of the user physical examination data record, and fill the corresponding fields of the user physical examination data record with the normalized standard test item name, the corresponding test value text fragment, the test unit text fragment, and the user identifier extracted from the document metadata or a specified location;

[0016] Logical validation is performed on the populated user physical examination data records to check the completeness of required fields and the validity of numerical fields. Records that pass the validation are stored in a structured set of user physical examination data records.

[0017] As a further aspect of the present invention, semantic annotation is performed on each user physical examination data record based on the aforementioned medical knowledge ontology to generate an enhanced physical examination data record with semantic tags, including:

[0018] Query the ontology nodes corresponding to the standard test item names in the structured user physical examination data records from the medical knowledge ontology base;

[0019] Obtain all attributes defined on the ontology node, including the physiological system classification to which the detection item belongs, the main clinical significance, and the health dimension labels that it affects;

[0020] The obtained physiological system classification, clinical significance, and health dimension labels are attached as semantic labels to the corresponding user physical examination data records.

[0021] Based on the item hierarchy defined in the medical knowledge ontology, the parent or child items of the current detection item are searched, and the existing hierarchy information is also attached as associated semantic tags.

[0022] User health check data records with all semantic tags attached are marked as the enhanced health check data records.

[0023] As a further aspect of the present invention, based on the outlier determination logic rules defined in the medical knowledge ontology, abnormal detection items deviating from the reference value range are identified, including:

[0024] Read the test item values ​​and their corresponding reference ranges from the enhanced physical examination data record. The reference ranges include the lower limit of normal values ​​and the upper limit of normal values.

[0025] The values ​​of the tested items are compared with the lower limit and upper limit of the normal value, respectively.

[0026] If the value of the detected item is less than the lower limit of the normal value, it is determined to be abnormal and below the reference range, and the abnormal value judgment logic rule below the reference range is triggered.

[0027] If the value of the detected item is greater than the upper limit of the normal value, it is determined to be abnormal and the abnormal value judgment logic rule of being abnormal and above the reference range is triggered.

[0028] For detection items with defined threshold values, the threshold value range is read from the medical knowledge ontology. If the value of the detection item falls into the threshold value range, it is determined to be a critical anomaly, and the anomaly value judgment logic rule of the critical anomaly is triggered.

[0029] For each judgment result, a corresponding preliminary anomaly identifier is generated, which includes the anomaly type and the degree of deviation.

[0030] As a further aspect of the present invention, correlation analysis is performed on the abnormal detection items carrying the preliminary abnormality identifier to find abnormal item combinations formed by multiple abnormal detection items that are related along the physiological pathway, including:

[0031] Starting with each abnormal detection item carrying the preliminary abnormality identifier, query the medical knowledge ontology for other detection items directly associated with it on the physiological pathway;

[0032] Verify whether other directly related test items also exist in the current user's enhanced physical examination data record, and whether they also carry the aforementioned preliminary abnormality indicator;

[0033] If there are two or more detection items that are directly related in the physiological pathway and both carry the preliminary abnormality identifier, the detection items will be grouped into a candidate abnormality item combination.

[0034] The path depth of the candidate abnormal item combination is expanded, and the indirect related items of the items in the combination on the physiological path are further queried. The abnormal status of the indirect related items is checked, and the indirect related items that meet the conditions are included in the candidate abnormal item combination to form a complete physiological path abnormal chain.

[0035] Assign a globally unique associated anomaly code to each complete physiological path anomaly chain, and bind the associated anomaly code to all anomaly detection items within the chain.

[0036] As a further aspect of the present invention, it also includes the step of generating a personalized risk trend map based on historical health data:

[0037] The enhanced physical examination data records under the same user identifier are obtained through the Internet interface and sorted by the examination time.

[0038] For the key testing items selected by the user, the test values ​​and testing times of each test item are extracted from the sorted historical enhanced physical examination data records to form the time series values ​​of the test items.

[0039] The time series values ​​are smoothed and trend-fitted to calculate the slope and acceleration of the changes in the values ​​of the detected items over time.

[0040] By combining the clinical progress knowledge of the test items in the medical knowledge ontology, the slope and acceleration of the change are interpreted, and the level that the test item values ​​will reach at a specific point in the future is predicted;

[0041] The predicted level of the selected project is compared with the reference range of the selected project. Projects that will enter the abnormal range in the future are marked. The prediction and marking results of all selected projects are integrated to generate the individualized risk trend map.

[0042] As a further aspect of the present invention, the time series values ​​are subjected to smoothing filtering and trend fitting processing, and the slope and acceleration of the change of the detected item values ​​over time are calculated, including:

[0043] Outliers in the time series data are identified and corrected, and the corrected time series data are smoothed using a moving average algorithm to obtain a denoised time series.

[0044] Using detection time as the independent variable and the value of the detected items in the denoised time series as the dependent variable, a multinomial regression model is used to fit the data to obtain the fitting curve equation describing the change of the value over time.

[0045] The first derivative of the fitted curve equation is obtained to get the slope of change, which represents the rate of change of the detected item value per unit time.

[0046] The second derivative of the fitted curve equation is obtained to obtain the acceleration, which characterizes the rate of change of the measured value.

[0047] Record the value of the current time point on the fitted curve, the slope of change, and the acceleration as the quantitative output of the trend analysis.

[0048] As a further aspect of the present invention, the method also includes the step of generating a structured health summary report and a list of intervention recommendations:

[0049] Summarize all enhanced physical examination data records of the current user, the identified abnormal detection items and their preliminary abnormality indicators, and all combinations of identified abnormal items and their associated abnormality codes;

[0050] Based on the physiological system classification defined in the medical knowledge ontology, the summarized information is organized to form a draft of a structured health summary report divided into chapters according to physiological systems;

[0051] For each abnormal detection item or combination of abnormal items listed in the structured health summary report draft, a matching intervention suggestion item is retrieved from a pre-set health intervention knowledge graph, which stores suggestions at different levels, from abnormality detection to lifestyle, follow-up examination, and medical treatment.

[0052] The retrieved intervention recommendations are sorted and categorized according to preset priorities and classification rules to generate an intervention recommendation list corresponding to the content of the structured health summary report draft.

[0053] The final version of the structured health summary report is associated and packaged with the intervention recommendation list to form an outputtable health management document.

[0054] As a further aspect of the present invention, the step of retrieving matching intervention suggestion entries from a pre-set health intervention knowledge graph includes:

[0055] The standard detection item name, abnormality type, and physiological system classification of the abnormality detection items are used as the first set of query keywords and input into the health intervention knowledge graph.

[0056] The abnormal physiological path meaning represented by the associated abnormal code of the abnormal item combination is used as the second set of query keywords and input into the health intervention knowledge graph.

[0057] In the health intervention knowledge graph, a graph traversal search based on the first set of query keywords and the second set of query keywords is performed in parallel to find all the intervention measure nodes connected to it.

[0058] The identified intervention nodes are deduplicated and merged, and the attribute information of each intervention node is extracted. The attribute information includes the specific content of the suggestion, the type of suggestion, the level of evidence, and the characteristics of the applicable population.

[0059] Based on the degree of deviation of the anomaly detection items, the user's demographic information, and the evidence level of the intervention measure nodes, the extracted intervention measure nodes are filtered and sorted to form the list of intervention suggestion items.

[0060] As a further aspect of the present invention, the step of establishing an anonymous group data analysis model is also included:

[0061] After obtaining user authorization, user identifiers that can directly identify individuals are extracted from the enhanced physical examination data records of all users to generate an anonymous group physical examination dataset;

[0062] The anonymous group physical examination dataset is grouped according to preset demographic dimensions, including age range, gender, and region.

[0063] Within each group, statistical distribution analysis was performed on the values ​​of each test item to calculate the mean, standard deviation, percentiles, and abnormal detection rate.

[0064] The statistical distribution of each test item in the current group is compared with the statistical distribution of the standard reference population. The standardized difference value is calculated to identify the test items that differ between the group populations.

[0065] Based on the differentiated detection items and their associated physiological systems, and combined with the medical knowledge ontology, a population health analysis brief describing the health characteristics of grouped populations is generated.

[0066] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0067] The extracted raw physical examination data records, containing user identifiers, test item entries, test item values, test units, and reference ranges, are matched against a pre-built medical knowledge ontology. This ontology explicitly defines the medical meaning of all standard test items, the physiological relationships between items, and the logical rules for outlier determination. Based on this ontology, each user's physical examination data record is semantically annotated, generating enhanced physical examination data records with semantic tags. This technology endows isolated physical examination data entries with clear medical semantic attributes, reflecting the physiological relationships between test items. It overcomes the limitations of conventional technologies where physical examination data consists only of surface values ​​and item names, lacking medical semantic support. It transforms physical examination data from simple raw data into interpretable, related structured data, eliminating semantic barriers in the data interpretation process.

[0068] This technology analyzes the numerical values ​​of test items in enhanced physical examination data records with semantic tags, comparing these values ​​with their corresponding reference ranges. Based on the outlier judgment logic rules defined in a medical knowledge ontology, it identifies abnormal test items that deviate from the reference range and generates a preliminary anomaly label for each identified abnormal test item. Then, based on the physiological relationships between items defined in the medical knowledge ontology, it performs correlation analysis on the abnormal test items carrying preliminary anomaly labels, seeking combinations of abnormal items formed by multiple abnormal test items related along physiological pathways. This technology overcomes the limitations of conventional techniques that can only identify single abnormal items and cannot perform correlation analysis. By exploring the intrinsic relationships between abnormal items from a physiological mechanism perspective, it discovers hidden health risks behind single abnormal items, which are composed of multiple related abnormalities. This makes anomaly analysis more aligned with human physiological patterns, avoiding potential health problems missed by isolated anomaly judgments in conventional techniques, and making the anomaly analysis of physical examination data more systematic and comprehensive. Attached Figure Description

[0069] Figure 1 This is a sequence diagram of the Internet-based health checkup report data analysis and management system described in this invention;

[0070] Figure 2This is a flowchart for parsing and extracting content from the original user medical examination report document;

[0071] Figure 3 A flowchart for generating anomaly item combinations through correlation analysis of anomaly detection items;

[0072] Figure 4 A time-varying trend graph of liver function indicators;

[0073] Figure 5 Trend charts showing the abnormality rates of the four major systems in different age groups. Detailed Implementation

[0074] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0075] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0076] See Figure 1 Through multi-layered data processing and knowledge fusion, unstructured physical examination reports are transformed into health information with deep semantic connections and clinical insights. The system receives the original physical examination report document uploaded by the user and processes it through the report parsing and extraction module, transforming it into a structured user physical examination data record containing user identifiers, standard test item entries, test item values, test units, and reference ranges. The semantic annotation module matches these records with a pre-built medical knowledge ontology, adding semantic tags such as the physiological system and clinical significance to each test item, generating enhanced physical examination data records. The anomaly identification module compares the test values ​​with the reference ranges according to the logical rules defined in the medical knowledge ontology, identifying abnormal test items and generating preliminary anomaly markers. Based on this, the association analysis module uses the physiological associations defined in the medical knowledge ontology to perform association analysis on multiple abnormal test items, searching for combinations of abnormal items related to physiological pathways, thereby revealing deeper clues to health problems. In one embodiment of the invention, the report parsing and extraction module receives the original document of the user's physical examination report, referencing... Figure 2The system identifies the file format of the user's original medical examination report and calls the corresponding document parser to read the document content. Using a pre-trained named entity recognition model, it identifies text fragments belonging to the detection item name, detection value, and detection unit from the read document content. The named entity recognition model is built on a bidirectional long short-term memory network and conditional random field architecture to accurately capture entity boundaries and categories in the text. The identified detection item name text fragments are then fuzzily matched with a pre-set standard detection item name dictionary. The fuzzy matching process calculates the similarity between the detection item name text fragment and each entry in the standard detection item name dictionary, and the best match is determined based on the similarity score. The similarity calculation formula is expressed as:

[0077]

[0078] in: This represents the similarity value between the text fragment T containing the name of the detected item and the entry D in the standard dictionary of detected item names. Let T represent the longest common subsequence of text segment T and word D. Indicates the length of text segment T. This indicates the length of term D. Matches with similarity values ​​exceeding a preset threshold are accepted. The text fragment of the detected item name is normalized to the corresponding standard detected item name. A data structure for user health checkup data records is established, including a user identifier field, a standard detected item name field, a detected value field, a detected unit field, and a reference value range field. The normalized standard detected item name, the corresponding detected value text fragment, the detected unit text fragment, and the user identifier extracted from document metadata or a specified location are filled into the corresponding fields of the user health checkup data records. Logical validation is performed on the filled user health checkup data records. The logical validation checks that the user identifier field is not empty, the detected value field is in a valid numeric format, and the detected unit field is consistent with the predefined unit of the standard detected item name. Records that pass the validation are stored in a structured user health checkup data record set.

[0079] For heterogeneous physical examination reports from different medical institutions, the system employs multi-constraint matching logic based on text similarity and unit standard consistency during normalization mapping. After obtaining the item name text using a named entity recognition model, the system not only calculates its edit distance to standard terms in the medical knowledge ontology, but also calls logical verification rules to check whether the extracted unit of the original item conforms to the preset standard unit standard of the corresponding ontology node. When the name similarity reaches the judgment threshold and the unit verification logic matches, the system maps the original record to the corresponding standard item code, thereby effectively solving the semantic conflict problem caused by differences or abbreviations in institution naming, and ensuring the standardization and comparability of physical examination data in subsequent analysis processes.

[0080] In some embodiments, the semantic annotation module queries the medical knowledge ontology base for the ontology nodes corresponding to the standard test item names in the structured user physical examination data records. The medical knowledge ontology base is stored in a graph structure, where nodes represent test items and edges represent relationships between items. It retrieves all attributes defined on the ontology nodes, including the physiological system classification to which the test item belongs, its main clinical significance, and the health dimension tags it affects. Physiological system classifications include cardiovascular and metabolic systems; clinical significance includes inflammatory markers and renal function indicators; and health dimension tags include blood lipid health and liver function health. The retrieved physiological system classification, clinical significance, and health dimension tags are appended as semantic tags to the corresponding user physical examination data records. The appending process involves adding a semantic tag field to the user physical examination data records and writing the corresponding value. Based on the item hierarchy defined in the medical knowledge ontology base, it searches for the parent or child items for the current test item. For example, if the standard test item name is "low-density lipoprotein cholesterol," it searches for the parent item "blood lipid indicators," and appends the existing hierarchy information as associated semantic tags. The appending method involves adding a hierarchy relationship field to the user physical examination data records to record the parent or child item identifier. User health check data records with all semantic tags attached are marked as enhanced health check data records, which are used for in-depth analysis in subsequent modules.

[0081] In this system, the medical knowledge ontology adopts a graph-structured storage architecture. Each node in the ontology represents a standard test item and encapsulates its globally unique item code, standard name string, synonym set, and standardized unit identifier. Nodes are linked through various types of semantic edges, including hierarchical edges representing the subordinate relationships of test items, path-related edges representing the linkage of physiological functions, and joint judgment edges representing the co-occurrence logic of multiple abnormalities. This structured storage method provides a multi-dimensional reference benchmark for the normalization matching of heterogeneous physical examination data, enabling computers to automatically align complex medical meanings based on the graph logic of the ontology.

[0082] Optionally, the named entity recognition model is trained using a labeled dataset of physical examination report texts, containing document samples in various formats to ensure the model's generalization ability. It is understood that the dictionary of standard test item names in fuzzy matching is continuously updated to cover variations of test items added by medical institutions. In specific implementations, logical verification also includes a logical consistency check between the test values ​​and reference ranges; for example, the test values ​​should not exceed medically reasonable extreme values. Optionally, the semantic tag appending process can process multiple user physical examination data records in parallel to improve system throughput. In some embodiments, the construction of the medical knowledge ontology is based on publicly available medical terminology standards and clinical guidelines to ensure the authority of attribute definitions. It is understood that the appending of hierarchical relationship information supports multi-level inheritance, allowing the tracing of the complete classification path of test items in enhanced physical examination data records.

[0083] In one embodiment of the present invention, the anomaly identification module reads the values ​​of test items and their corresponding reference value ranges from the enhanced physical examination data records. The reference value range includes a lower limit and an upper limit of normal values. The test item values ​​are compared with the lower limit and upper limit of normal values ​​respectively, and the comparison operation generates a logical judgment result of greater than, less than, or equal to. If the test item value is less than the lower limit of normal values, it is determined to be below the reference range anomaly, and the anomaly value judgment logic rule for below the reference range anomaly is triggered. The anomaly value judgment logic rule for below the reference range anomaly includes recording and marking the "too low" state. If the test item value is greater than the upper limit of normal values, it is determined to be above the reference range anomaly, and the anomaly value judgment logic rule for above the reference range anomaly is triggered. The anomaly value judgment logic rule for above the reference range anomaly includes recording and marking the "too high" state. For test items with defined threshold values, the threshold value interval is read from the medical knowledge ontology. The medical knowledge ontology stores independent lower and upper limits of threshold values ​​for specific items. If the test item value falls into the threshold value interval, it is determined to be a critical anomaly, and the anomaly value judgment logic rule for critical anomalies is triggered. For each judgment result, a corresponding preliminary anomaly label is generated. The preliminary anomaly label includes the anomaly type and the degree of deviation. The degree of deviation is quantified by calculating the percentage of the relative distance between the detected item value and the reference value boundary. The specific calculation can be expressed as follows:

[0084]

[0085] in: Indicates the percentage of standardization deviation. This indicates the numerical value of the test item. This indicates the boundary value (upper limit or lower limit of normal value) within the reference range. Based on the percentage of deviation, the abnormality can be classified as "mild", "moderate" or "severe".

[0086] In some embodiments, see Figure 3 The association analysis module starts with each abnormal detection item carrying a preliminary abnormality marker. It queries the medical knowledge ontology for other detection items directly associated with it along the physiological pathway. The medical knowledge ontology defines direct physiological relationships between items through edges of types such as "causal association," "metabolic association," and "regulatory association." It then verifies whether other directly associated detection items also exist in the current user's enhanced physical examination data record and whether they also carry preliminary abnormality markers. This verification process is completed by comparing the detection item name with the abnormality marker status. If two or more detection items are directly associated along the physiological pathway and both carry preliminary abnormality markers, these detection items are grouped into a candidate abnormal item combination. For example, abnormal fasting blood glucose and abnormal glycated hemoglobin are grouped together because they both belong to the glucose metabolism pathway. The path depth of candidate abnormal item combinations is expanded to continue querying indirectly related items along the physiological pathway. Indirectly related items refer to detection items connected through one or more intermediate items. The abnormal status of these indirectly related items is verified, and those that meet the criteria are included in the candidate abnormal item combinations, forming a complete physiological pathway abnormal chain. For example, extending from fasting blood glucose abnormality to urinary microalbumin abnormality forms an abnormal chain of "abnormal blood glucose - early kidney damage." A globally unique associated abnormal code is assigned to each complete physiological pathway abnormal chain. This associated abnormal code is generated using a specific prefix plus a sequence number, and is bound to all abnormal detection items within the chain. The binding relationship is stored in the system's association mapping table.

[0087] The association analysis module executes a graph-constrained path traversal algorithm. The system uses nodes carrying preliminary anomaly markers as the starting point for the search, performing constrained expansion within the physiological association graph of the medical knowledge ontology. The search logic is limited by a pre-defined whitelist of relationship types and maximum path depth constraints to ensure that identified abnormal chains have clear physiological significance. During the search process, the system continuously checks the status of other test items involved in the associated path within the current enhanced physical examination data record. Only when multiple mutually corroborating nodes exist on the path and are all within the abnormal range is it determined to be a physiological abnormal chain. Subsequently, the system assigns a globally unique association anomaly code to the complete path, achieving a technological leap from isolated anomaly detection to systematic risk mining.

[0088] Optionally, the outlier determination logic rules are configurable, allowing for differentiated reference value ranges and critical value intervals based on different genders and age groups. It can be understood that the priority for determining critical anomalies is higher than normal but lower than clearly defined out-of-reference-range anomalies. In specific implementations, the formation process of the physiological pathway anomaly chain allows setting a maximum path depth to prevent infinite expansion; for example, the maximum depth can be set to 3. Optionally, the associated anomaly code is not only bound to the anomaly detection item but also associated with the user identifier that generated the anomaly chain to support user-based queries. In some embodiments, for a collection of candidate anomaly item combinations, the system calculates the anomaly directionality of the items within the combination; for example, all items are "high" or all are "low". It can be understood that path depth expansion is a recursive process until no new indirect associated items carrying the initial anomaly identifier are found or the preset depth is reached.

[0089] In one embodiment of the present invention, historical enhanced physical examination data records under the same user identifier are obtained through an internet interface, sorted by examination time, and a record list is generated in ascending order of timestamps. For the key examination items selected by the user, the values ​​and examination times of each examination item are extracted from the sorted historical enhanced physical examination data records to form a time series value for the examination item. The time series value is represented as a set of value pairs arranged in chronological order. Smoothing filtering and trend fitting are performed on the time series value, and the slope and acceleration of the change in the examination item value over time are calculated. Combined with clinical progress knowledge of the examination item in a medical knowledge ontology, the slope and acceleration are interpreted to predict the level that the examination item value will reach at a specific future time point. The predicted level of the selected item is compared with the reference range of the selected item. Items that will enter the abnormal range in the future are marked. The prediction and marking results of all selected items are integrated to generate an individualized risk trend map. The individualized risk trend map displays the historical trajectory, fitted curve, and future prediction point of each item in a visual chart format.

[0090] In some embodiments, the process of smoothing and trend fitting time series values, and calculating the slope and acceleration of the changes in the detected item values ​​over time, involves identifying and correcting outliers in the time series values, and then smoothing the corrected time series values ​​using a moving average algorithm to obtain a denoised time series. The window size of the moving average is adaptively determined based on the density of the time series data points. Using the detection time as the independent variable and the detected item values ​​in the denoised time series as the dependent variable, a multinomial regression model is used to fit the data to obtain a fitting curve equation describing the changes in values ​​over time. The general form of the multinomial regression model is:

[0091]

[0092] in: This indicates the numerical value of the test item. Indicates the detection time. These represent the coefficients fitted by the polynomial regression model. The order of the multinomial regression model is determined using the least squares method. The slope of change is obtained by taking the first derivative of the fitted curve equation. slope of change The rate of change of the measured value per unit time is used to characterize the acceleration. The second derivative of the fitted curve equation is then taken to obtain the acceleration. acceleration Characterizes the rate of change of the measured values. Records the values ​​and slopes of the fitted curve at the current time point. and acceleration This serves as a quantitative output for trend analysis.

[0093] Optionally, outlier identification is based on statistical methods, such as classifying data points that deviate from the moving median by more than three times the absolute deviation as outliers and then correcting for them. This is understandable given the order of the multinomial regression model. Selection is achieved through cross-validation to avoid overfitting or underfitting. In practice, clinical progress knowledge includes the clinical significance of changes in test item values; for example, an accelerated rate of increase in serum creatinine may indicate a faster progression of renal function decline. Optionally, predictions of specific future time points are made by considering future time... Substitute into the fitted curve equation Calculated. In some embodiments, the individualized risk trend map distinguishes different detection items using curves of different colors, and highlights or marks items predicted to enter the abnormal range on the chart. It can be understood that the slope of change... With acceleration The symbols together indicate the direction and curvature of the trend, with a positive slope indicating a change in trend. Combined with positive acceleration This indicates an accelerating upward trend.

[0094] In one embodiment of the present invention, all enhanced physical examination data records of the current user, identified abnormal detection items and their preliminary abnormality identifiers, and all identified abnormal item combinations and their associated abnormality codes are aggregated to generate an intermediate aggregate dataset containing all relevant data. According to the physiological system classification defined in the medical knowledge ontology, the information in the intermediate aggregate dataset is organized to form a structured health summary report draft divided into chapters by physiological system. For example, "Cardiovascular System," "Metabolic and Endocrine System," and "Liver Function" each become independent chapters, and each chapter lists the abnormal detection items and abnormal item combinations belonging to that system. For each abnormal detection item or abnormal item combination listed in the structured health summary report draft, matching intervention suggestion entries are retrieved from a pre-set health intervention knowledge graph. The health intervention knowledge graph is stored in a graph structure, with nodes including abnormal entity nodes and intervention measure nodes, and edges representing the applicability relationship between them. The retrieved intervention suggestion entries are sorted and categorized according to preset priority and classification rules to generate an intervention suggestion list corresponding to the content of the structured health summary report draft. The entries in the intervention suggestion list are categorized into categories such as "Lifestyle Adjustment," "Medical Follow-up Suggestions," and "Medical Treatment Guidance." The final structured health summary report and the list of intervention recommendations are linked and packaged to form an output health management document. The linking and packaging method can be to package the two documents into a compressed file or to generate an interactive document with internal links.

[0095] The system retrieves matching intervention recommendations from a pre-built health intervention knowledge graph. This involves inputting the standard detection item name, abnormality type, and associated physiological system classification of the abnormality detection item as the first set of query keywords. The second set of query keywords is the meaning of the physiological pathway abnormality represented by the associated abnormality code of the abnormal item combination. A graph traversal search based on the first and second sets of query keywords is performed in parallel within the health intervention knowledge graph to find all connected intervention measure nodes. The graph traversal search starts from the entity node representing the query keyword and searches for directly connected intervention measure nodes along edges of the "applicable recommendation" type. The found intervention measure nodes are deduplicated and merged, and the attribute information of each intervention measure node is extracted. The attribute information includes the specific content of the recommendation, the recommendation type, the level of evidence, and the characteristics of the applicable population. Based on the deviation degree of the abnormality detection item, the user's demographic information, and the level of evidence of the intervention measure node, the extracted intervention measure nodes are filtered and sorted to form a list of intervention recommendation items. The sorting process calculates a ranking score for each node.

[0096]

[0097] in: This indicates the ranking score of the intervention measure node. Numerical representation of evidence levels, for example: Level I evidence is 3, Level II is 2, and Level III is 1. Numerical representation of the degree of deviation of anomaly detection items, for example: severe deviation is 0.3, moderate deviation is 0.2, and slight deviation is 0.1. The final list is sorted by score. Sort in descending order.

[0098] In some embodiments, the attribute information of intervention measure nodes in the health intervention knowledge graph is stored in structured fields, see Table 1.

[0099] Table 1: Node Attribute Table of Intervention Measures

[0100] Node ID Suggested specific content Recommendation type Level of evidence Target audience characteristics IN001 Engage in at least 150 minutes of moderate-intensity aerobic exercise per week. Lifestyle Adjustment II In general adults without cardiopulmonary dysfunction IN002 It is recommended to have fasting blood glucose and glycated hemoglobin rechecked within 3 months. Medical follow-up recommendations II Those with borderline or slightly elevated fasting blood glucose IN003 It is recommended to visit an endocrinology clinic to assess the risk of diabetes. Medical guidance I Those with abnormal fasting blood glucose and glycated hemoglobin

[0101] Optionally, user demographic information, including age, gender, and past medical history tags, is used to filter intervention nodes that do not match the characteristics of the applicable population. It is understood that the numerical representation rules for evidence levels can be adjusted based on updates to clinical guidelines. In specific implementation, the meaning of physiological pathway abnormalities is obtained from the medical knowledge ontology through association anomaly codes; for example, the association anomaly code "AM001" corresponds to "abnormal glucose metabolism - early kidney injury association." Optionally, the generation of the intervention recommendation list supports personalized configuration, allowing users to choose to focus on "lifestyle" or "medical" recommendations, and the system pre-filters accordingly before sorting. In some embodiments, the generation of a structured health summary report draft uses a report template engine to populate the organized information into preset chapter templates. It is understood that the health management document formed by association encapsulation can be output as PDF, HTML, or a format conforming to specific medical data exchange standards.

[0102] See Figure 4This is a time-varying trend chart of liver function indicators, showing the changes in three key liver function indicators for the user from June 2021 to June 2024 during the risk trend analysis phase. Alanine aminotransferase (ALT) showed an overall upward trend, rising from 45 U / L in June 2021 to 55 U / L in June 2024, reaching 52 U / L in June 2022, and briefly falling back to 48 U / L in June 2023. Aspartate aminotransferase (AST) also showed an upward trend, rising from 38 U / L in June 2021 to 45 U / L in June 2024, with a slight decrease to 40 U / L in June 2023. Bilirubin fluctuated less, generally fluctuating slightly within the range of 15-19 μmol / L, reaching 19 μmol / L in June 2024. The persistent abnormality of the three indicators, especially the simultaneous elevation of ALT and AST, suggests that the user's liver health is at risk of deterioration, and further clinical investigation is needed to rule out causes such as fatty liver, viral hepatitis, or drug-induced liver injury.

[0103] In one embodiment of the present invention, after obtaining user authorization, user identifiers that can directly identify an individual are extracted from the enhanced physical examination data records of all users. These user identifiers include name, ID number, and phone number, generating an anonymous group physical examination dataset. The anonymization process employs irreversible hash replacement or deletion operations. A strong hash algorithm is used to perform irreversible mapping transformations on sensitive fields involving user identity, replacing the original identifiers with anonymized hash labels. For quasi-identifiers such as region and age, the system performs attribute generalization processing according to the privacy classification strategy preset in the medical knowledge ontology. For example, precise geographical coordinates are generalized to administrative division codes, and specific birth dates are converted into corresponding age range identifiers, thereby eliminating the risk of individual identity re-identification caused by the combination of multiple quasi-identifiers. The anonymous group physical examination dataset is grouped according to preset demographic dimensions, including age range, gender, and region. Age ranges are divided in 10-year increments, and regions are divided by provincial-level administrative regions. Each specific combination of age range, gender, and region constitutes an independent group. Within each group, statistical distribution analysis was performed on the values ​​of each test item to calculate the mean, standard deviation, percentiles, and abnormal detection rate. The abnormal detection rate was calculated based on the reference range defined in the medical knowledge ontology, using the following formula:

[0104]

[0105] in: This indicates the abnormal detection rate of a certain test item within the current group. This indicates the number of records within this group whose values ​​for this detection item were determined to be abnormal. This indicates the total number of records containing this test item within the group. For the numerical fields of the test items, the system introduces a numerical discretization mapping mechanism. Especially for extreme test values ​​that exceed the normal physiological range, the system maps them to a preset numerical range based on the health feature profiles defined in the ontology, rather than retaining the original high-precision measurement values. This range-based processing effectively prevents malicious queryers from tracing back to specific users by searching for extreme outliers, ensuring that the group dataset retains medical statistical characteristics while obscuring the precise identification of individual values. The statistical distribution of each test item within the current group is compared with the statistical distribution of the standard reference population, and the standardized difference value is calculated. The statistical distribution data of the standard reference population comes from authoritative health statistics yearbooks published by the state or industry. The standardized difference value is:

[0106]

[0107] in: Indicates standardized difference value, This represents the mean of the current group of samples. This represents the mean of the standard reference population. This represents the standard deviation of the current group of samples. This represents the standard deviation of the reference population. It identifies test items that show differences between grouped populations; the identification criterion is the standardized difference value. The absolute value of the value is greater than a preset threshold. Before outputting the statistical distribution analysis results, the system performs k-anonymity logic verification and differential privacy noise processing. By calculating the number of records in each group, it checks whether the preset safety threshold is met. If the sample size in a specific group (such as an elderly group in a specific region) is too small, the system will automatically trigger dimension merging logic to merge adjacent group intervals until the indistinguishability requirement is met. In addition, when calculating aggregate indicators such as mean and standard deviation, the system introduces a small amount of Laplace noise as needed to ensure that the content of the statistical summary does not reveal any individual's physiological data details. Based on the differentially detected items and their associated physiological systems, combined with a medical knowledge ontology, a group health analysis summary describing the health characteristics of the group population is generated. The group health analysis summary is presented in text and chart form, indicating which physiological system indicators of the group significantly deviate from the standard population.

[0108] In some embodiments, the anonymization process ensures that the generated dataset cannot be used to infer an individual's identity from the remaining fields; for example, the processed records retain only the age range, gender, region code, and detection value. It is understood that grouping operations can be multi-level nested, for example, first grouping by region, and then further subdividing by age and gender within each region group. In specific implementations, percentile calculations include the 5th, 25th, 50th, 75th, and 95th percentiles, used to describe the data distribution pattern. Optionally, standardized variance values... The preset threshold can be set according to the required analytical precision, for example, setting the threshold to 0.2. Differences are considered to exist at that time. In some embodiments, the identification results of the differentially identified test items are mapped to physiological systems defined in the medical knowledge ontology. For example, if differences are identified in fasting blood glucose, glycated hemoglobin, and triglycerides, they are mapped to the "metabolic system." It can be understood that the content generation of the population health analysis briefing uses a template-filling method, which includes the identified differential items, associated physiological systems, and standardized difference values. Statistical descriptions are automatically filled into a preset analysis report template. Optionally, the anonymized population data analysis model supports periodic incremental updates; when new authorized enhanced physical examination data records are added, the anonymized population physical examination dataset is automatically updated and reanalyzed.

[0109] See Figure 5 This is a trend chart of abnormality rates in the four major systems across different age groups. It shows the changing trends in the detection rates of abnormalities in the metabolic, cardiovascular, digestive, and respiratory systems during the population data analysis phase, and can be used for population health profile analysis. The metabolic system has the highest abnormality rate among all age groups, and it increases significantly linearly with age, rising from 15% in the 20-29 age group to 65% in the 60-69 age group, making it the system with the largest increase. The cardiovascular system's abnormality rate follows closely behind the metabolic system, rising from 10% in the 20-29 age group to 60% in the 60-69 age group, with a particularly significant increase after age 50. The digestive system's abnormality rate has a relatively slow overall growth rate, increasing from 8% in the 20-29 age group to 45% in the 60-69 age group, consistently lower than that of the metabolic and cardiovascular systems. The respiratory system has the lowest abnormality rate among all systems, increasing from 5% in the 20-29 age group to 35% in the 60-69 age group, with a relatively moderate increase.

[0110] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A health checkup report data analysis and management system based on the Internet, characterized in that, The system includes: The report parsing and extraction module receives the original document of the user's physical examination report, performs format parsing and content extraction on the original document of the user's physical examination report, and obtains user physical examination data records containing user identifier, test item items, test item values, test units and reference value ranges; The semantic annotation module matches the user's physical examination data record with a pre-built medical knowledge ontology. The medical knowledge ontology defines the medical meaning of all standard test items, the physiological relationship between items, and the logical rules for judging outliers. Based on the medical knowledge ontology, the module performs semantic annotation on each user's physical examination data record to generate an enhanced physical examination data record with semantic tags. The anomaly identification module analyzes the values ​​of the test items in the enhanced physical examination data record, compares the values ​​of the test items with their corresponding reference value ranges, identifies abnormal test items that deviate from the reference value ranges according to the anomaly judgment logic rules defined in the medical knowledge ontology, and generates a preliminary anomaly label for each identified abnormal test item. The association analysis module, based on the physiological associations between items defined in the medical knowledge ontology, performs association analysis on the abnormal detection items carrying the preliminary abnormality identifier, and finds abnormal item combinations formed by multiple abnormal detection items that are related on the physiological path.

2. The Internet-based health checkup report data analysis and management system according to claim 1, characterized in that, The original user health check report document is parsed and its content extracted to obtain user health check data records containing user identifiers, test item entries, test item values, test units, and reference ranges, including: Identify the file format of the user's physical examination report original document, and call the document parser corresponding to the file format to read the document content; Using a pre-trained named entity recognition model, text fragments belonging to the detection item name, detection value, and detection unit are identified from the read document content; The identified text fragments of the detection item name are fuzzily matched with a pre-set standard detection item name dictionary, and the text fragments of the detection item name are normalized to standard detection item names. Establish the data structure of the user physical examination data record, and fill the corresponding fields of the user physical examination data record with the normalized standard test item name, the corresponding test value text fragment, the test unit text fragment, and the user identifier extracted from the document metadata or a specified location; Logical validation is performed on the populated user physical examination data records to check the completeness of required fields and the validity of numerical fields. Records that pass the validation are stored in a structured set of user physical examination data records.

3. The Internet-based health checkup report data analysis and management system according to claim 2, characterized in that, Based on the aforementioned medical knowledge ontology, each user's physical examination data record is semantically annotated to generate an enhanced physical examination data record with semantic tags, including: Query the ontology nodes corresponding to the standard test item names in the structured user physical examination data records from the medical knowledge ontology base; Obtain all attributes defined on the ontology node, including the physiological system classification to which the detection item belongs, the main clinical significance, and the health dimension labels that it affects; The obtained physiological system classification, clinical significance, and health dimension labels are attached as semantic labels to the corresponding user physical examination data records. Based on the item hierarchy defined in the medical knowledge ontology, the parent or child items of the current detection item are searched, and the existing hierarchy information is also attached as associated semantic tags. User health check data records with all semantic tags attached are marked as the enhanced health check data records.

4. The Internet-based health checkup report data analysis and management system according to claim 3, characterized in that, Based on the outlier determination logic rules defined in the medical knowledge ontology, abnormal detection items that deviate from the reference value range are identified, including: Read the test item values ​​and their corresponding reference ranges from the enhanced physical examination data record. The reference ranges include the lower limit of normal values ​​and the upper limit of normal values. The values ​​of the tested items are compared with the lower limit and upper limit of the normal value, respectively. If the value of the detected item is less than the lower limit of the normal value, it is determined to be abnormal and below the reference range, and the abnormal value judgment logic rule below the reference range is triggered. If the value of the detected item is greater than the upper limit of the normal value, it is determined to be abnormal and the abnormal value judgment logic rule of being abnormal and above the reference range is triggered. For detection items with defined threshold values, the threshold value range is read from the medical knowledge ontology. If the value of the detection item falls into the threshold value range, it is determined to be a critical anomaly, and the anomaly value judgment logic rule of the critical anomaly is triggered. For each judgment result, a corresponding preliminary anomaly identifier is generated, which includes the anomaly type and the degree of deviation.

5. The Internet-based health checkup report data analysis and management system according to claim 4, characterized in that, A correlation analysis is performed on the abnormal detection items carrying the preliminary abnormality markers to identify abnormal item combinations formed by multiple abnormal detection items that are related along the physiological pathway, including: Starting with each abnormal detection item carrying the preliminary abnormality identifier, query the medical knowledge ontology for other detection items directly associated with it on the physiological pathway; Verify whether other directly related test items also exist in the current user's enhanced physical examination data record, and whether they also carry the aforementioned preliminary abnormality indicator; If there are two or more detection items that are directly related in the physiological pathway and both carry the preliminary abnormality identifier, the detection items will be grouped into a candidate abnormality item combination. The path depth of the candidate abnormal item combination is expanded, and the indirect related items of the items in the combination on the physiological path are further queried. The abnormal status of the indirect related items is checked, and the indirect related items that meet the conditions are included in the candidate abnormal item combination to form a complete physiological path abnormal chain. Assign a globally unique associated anomaly code to each complete physiological path anomaly chain, and bind the associated anomaly code to all anomaly detection items within the chain.

6. The Internet-based health checkup report data analysis and management system according to claim 5, characterized in that, It also includes the step of generating individualized risk trend maps based on historical health data: The enhanced physical examination data records under the same user identifier are obtained through the Internet interface and sorted by the examination time. For the key testing items selected by the user, the test values ​​and testing times of each test item are extracted from the sorted historical enhanced physical examination data records to form the time series values ​​of the test items. The time series values ​​are smoothed and trend-fitted to calculate the slope and acceleration of the changes in the values ​​of the detected items over time. By combining the clinical progress knowledge of the test items in the medical knowledge ontology, the slope and acceleration of the change are interpreted, and the level that the test item values ​​will reach at a specific point in the future is predicted; The predicted level of the selected project is compared with the reference range of the selected project. Projects that will enter the abnormal range in the future are marked. The prediction and marking results of all selected projects are integrated to generate the individualized risk trend map.

7. The Internet-based health checkup report data analysis and management system according to claim 6, characterized in that, The time series values ​​are smoothed and trend-fitted to calculate the slope and acceleration of the changes in the detected item values ​​over time, including: Outliers in the time series data are identified and corrected, and the corrected time series data are smoothed using a moving average algorithm to obtain a denoised time series. Using detection time as the independent variable and the value of the detected items in the denoised time series as the dependent variable, a multinomial regression model is used to fit the data to obtain the fitting curve equation describing the change of the value over time. The first derivative of the fitted curve equation is obtained to get the slope of change, which represents the rate of change of the detected item value per unit time. The second derivative of the fitted curve equation is obtained to obtain the acceleration, which characterizes the rate of change of the measured value. Record the value of the current time point on the fitted curve, the slope of change, and the acceleration as the quantitative output of the trend analysis.

8. The Internet-based health checkup report data analysis and management system according to claim 7, characterized in that, It also includes the steps of generating structured health summary reports and a list of intervention recommendations: Summarize all enhanced physical examination data records of the current user, the identified abnormal detection items and their preliminary abnormality indicators, and all combinations of identified abnormal items and their associated abnormality codes; Based on the physiological system classification defined in the medical knowledge ontology, the summarized information is organized to form a draft of a structured health summary report divided into chapters according to physiological systems; For each abnormal detection item or combination of abnormal items listed in the structured health summary report draft, a matching intervention suggestion item is retrieved from a pre-set health intervention knowledge graph, which stores suggestions at different levels, from abnormal detection to lifestyle, follow-up examination, and medical treatment. The retrieved intervention recommendations are sorted and categorized according to preset priorities and classification rules to generate an intervention recommendation list corresponding to the content of the structured health summary report draft. The final version of the structured health summary report is associated and packaged with the intervention recommendation list to form an outputtable health management document.

9. The Internet-based health checkup report data analysis and management system according to claim 8, characterized in that, The step of retrieving matching intervention suggestion entries from a pre-set health intervention knowledge graph includes: The standard detection item name, abnormality type, and physiological system classification of the abnormality detection items are used as the first set of query keywords and input into the health intervention knowledge graph. The abnormal physiological path meaning represented by the associated abnormal code of the abnormal item combination is used as the second set of query keywords and input into the health intervention knowledge graph. In the health intervention knowledge graph, a graph traversal search based on the first set of query keywords and the second set of query keywords is performed in parallel to find all the intervention measure nodes connected to it. The identified intervention nodes are deduplicated and merged, and the attribute information of each intervention node is extracted. The attribute information includes the specific content of the suggestion, the type of suggestion, the level of evidence, and the characteristics of the applicable population. Based on the degree of deviation of the anomaly detection items, the user's demographic information, and the evidence level of the intervention measure nodes, the extracted intervention measure nodes are filtered and sorted to form the list of intervention suggestion items.

10. The Internet-based health checkup report data analysis and management system according to claim 9, characterized in that, It also includes the steps of building an anonymous group data analysis model: After obtaining user authorization, user identifiers that can directly identify individuals are extracted from the enhanced physical examination data records of all users to generate an anonymous group physical examination dataset; The anonymous group physical examination dataset is grouped according to preset demographic dimensions, including age range, gender, and region. Within each group, statistical distribution analysis was performed on the values ​​of each test item to calculate the mean, standard deviation, percentiles, and abnormal detection rate. The statistical distribution of each test item in the current group is compared with the statistical distribution of the standard reference population. The standardized difference value is calculated to identify the test items that differ between the group populations. Based on the differentiated detection items and their associated physiological systems, and combined with the medical knowledge ontology, a population health analysis brief describing the health characteristics of grouped populations is generated.

Images