User health record construction method and system, electronic device and storage medium

Abstract

The application provides a user health record construction method and system, electronic equipment and storage medium, relates to the medical health information processing technical field, and identifies key detection indexes, analysis results, health grades and disease progression trends of users through detection index data sequences and key semantic vector libraries; and further constructs user health records. The application realizes the integration of multi-source heterogeneous health data, and further extracts time sequence change information and context semantic weight of detection indexes, which is beneficial to improve the accuracy of determining user health records. The application differentiates and analyzes detection index data sequences and key semantic vector libraries from multiple dimensions, and further obtains key detection indexes, analysis results of key detection indexes, health grades and disease progression trends, realizes the deep mining of user health data, and improves the comprehensiveness and accuracy of user health records.

Description

Technical Field

[0001] This invention relates to the field of medical and health information processing technology, and in particular to a method, system, electronic device and storage medium for constructing user health records. Background Technology

[0002] With the widespread adoption of Electronic Health Records (EHRs) and the acceleration of healthcare digitization, individuals accumulate a vast amount of healthcare data during their daily medical visits. This data comes in various forms, including but not limited to laboratory reports, imaging reports, and outpatient and inpatient medical records, forming a multi-sourced and heterogeneous personal health record. Providing users with a comprehensive, accurate, and traceable assessment of their health status requires the effective integration and analysis of this complex data spanning multiple time periods and institutions.

[0003] Existing methods for constructing user health records include medical text understanding methods based on pre-trained language models, which utilize natural language processing (NLP) technology to extract information and semantically understand individual medical records or reports. Additionally, there are auxiliary analysis systems based on Large Language Models (LLM), which attempt to directly input users' health report text into the model to obtain analysis and suggestions. Meanwhile, some medical information systems employ search engine-based technologies to achieve keyword retrieval and querying of massive amounts of medical documents. When a patient has dozens or even more health attachments such as test reports and medical records, directly inputting the combined text of all attachments into a large language model for comprehensive analysis faces bottlenecks due to context length limitations and attention decay. On the one hand, the total text length of massive attachments easily exceeds the upper limit of the context window that the large language model can handle; on the other hand, even if the text length is within the window, excessively long input sequences can cause the model's attention mechanism to decay, significantly reducing its ability to capture the beginning or key information of the input content. This severely affects the accuracy and comprehensiveness of the final analysis results, making it impossible to achieve effective and high-quality comprehensive evaluation of massive historical data.

[0004] In conclusion, the accuracy of analyzing user health records based on massive amounts of historical data is low. Summary of the Invention

[0005] This invention provides a method, system, electronic device, and storage medium for constructing user health records, in order to solve the problem of low accuracy in the analysis of user health records based on massive historical data in the prior art, and to improve the accuracy of user health record analysis based on massive historical data.

[0006] This invention provides a method for constructing a user health record, comprising: Perform multi-level structured processing and index construction on at least one type of user health data to obtain the data sequence of user's various detection indicators changing over time; Extract key semantic information from various health data, generate key semantic vectors, and store them in a key semantic vector library; The importance of each detection indicator is determined based on the key semantic vector library in order to identify the key detection indicators in the data sequence of each detection indicator and obtain the parsing results of the key detection indicators. Based on the key semantic vector library and the data sequences of various detection indicators, the user's health level is determined. Based on the magnitude and abnormal patterns of changes in the data sequence of detection indicators, the user's disease progression trend can be determined; Based on key detection indicators, the analysis results of key detection indicators, health level, and disease progression trend, a user health profile is constructed.

[0007] According to the user health record construction method provided by the present invention, at least one type of user health data is subjected to multi-level structured processing and index construction to obtain the user's various detection indicators data sequences changing over time, including: Based on different extraction templates, structured information is extracted from unstructured text of various health data to obtain structured detection indicator data; the different extraction templates include examination report extraction templates and medical record extraction templates; The structured detection index data is split into detection index datasets for each detection index. The names of each detection indicator are aligned and the units are standardized to merge the detection indicator datasets with the same name, thus obtaining the preprocessed detection indicator datasets. Based on the detection time, the data of each detection indicator in the preprocessed detection indicator dataset are sorted by time to obtain the detection indicator data sequence.

[0008] According to the user health profile construction method provided by the present invention, key semantic information is extracted from various health data to generate various key semantic vectors, including: Obtain structured correlation information for each detection indicator; Based on structured association information, key semantic information is extracted from various health data to obtain various key semantic data. Based on the hash value of each key semantic data, duplicate key semantic data are removed to obtain each deduplicated key semantic data. Based on the deduplicated key semantic data, generate each key semantic vector.

[0009] According to the user health profile construction method provided by this invention, the importance of each detection indicator is determined based on a key semantic vector library to identify key detection indicators in each detection indicator data sequence, including: Obtain a set number of key semantic vectors from the key semantic vector library; The importance of each detection indicator is identified based on a set number of key semantic vectors. The scores for each detection indicator are calculated based on the importance of each indicator and the anomalies in the data sequences of each indicator. Based on the ranking of the scores, key detection indicators are identified.

[0010] According to the user health record construction method provided by the present invention, the user's disease progression trend is determined based on the variation amplitude and abnormal variation pattern of the detection indicator data sequence, including: Identify significantly changing detection indicators based on the magnitude of changes in the detection indicator data sequence; To obtain abnormal change patterns in the data sequences of detection indicators; Based on the significant change detection indicators, the abnormal change patterns of the detection indicators, and the deviation values ​​of each detection indicator data in the detection indicator data sequence, user prompt words are generated. Input user prompts into a large language model and obtain the disease progression trend output by the large language model; Among them, the large language model searches medical literature databases based on user-provided prompts and obtains disease progression trends based on the search results.

[0011] According to the user health profile construction method provided by the present invention, based on a key semantic vector library and various detection indicator data sequences, the user's health level is determined, including: Obtain the sequence of abnormal states of each detection indicator data relative to the normal detection indicator range in the detection indicator data sequence; Based on the abnormal state sequence, determine the type of abnormal change in the detection indicators; Based on a set number of key semantic vectors, the reference value of identifying the abnormal change types of each detection indicator; The user's health level is determined based on the type of abnormal change and its reference value.

[0012] According to the user health profile construction method provided by the present invention, the number of key semantic vectors is determined based on the following method: Each key semantic vector is aggregated into multiple key semantic vector groups according to the detection indicators; Within the key semantic vector group, multiple valid key semantic vectors are selected based on the importance score of each key semantic vector. The importance score of the key semantic vector is determined based on the anomaly index weight, timeliness weight, urgency weight, and detection index diversity constraints of the key semantic vector. Based on the effective key semantic vectors of each key semantic vector group, a set number of key semantic vectors are obtained.

[0013] This invention also provides a system for constructing user health records, comprising: a structured medical data analysis subsystem, a health memory management subsystem, a health analysis agent, a health level assessment agent, a disease progression analysis agent, and an analysis output layer. The structured medical data analysis subsystem is used to perform multi-level structured processing and indexing of at least one type of health data of users, and to obtain the data sequence of various test indicators of users changing over time. The health memory management subsystem is used to extract key semantic information from various health data, generate key semantic vectors, and store them in the key semantic vector library. A health analysis intelligent agent is used to determine the importance of each detection indicator based on a key semantic vector library, in order to identify key detection indicators in the data sequence of each detection indicator and obtain the analysis results of key detection indicators. A health level assessment agent is used to determine a user's health level based on a key semantic vector library and data sequences of various detection indicators. The disease progression analysis agent is used to determine the user's disease progression trend based on the magnitude and abnormal change patterns of the detection indicator data sequence. The analysis output layer is used to construct user health profiles based on key detection indicators, the analysis results of key detection indicators, health level, and disease progression trends.

[0014] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any of the above-described methods for constructing a user health record.

[0015] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-described methods for constructing a user health record.

[0016] This invention provides a method, system, electronic device, and storage medium for constructing user health records. It involves multi-level structured processing and indexing of at least one type of user health data to obtain time-varying data sequences of various detection indicators. Key semantic information is extracted from these health data, generating key semantic vectors and storing them in a key semantic vector library. The importance of each detection indicator is determined based on the key semantic vector library to identify key detection indicators in the data sequences and obtain their analytical results. The user's health level is determined based on the key semantic vector library and the data sequences of each detection indicator. The user's disease progression trend is determined based on the magnitude and abnormal change patterns of the detection indicator data sequences. Finally, a user health record is constructed based on the key detection indicators, their analytical results, health level, and disease progression trend. This invention generates detection indicator data sequences and a key semantic vector library from at least one type of health data, integrating multi-source heterogeneous health data and extracting time-series change information and contextual semantic weights of the detection indicators, thus improving the accuracy of determining user health records. This invention performs differentiated analysis on the detection indicator data sequence and key semantic vector library from multiple dimensions, thereby obtaining key detection indicators, the analysis results of key detection indicators, health level and disease progression trend, realizing in-depth mining of user health data and improving the comprehensiveness and accuracy of user health records. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is one of the flowcharts illustrating the method for constructing a user health record provided by the present invention.

[0019] Figure 2 This is the second flowchart illustrating the method for constructing user health records provided by this invention.

[0020] Figure 3 This is the third flowchart illustrating the method for constructing user health records provided by this invention.

[0021] Figure 4 This is a schematic diagram of the change curve constructed based on the detection index data sequence provided by the present invention.

[0022] Figure 5 This is a schematic diagram of the process for generating a user's health profile provided by the present invention.

[0023] Figure 6 This is a schematic diagram of the structure of the user health record construction system provided by the present invention.

[0024] Figure 7 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0026] Most existing user health profiles are built using a single model or a single analytical perspective to process health data, failing to provide a collaborative analysis of patient health status from multiple professional angles (such as health level assessment, key indicator analysis, and disease progression tracking). A single model struggles to balance comprehensive analysis with in-depth professional knowledge across various dimensions, resulting in biased or inaccurate analytical results.

[0027] Existing medical data lacks sufficient structure and its indicators are untraceable. Patient test reports are typically in the form of images, PDFs, or unstructured text, with varying formats across different times and hospitals. Current systems lack the ability to extract and store this heterogeneous data in a unified structure, making it impossible to effectively correlate and compare the values ​​of the same indicator at different points in time, hindering time-series tracking and trend analysis.

[0028] Meanwhile, large language models face context length limitations and attention decay issues when processing a large number of attachments. When users accumulate a large number of health attachments (such as dozens of test reports and medical records), directly inputting the complete content of all attachments into the large language model for analysis will encounter two core problems: first, it exceeds the model's context window length limit; second, even within the context window, excessively long input text will lead to attention decay in the large language model, reducing its ability to capture key information and significantly lowering the quality of analysis.

[0029] Existing health analysis systems often lack explicit citations and traceability of the basis for their analyses when providing conclusions. Users and doctors find it difficult to judge the reliability of these conclusions, and cannot trace which specific test data or medical references a particular conclusion is based on.

[0030] The existing user health profile construction lacks the ability to effectively remember and manage historical health data. The construction of existing user health profiles typically requires reloading and processing all historical data for each analysis, lacking the ability to effectively remember and incrementally update processed data. This results in low system efficiency and makes it difficult to maintain analytical quality as data volume increases.

[0031] This invention provides an intelligent analysis method and system for user health records based on multi-agent collaboration and health memory. The method and system utilize a multi-agent architecture to achieve a comprehensive and evidence-based analysis of the user's health status from multiple professional perspectives. This invention structurally extracts and summarizes the user's past and current test reports, medical records, etc., into a sequence of detection indicator data, enabling the tracking, calculation, and comparison of various detection indicators.

[0032] The method for constructing user health records provided by this invention achieves effective memory and storage of user health data through a health memory management subsystem. It reduces redundant text through aggregation and deduplication, and ensures effective analysis of a large number of attachments while reducing the length of the context.

[0033] The following is combined with Figures 1 to 7 This invention describes a method, apparatus, and electronic device for constructing user health records.

[0034] Figure 1 This is one of the flowcharts illustrating the method for constructing a user health record provided by the present invention. Figure 2 This is a second flowchart illustrating the method for constructing user health records provided by this invention, as shown below. Figure 1 and Figure 2 As shown, the method for constructing a user's health record includes steps S100 to S600, and the specific steps are as follows.

[0035] S100: Perform multi-level structured processing and indexing on at least one type of user health data to obtain the data sequence of various user detection indicators that change over time.

[0036] The implementing entity of this invention includes a system for constructing user health records. For example... Figure 2 As shown, the user health record construction system includes a data input layer, a structured medical data analysis subsystem, a health memory management subsystem, an intelligent agent health analysis subsystem, and an analysis output layer.

[0037] At least one type of health data includes test data in various formats, such as laboratory test reports (e.g., complete blood count, liver function tests, etc.), outpatient and inpatient medical records (including chief complaint and present illness), imaging reports (e.g., CT reports, ultrasound reports, etc.), symptom photos, etc. At least one type of health data includes the user's current test data, current medical history, previous test data, and past medical history.

[0038] Different levels of retrieval involve constructing different levels of search keys based on varying retrieval needs. Feature-based retrieval of all health data is then performed using these different search keys. These different levels of retrieval needs include searching by test report, searching by test indicators, etc.

[0039] The testing indicators include those already tested and associated with the user's condition, such as homocysteine ​​(HCY) levels, thyroid-stimulating hormone (TSH) levels, uric acid (UA) levels, and triglyceride (TG) levels.

[0040] The test indicator data sequence consists of multiple test indicator data points carrying time stamps. The test indicator data includes the test value, any abnormalities, and the reference range. For example, the test indicator data sequence includes the user's low-density lipoprotein cholesterol (LDL-C) test values, abnormalities, and reference ranges from the past five tests: January 15, 2022: 3.8 mmol / L (abnormal, reference range 0~3.37 mmol / L); June 20, 2022: 3.5 mmol / L (abnormal); January 10, 2023: 3.9 mmol / L (abnormal); July 15, 2023: 3.2 mmol / L (normal); January 20, 2024: 3.1 mmol / L (normal).

[0041] The data input layer receives at least one type of health data from the user. The structured medical data analysis subsystem performs multi-level structured processing and indexing on the user's at least one type of health data, extracts multiple detection indicators from the user, and obtains the data sequence of each detection indicator over time.

[0042] S200: Extract key semantic information from various health data, generate key semantic vectors, and store them in the key semantic vector library.

[0043] The health memory management subsystem extracts text summary data from at least one type of user health data, and performs key information retention and intelligent aggregation and deduplication on the extracted text summary data to obtain multiple key semantic data, thereby significantly reducing the amount of text while retaining the key information required for analysis.

[0044] Each key semantic vector is generated based on the key semantic data. All key semantic vectors and their corresponding key semantic data are stored in a key semantic vector library. In this library, key semantic vectors are used for retrieval and matching, while key semantic data is used for input analysis; the two are used in conjunction.

[0045] The key semantic vector library can serve as contextual information for the data sequence of detection metrics. Furthermore, it can be used to evaluate the importance of each detection metric, thereby determining its attention weight.

[0046] S300: Based on the key semantic vector library, determine the importance of each detection indicator to identify key detection indicators in each detection indicator data sequence and obtain the parsing results of key detection indicators.

[0047] The intelligent agent health analysis subsystem comprises multiple agents with different functions, a large language model, and execution tools. These agents include a health analysis agent, a health level assessment agent, and a disease progression analysis agent.

[0048] The detection indicator data sequence and key semantic vector library are input into the intelligent agent health analysis subsystem. The health analysis agent, in conjunction with the key semantic vector library, performs in-depth analysis of each detection indicator to determine its importance. This in-depth analysis includes the clinical significance of abnormal indicators in the detection indicator data sequence, assessment of potential health risks, and targeted health recommendations.

[0049] Based on the importance of each detection indicator, key detection indicators are selected from all detection indicator data sequences.

[0050] Obtain the analytical results of key detection indicators. Optionally, the intelligent agent health analysis subsystem constructs a user prompt keyword search medical literature database based on the key detection indicators to obtain the analytical results of the key detection indicators.

[0051] S400: Determines a user's health level based on a key semantic vector library and data sequences of various detection indicators.

[0052] The data sequences of each detection indicator and the key semantic vector library are input into the health level assessment agent. The health level assessment agent classifies each detection indicator data in the data sequence as normal or abnormal, obtaining an abnormal state sequence for each detection indicator data sequence. Based on the abnormal state sequence, the abnormal change type of each detection indicator is analyzed. Optionally, the abnormal change type includes persistent abnormality, most recent recovery to normal, multiple consecutive recoverys to normal, newly occurring abnormality, and others.

[0053] The reference value of determining the abnormal change types of each detection indicator based on the key semantic vectors can be assessed. Optionally, the timeliness and urgency of the abnormal change types of each detection indicator can be determined based on the key semantic vectors, thereby evaluating the reference value of the abnormal change types of each detection indicator.

[0054] By combining the types of abnormal changes in various detection indicators and their reference value, the user's health level is determined. The user's health level includes normal, low risk, medium risk, high risk, and very high risk.

[0055] Specifically, when a patient's overall indicators are stable recently, with mild abnormalities or a clear trend of improvement, they are prioritized for assessment as low-risk or medium-risk, even if they have a history of multiple abnormalities. When a patient currently has persistent and not mild abnormalities, and the impact of these abnormalities extends across dimensions and systems, without clear signs of improvement, they are assessed as high-risk. When a patient currently faces a highly urgent health threat, and their condition may deteriorate rapidly in the short term, requiring immediate high-level medical intervention, they are assessed as extremely high-risk. Therefore, the health level is not directly determined by the number of abnormalities, the number of reports, or accumulated historical problems, but rather by a comprehensive stratified assessment based on the patient's current overall risk status, abnormal trends, and potential systemic impacts.

[0056] S500: Determine the user's disease progression trend based on the variation amplitude and abnormal change patterns of the detection indicator data sequence.

[0057] The data sequences of various detection indicators and the key semantic vector library are input into the disease progression analysis agent. The data sequences of detection indicators include the magnitude of change and the pattern of abnormal changes. The formula for calculating the magnitude of change is as follows.

[0058] ; When the change exceeds 20% of the preset threshold, the detection index is marked as a significant change detection index.

[0059] Based on the abnormal change patterns of the detection indicator data sequence, identify the abnormal change patterns of the detection indicators.

[0060] By combining significant changes in detection indicators and patterns of abnormal changes in these indicators, the disease progression trend of users can be determined.

[0061] S600: Constructs user health records based on key detection indicators, the analysis results of key detection indicators, health level, and disease progression trends.

[0062] Enter the key detection indicators, their analysis results, health level, and disease progression trend into the corresponding positions in the health record template to obtain the final user health record. Optionally, the key detection indicator is homocysteine. The user health record includes historical homocysteine ​​(Hcy) detection data, the clinical significance of the indicator (analysis results of the key detection indicator), health level, disease progression trend, reasons for concern (analysis results of the key detection indicator), and intervention recommendations (analysis results of the key detection indicator).

[0063] The user's health record includes historical homocysteine ​​(Hcy) test data, as shown in Table 1.

[0064] Table 1

[0065] Clinical significance of the indicator: Homocysteine ​​is an indicator closely related to atherosclerosis, thrombosis risk, and cardiovascular and cerebrovascular diseases. Excessively high levels may indicate underlying vascular health issues. Elevated Hcy levels can damage vascular endothelial cells, promote platelet aggregation, and increase the risk of thrombosis.

[0066] Disease progression trend: Previously in an abnormal state (August 2020-2025), but the most recent test result (December 4, 2025) returned to normal. This is a positive change, indicating that the health condition is developing in a more stable direction. It is recommended to continue to observe and maintain the good condition, and not to let down one's guard.

[0067] Reason for concern: Elevated levels of this indicator may affect cardiovascular function, and its long-term trend needs to be monitored. Even if it has returned to normal, regular monitoring is still necessary to prevent a rebound. It is recommended to check Hcy levels every 3 months and continue to supplement with folic acid, vitamin B6, and B12.

[0068] Intervention recommendations: 1. Supplement with folic acid; 2. Supplement with vitamins B6 and B12; 3. Increase intake of leafy green vegetables, legumes, and whole grains; 4. Limit methionine intake (red meat); 5. Quit smoking and limit alcohol consumption; 6. Check Hcy levels every 3 months.

[0069] The method for constructing a user health record provided in this invention involves multi-level structured processing and indexing of at least one type of user health data to obtain a sequence of detection indicator data showing the changes of various detection indicators over time; extracting key semantic information from various health data to generate key semantic vectors, which are then stored in a key semantic vector library; determining the importance of each detection indicator based on the key semantic vector library to identify key detection indicators in each detection indicator data sequence and obtaining the analytical results of the key detection indicators; determining the user's health level based on the key semantic vector library and the data sequences of each detection indicator; determining the user's disease progression trend based on the magnitude and abnormal change patterns of the detection indicator data sequences; and constructing a user health record based on the key detection indicators, the analytical results of the key detection indicators, the health level, and the disease progression trend. This invention generates a sequence of detection indicator data and a key semantic vector library based on at least one type of health data, achieving the integration of multi-source heterogeneous health data, and further extracting the temporal change information and contextual semantic weights of the detection indicators, which helps improve the accuracy of determining the user's health record. This invention performs differentiated analysis on the detection indicator data sequence and key semantic vector library from multiple dimensions, thereby obtaining key detection indicators, the analysis results of key detection indicators, health level and disease progression trend, realizing in-depth mining of user health data and improving the comprehensiveness and accuracy of user health records.

[0070] Based on the above embodiments, multi-level structured processing and indexing are performed on at least one type of user health data to obtain the data sequence of various detection indicators of the user changing over time, including the following steps: Based on different extraction templates, structured information is extracted from unstructured text of various health data to obtain structured detection indicator data; the different extraction templates include examination report extraction templates and medical record extraction templates; The structured detection index data is split into detection index datasets for each detection index. The names of each detection indicator are aligned and the units are standardized to merge the detection indicator datasets with the same name, thus obtaining the preprocessed detection indicator datasets. Based on the detection time, the data of each detection indicator in the preprocessed detection indicator dataset are sorted by time to obtain the detection indicator data sequence.

[0071] Health data comes in various formats, and its time stamps include the current and previous testing times. A single set of health data may include multiple examination attachments and medical history.

[0072] The structured medical data analysis subsystem performs multi-level structured processing and indexing on at least one type of health data of the user to obtain the data sequence of the user's various test indicators over time.

[0073] This invention constructs a multi-layered medical data indexing architecture based on Elasticsearch.

[0074] This system identifies and extracts unstructured text from various health data sets. When acquiring unstructured text, the search key can be determined using the attachment ID (attachment_id) + file ID (file_id). After acquiring the unstructured text, the original health data is retained, and backtracking of the unstructured text is also supported.

[0075] Obtain structured relational information from unstructured text. Based on different extraction templates and combined with structured relational information, structured information is extracted from the unstructured text of various health data to obtain structured detection indicator data. Different extraction templates include examination report extraction templates and medical record extraction templates. Optionally, the examination report extraction template is used to extract fields such as report category, report title, report date, medical institution, patient information, diagnosis results, list of examination items (including item name, test value, unit, reference range, whether abnormal, direction of abnormality, etc.), and findings and suggestions. The medical record extraction template is used to extract fields such as medical record type, department, doctor information, consultation time, chief complaint, present illness, past medical history, family history, allergy history, personal history, physical examination, auxiliary examinations, diagnosis results, and medical orders.

[0076] Structured testing data includes structured test reports and structured medical records.

[0077] For structured test reports, based on the detection indicators, the report is split into individual indicator datasets for each indicator. During this splitting process, an index key for each indicator is constructed using the attachment ID (attachment_id), file ID (file_id), and item name (item_name). Unstructured text is then split into individual indicator datasets based on this index key. Structured medical records are used to supplement the explanations of each indicator.

[0078] Because the same test indicator may be described with different names in test reports from different medical institutions and at different times (such as "alanine aminotransferase", "alanine aminotransferase", "ALT" etc. all refer to the same indicator), this invention designs a comprehensive standardization and alignment mechanism for test indicator names.

[0079] Construct a synonym mapping table. The synonym mapping table covers multiple categories of indicators such as routine blood tests, liver function tests, kidney function tests, blood lipids, thyroid function tests, blood glucose tests, urinalysis tests, electrolytes, and inflammatory markers, and maps various variant names of the test indicators to standard names.

[0080] The names of the testing indicators are standardized, including removing spaces, standardizing parenthesis format, and standardizing dash format.

[0081] Based on the thesaurus and the normalization of the names of the detection indicators, the names of each detection indicator are aligned and units are standardized to merge datasets of detection indicators with the same name, resulting in preprocessed datasets of each detection indicator. Through this step, the structured medical data analysis subsystem can perform time-series tracking of each detection indicator for the user (patient).

[0082] Furthermore, such as Figure 5 As shown, the multi-layered structured processing and indexing construction of this invention includes four layers. The first layer is used to retrieve raw data (at least one existing health data point) and extract unstructured text. Its index primary key is constructed based on `attachment_id` and `file_id`. The second layer is used to retrieve structured data, including a report sub-index and a medical record sub-index: the report sub-index obtains structured detection indicator data based on structured association information. The medical record sub-index obtains structured medical records based on structured association information. The index primary key of the report sub-index is constructed based on `attachment_id` and `file_id`. The index primary key of the medical record sub-index is constructed based on `attachment_id` and `file_id`. The third layer is used to retrieve detection indicator-level data, including detection indicator index keys. The detection indicator index keys obtain detection indicator datasets according to the detection indicators. The index primary key of the detection indicator index keys is constructed based on `attachment_id`, `file_id`, and `item_name`. The fourth layer is used for aggregated profiling, including user search keys. The user search keys obtain the user's health profile based on the detection indicator dataset and structured medical records. The user search key is constructed based on the user ID (patient_id).

[0083] Based on the detection time, the data for each detection indicator in the preprocessed detection indicator dataset are sorted chronologically to obtain a sequence of detection indicator data. For example, sorting the data for each detection indicator in the preprocessed detection indicator dataset in ascending order of detection time yields a sequence of detection indicator data.

[0084] Furthermore, the data sequence of detection indicators also includes the magnitude of change for each indicator. The structured medical data analysis subsystem calculates the magnitude of change for each indicator. When the magnitude of change for an indicator exceeds a preset threshold, the indicator is marked as a significantly changed indicator.

[0085] Furthermore, the detection indicator data sequence also includes abnormal or normal labels for each detection indicator data.

[0086] Furthermore, the detection indicator data sequence is constructed into a structured table or graph, including the detection time (e.g., detection date), detection value, reference range, abnormal or normal label, and the most recent detection value. For example, Figure 4 The most recent detection value of homocysteine ​​(HCY) content in the sequence was 9.97 μmol / L.

[0087] like Figure 4 As shown, a schematic diagram of the change curves was constructed based on the data sequences of homocysteine ​​(HCY), thyroid-stimulating hormone (TSH), uric acid (UA), triglyceride (TG), and low-density lipoprotein cholesterol (LDL-C).

[0088] This invention progressively organizes health data from unstructured text into preprocessed datasets of various detection indicators through different levels of retrieval. By constructing a sequence of detection indicator data, it enables time-series tracking of the detection indicators.

[0089] Based on the above embodiments, key semantic information is extracted from various health data to generate key semantic vectors, including the following steps: Obtain structured correlation information for each detection indicator; Based on structured association information, key semantic information is extracted from various health data to obtain various key semantic data. Based on the hash value of each key semantic data, duplicate key semantic data are removed to obtain each deduplicated key semantic data. Based on the deduplicated key semantic data, generate each key semantic vector.

[0090] Extracting key semantic information from various health data, generating key semantic vectors, and storing them in the key semantic vector library is part of the writing process of the health memory management subsystem.

[0091] The health memory management subsystem extracts key semantic information from various health data and generates key semantic vectors based on the extracted key semantic information. All key semantic information and their corresponding key semantic vectors are stored in a key semantic vector library. This library addresses the issues of context length limitations and attention decay when health management data contains numerous attachments.

[0092] The structured medical data analysis subsystem extracts structured information from the unstructured text of all health data, obtaining structured correlation information for each detection indicator. The health memory management subsystem then acquires this structured correlation information. This structured correlation information represents the relationships between various detection indicators, and further, it can represent the relationships between various key semantic vectors in the key semantic vector library.

[0093] Key semantic information is extracted from various health data based on structured relational information, resulting in key semantic data. Key semantic data includes abnormal situations, user medical history, timeliness weight, data source, and search semantics. The data source can be determined through structured relational information. Optionally, the timeliness weight of each type of health data is determined based on its urgency and timeliness. For example, for routine examination reports (R1 type), different validity periods are set according to the examination type (e.g., 6 months for blood routine tests, 12 months for imaging reports, etc.). For urgent examination reports (C1 type), the validity period is shortened accordingly. Reports exceeding their validity period are not completely excluded but are assigned a lower timeliness weight (e.g., 0.3-0.5) to reduce their impact in the analysis while retaining their historical reference value.

[0094] The hash value of key semantic data can be calculated based on the user ID (patient_id), the file ID of the health data (file_id), the detection time (report_date), and the content hash value (content_hash) of the key semantic data. The content hash value of the key semantic data is calculated using different methods depending on the document type. For example, the content hash value for report type documents is calculated based on the name and value of the abnormal indicator, while the content hash value for medical record type documents is calculated based on the chief complaint and diagnosis results. Based on the hash value of each key semantic data item, duplicate key semantic data is removed to obtain deduplicated key semantic data. Each deduplicated key semantic data item represents a key semantic aspect of the detection indicator.

[0095] Each deduplicated key semantic data point is converted into a vector, and each key semantic vector is obtained. A key semantic vector library is then created based on all key semantic vectors and their associated key semantic data.

[0096] This invention extracts key semantic data through structured association information, achieving key semantic condensation of various health data. By using the hash value of the key semantic data, duplicate key semantic data is removed, eliminating redundant information and simplifying the data structure of the key semantic vector library.

[0097] Based on the above embodiments, the importance of each detection indicator is determined based on the key semantic vector library to identify key detection indicators in each detection indicator data sequence, including the following steps: Obtain a set number of key semantic vectors from the key semantic vector library; The importance of each detection indicator is identified based on a set number of key semantic vectors. The scores for each detection indicator are calculated based on the importance of each indicator and the anomalies in the data sequences of each indicator. Based on the ranking of the scores, key detection indicators are identified.

[0098] The system retrieves a predetermined number of key semantic vectors from a key semantic vector library. Specifically, the health memory management subsystem reads a predetermined number of key semantic vectors from the library and provides them to the agent health analysis subsystem. The agent health analysis subsystem then retrieves the predetermined number of key semantic vectors from the health analysis agent.

[0099] The health analysis agent identifies the importance of each detection indicator based on a set number of key semantic vectors. The importance of the detection indicator includes its urgency (e.g., routine check vs. emergency check), timeliness (whether it is within its validity period), and proximity of the detection time.

[0100] Anomalies in the detection index data sequence include whether there are abnormal detection values ​​in the detection index data sequence, and the number of times abnormal detection values ​​occur.

[0101] Based on the importance of each testing indicator and the anomalies in the data sequences of each indicator, a score is calculated for each testing indicator. For example, the scoring rules for testing indicators are as follows: 3 points are added for indicators with abnormal values; 2 points are added for indicators used in routine inspections; 1 point is added for indicators used in emergency inspections; 1 point is added for indicators within their validity period; 2 points are added for indicators tested within the last 3 months; and 1 point is added for indicators tested within the last 6 months. The score for each testing indicator is calculated according to these scoring rules.

[0102] Based on the scores of the detection indicators, the indicators are ranked. Then, a set number (e.g., 1-5) of the indicators with higher scores are selected as key detection indicators.

[0103] This invention uses key semantic vectors as contextual information for detection indicators, enabling accurate identification of the importance of each indicator. By combining the importance of the indicators with anomalies in the indicator data sequence, a score is calculated for each indicator, allowing for accurate screening of the key detection indicators most clinically significant for the patient's health status.

[0104] Furthermore, the health analysis agent acquires the analytical results of key detection indicators. The health analysis agent is equipped with a medical literature retrieval tool (e.g., Bocha Search). A maximum inference step limit is set for the health analysis agent, allowing for multiple rounds of inference and invocation of the medical literature retrieval tool. The health analysis agent acquires and analyzes key detection indicators and important abnormal indicator values, forming a preliminary judgment result. Based on the preliminary judgment result, user prompts are constructed. For example, if the preliminary judgment result is "high homocysteine," the user prompt would be "What are the dangers of high homocysteine?"

[0105] The health analysis agent inputs user prompts into a pre-configured large language model. The large language model performs in-depth analysis and consideration of the prompts, and determines whether to invoke medical literature retrieval tools based on the results. If so, it calls the API of the medical literature database to perform a search and retrieve the results. Key information is extracted and organized from the search results. The extracted and organized key information is then analyzed and considered in depth to generate analytical results for key detection indicators.

[0106] For example, the user prompt message is "What are the dangers of high homocysteine ​​levels?"

[0107] In-depth analysis and consideration of user prompts (Thought 1): The user is asking about the dangers of high homocysteine ​​(Hcy) levels. I need to first understand the basic information about Hcy and its health risks. I should use a search engine to obtain authoritative medical information.

[0108] Based on the results of the analysis, decide whether to use the medical literature search tool (Action 1): Use the Bocha search tool to search for "high homocysteine ​​levels are harmful to cardiovascular diseases".

[0109] Using the medical literature search tool: Perform a search and return 5 relevant medical literature results.

[0110] Key information was extracted and organized from the search results (Observation): Elevated Hcy is an independent risk factor for cardiovascular and cerebrovascular diseases [1]. Normal reference range: 5-15 μmol / L [2]. Hcy>15 μmol / L increases the risk of atherosclerosis [3]. Among them, [1], [2] and [3] are the cited data sources.

[0111] Thought2: I have obtained basic information about the dangers of Hcy. Now I have enough information to answer the user's question without needing to use other tools.

[0112] Analysis of key detection indicators (Answer): Elevated homocysteine ​​(Hcy) can cause the following health hazards. 1. Risk of cardiovascular and cerebrovascular diseases: Elevated Hcy is an independent risk factor for cardiovascular and cerebrovascular diseases, which can damage vascular endothelial cells and increase the risk of atherosclerosis[1]. 2. Normal reference range: 5-15 μmol / L, and a value exceeding 15 μmol / L is considered hyperhomocysteinemia[2].

[0113] This invention constructs user prompts using key detection indicators and their important abnormal values, thereby compressing information about these key indicators. Furthermore, by utilizing medical literature retrieval tools to obtain the analytical results of these key detection indicators, this invention achieves an expansion and accurate interpretation of their meanings.

[0114] Based on the above embodiments, the key semantic vectors for setting the quantity are determined in the following manner: Each key semantic vector is aggregated into multiple key semantic vector groups according to the detection indicators; Within the key semantic vector group, multiple valid key semantic vectors are selected based on the importance score of each key semantic vector. The importance score of the key semantic vector is determined based on the anomaly index weight, timeliness weight, urgency weight, and detection index diversity constraints of the key semantic vector. Based on the effective key semantic vectors of each key semantic vector group, a set number of key semantic vectors are obtained.

[0115] Determining the set number of key semantic vectors is part of the reading process of the health memory management subsystem. When the user's health record construction system is configured to enable health memory mode, the health memory management subsystem determines and reads the set number of key semantic vectors.

[0116] When the number of key semantic vectors in the key semantic vector library is less than a set number, the health memory management subsystem outputs all key semantic vectors. When the number of key semantic vectors in the key semantic vector library is greater than a set number, the health memory management subsystem selects a set number of key semantic vectors. This ensures that the key semantic vectors with the best value are retained within a limited context window.

[0117] Optionally, when the total text volume of the key semantic vectors in the key semantic vector library is less than a text volume threshold (e.g., 80,000 characters), the health memory management subsystem outputs all key semantic vectors. When the total text volume of the key semantic vectors in the key semantic vector library is greater than the text volume threshold, the health memory management subsystem selects multiple key semantic vectors, ensuring that the total text volume of the selected multiple key semantic vectors is less than the text volume threshold.

[0118] The health memory management subsystem aggregates each key semantic vector into multiple key semantic vector groups according to detection indicators. One detection indicator corresponds to one key semantic vector group.

[0119] The anomaly indicator weight represents whether the key semantic vector contains outliers of the detected indicator. If the detected indicator contains outliers, the anomaly indicator weight is high. If the detected indicator does not contain outliers, the anomaly indicator weight is low.

[0120] Timeliness weight represents the timeliness of key semantic vectors. Key semantic vectors within their validity period have a high timeliness weight, while those outside their validity period have a low timeliness weight. Furthermore, key semantic vectors from more recent times have a higher timeliness weight, while those from more recent times have a lower timeliness weight.

[0121] The urgency weight represents the urgency level of a key semantic vector. Key semantic vectors belonging to urgent check types (e.g., C1 type, which belongs to the highest urgency level) have high urgency weights. Key semantic vectors not belonging to urgent check types have low urgency weights.

[0122] The diversity constraint for detection metrics is achieved by limiting the maximum number of records retained for each metric. To ensure that each metric has a record, and at most a certain number of records are retained, a diversity constraint is set. If a detection metric has only one key semantic vector, the diversity constraint for that key semantic vector is high. If the number of key semantic vectors for a detection metric exceeds the record count threshold for that metric, the diversity constraint for that key semantic vector is low.

[0123] The importance score of a key semantic vector is determined based on the weights of its anomaly indicators, timeliness, urgency, and the diversity of detection indicators. Optionally, the importance score of a key semantic vector can be obtained by summing the weights of its anomaly indicators, timeliness, urgency, and the diversity of detection indicators.

[0124] Within the key semantic vector group, multiple valid key semantic vectors are selected based on their importance scores. Within the key semantic vector group, each key semantic vector is then ranked according to its importance score. Based on the ranking from highest to lowest importance score, the top N key semantic vectors are selected as valid key semantic vectors.

[0125] Based on the valid key semantic vectors of all key semantic vector groups, a set number of key semantic vectors are obtained.

[0126] This invention, by dividing key semantic vectors into groups and filtering effective key semantic vectors based on their importance scores, ensures that the most valuable key semantic vectors are retained within a limited context window, while also avoiding the attention decay problem of the agent health analysis subsystem.

[0127] Based on the above embodiments, the disease progression trend of a user is determined according to the change amplitude and abnormal change pattern of the detection indicator data sequence, including the following steps: Identify significantly changing detection indicators based on the magnitude of changes in the detection indicator data sequence; To obtain abnormal change patterns in the data sequences of detection indicators; Based on the significant change detection indicators, the abnormal change patterns of the detection indicators, and the deviation values ​​of each detection indicator data in the detection indicator data sequence, user prompt words are generated. Input user prompts into a large language model and obtain the disease progression trend output by the large language model; Among them, the large language model searches medical literature databases based on user-provided prompts and obtains disease progression trends based on the search results.

[0128] Based on the magnitude of changes in the data sequence of detection indicators, significant change detection indicators are identified. When the magnitude of change exceeds 20% of a preset threshold, the detection indicator is marked as a significant change detection indicator.

[0129] Obtain the abnormal variation patterns of the detection indicator data sequence. The abnormal variation patterns include information such as abnormal state sequence, abnormal variation type, total number of tests, number of abnormalities, and the most recent consecutive number of normal / abnormal results.

[0130] The deviation value for each test indicator data point is calculated within the test indicator data sequence. When the test value is higher than the upper limit of the reference range, the deviation value equals the test value minus the upper limit of the reference range, and the deviation value is positive. When the test value is lower than the lower limit of the reference range, the deviation value equals the test value minus the lower limit of the reference range, and the deviation value is negative. For qualitative test indicators (such as Helicobacter pylori antibody positive / negative), the disease progression analysis agent constructs a time-series table containing the examination date, examination results, and status changes.

[0131] Based on the significant change detection indicators, the abnormal change patterns of the detection indicators, and the deviation values ​​of each detection indicator data in the detection indicator data sequence, user prompt words are generated.

[0132] The disease progression analysis agent inputs user-provided prompts into a large language model and obtains the disease progression trend output by the model. Specifically, the large language model searches a medical literature database based on the user-provided prompts and obtains the disease progression trend based on the search results.

[0133] The large language model performs in-depth analysis and consideration of user prompts, and decides whether to invoke medical literature retrieval tools based on the results. If so, it calls the API of the medical literature database to perform a search and obtain the results. Key information is extracted and organized from the search results. The extracted and organized key information is then analyzed and considered in depth to generate a disease progression trend. For example, the disease progression trend might be: "This detection indicator was previously in an abnormal state for a long period (August 2020-2025), but the most recent test result (December 4, 2025) has returned to normal. This is a positive change, indicating that the health condition is developing in a more stable direction. It is recommended to continue to observe and maintain the good condition, and not to let down one's guard."

[0134] This invention constructs user prompt words by analyzing significant changes in detection indicators, abnormal change patterns, and deviations of individual detection indicator data within the data sequence. This enables the extraction and compression of key information from the detection indicator data sequence. By retrieving medical literature databases using a large language model, it achieves in-depth and accurate analysis of significant changes in detection indicators and abnormal change patterns, thus accurately identifying disease progression trends.

[0135] Based on the above embodiments, the user's health level is determined based on the key semantic vector library and the data sequences of various detection indicators, including the following steps: Obtain the sequence of abnormal states of each detection indicator data relative to the normal detection indicator range in the detection indicator data sequence; Based on the abnormal state sequence, determine the type of abnormal change in the detection indicators; Based on a set number of key semantic vectors, the reference value of identifying the abnormal change types of each detection indicator; The user's health level is determined based on the type of abnormal change and its reference value.

[0136] Obtain the sequence of abnormal states for each test indicator data point relative to the normal test indicator range within the test indicator data sequence. For example, the test indicator data sequence is: "2022-01-15 Test value 3.8 mmol / L (abnormal, reference range 0~3.37 mmol / L), 2022-06-20 Test value 3.5 mmol / L (abnormal), 2023-01-10 Test value 3.9 mmol / L (abnormal), 2023-07-15 Test value 3.2 mmol / L (normal), 2024-01-20 Test value 3.1 mmol / L (normal)". The abnormal state sequence for the test indicator is [abnormal, abnormal, abnormal, normal, normal].

[0137] Abnormal change types include persistent abnormality, most recent recovery to normal, multiple consecutive recovery to normal, newly occurring abnormality, and others. The criteria for persistent abnormality are: current abnormality and at least two recent consecutive abnormal occurrences. The criteria for most recent recovery to normal are: current normality, historical abnormality records, and exactly one recent consecutive normal occurrence. The criteria for multiple consecutive normal occurrences are: current normality, historical abnormality records, and at least two recent consecutive normal occurrences. The criteria for newly occurring abnormality are: current abnormality but no historical abnormality records. Situations not meeting any of the above conditions are classified as "other."

[0138] Based on the abnormal state sequence, the type of abnormal change in the detection index is determined. The abnormal state sequence is analyzed to obtain the results. The results include: total number of detections, number of abnormalities, number of most recent consecutive normal values, number of most recent consecutive abnormal values, whether the most recent detection was abnormal, and whether there are any historical abnormal records. Based on the analysis results, the type of abnormal change in the detection index is determined.

[0139] For example, the analysis results show: a total of 5 tests, 3 abnormal tests, 2 most recent consecutive normal tests, 0 most recent consecutive abnormal tests, a current normal test, and historical abnormal records. The corresponding abnormal change type of the detection indicator is "consecutive returns to normal."

[0140] Based on a predetermined number of key semantic vectors, the reference value of abnormal change types for each detection indicator is identified. Reference value characterizes the importance and relevance of the abnormal change type of the detection indicator. Based on the predetermined number of key semantic vectors, the timeliness and urgency of the abnormal change type of each detection indicator are identified, thereby determining the reference value of each abnormal change type.

[0141] A user's health level is categorized as normal, low risk, medium risk, high risk, and very high risk. The user's health level is determined based on the type of abnormal change and its reference value. The health level assessment agent follows the principles of "prioritizing recent status, correcting for historical information, and avoiding mechanical accumulation" to determine the user's health level. When a user's overall indicators are stable recently, the degree of abnormality is mild, or there is a clear trend of improvement, even if there are multiple previous abnormal records, they are prioritized for low or medium risk. When a patient currently has persistent and not mild abnormalities, and the impact of the abnormalities extends across dimensions and systems, and there is a lack of clear signs of improvement, they are assessed as high risk. When a patient currently faces a highly urgent health threat, and may rapidly deteriorate in the short term, requiring immediate high-level medical intervention, they are assessed as very high risk.

[0142] This invention, by setting a number of key semantic vectors, can accurately obtain historical contextual information for each detection indicator, thereby identifying the reference value of abnormal change types for each indicator. By combining the abnormal change types and their reference value, this invention can accurately determine a user's health level.

[0143] like Figure 3 As shown, the present invention provides a method for constructing a user health record, which includes the following steps.

[0144] (1) The structured medical data analysis subsystem acquires at least one type of health data.

[0145] (2) The structured medical data analysis subsystem acquires unstructured text.

[0146] (3) The structured medical data analysis subsystem extracts structured related information.

[0147] The structured medical data analysis subsystem extracts structured detection indicator data based on structured correlation information and different extraction templates. This structured detection indicator data is then split into individual detection indicator datasets. The names of the detection indicators are aligned and units are standardized to obtain a preprocessed detection indicator dataset. Finally, the preprocessed dataset is sorted by time to obtain the data sequences for each detection indicator.

[0148] The health memory management subsystem initiates a health memory update. It extracts key semantic data from all health data based on structured association information. Duplicate key semantic data is removed based on the hash value of the key semantic data, and a key semantic vector is constructed.

[0149] (4) Trigger health analysis by combining the data sequences of various detection indicators and the key semantic vector library.

[0150] The health level assessment agent performs in-depth analysis and consideration of the data sequences of various detection indicators and the key semantic vector library, and at the same time calls medical literature retrieval tools to output the health level and assessment basis.

[0151] The health analysis agent performs in-depth analysis and analysis on the data sequences of various detection indicators and the key semantic vector library, while calling medical literature retrieval tools to output key detection indicators and their analysis results.

[0152] Significantly changing indicators and their abnormal patterns are calculated based on the data sequences of each detection indicator. These significantly changing indicators and abnormal patterns are then input into the disease progression analysis agent to obtain the disease progression trend.

[0153] (5) Construct user health records based on health level, key detection indicators and their analysis results and disease progression trends.

[0154] This invention combines the reasoning capabilities of a large language model with the invocation of external medical literature retrieval tools. During the analysis process, the intelligent agent can proactively search medical literature databases, query historical data, and perform numerical calculations, ensuring that the analytical conclusions are fully supported by data and medical evidence, rather than relying solely on the model's internal knowledge. A comprehensive citation management mechanism allows users and doctors to trace the source of every analytical conclusion.

[0155] The user health record construction system provided by the present invention will be described below. The user health record construction system described below can be referred to in correspondence with the user health record construction method described above.

[0156] like Figure 6 As shown, a user health record construction system includes: a structured medical data analysis subsystem, a health memory management subsystem, a health analysis agent, a health level assessment agent, a disease progression analysis agent, and an analysis output layer. The structured medical data analysis subsystem is used to perform multi-level structured processing and indexing of at least one type of health data of users, and to obtain the data sequence of various test indicators of users changing over time. The health memory management subsystem is used to extract key semantic information from various health data, generate key semantic vectors, and store them in the key semantic vector library. A health analysis intelligent agent is used to determine the importance of each detection indicator based on a key semantic vector library, in order to identify key detection indicators in the data sequence of each detection indicator and obtain the analysis results of key detection indicators. A health level assessment agent is used to determine a user's health level based on a key semantic vector library and data sequences of various detection indicators. The disease progression analysis agent is used to determine the user's disease progression trend based on the magnitude and abnormal change patterns of the detection indicator data sequence. The analysis output layer is used to construct user health profiles based on key detection indicators, the analysis results of key detection indicators, health level, and disease progression trends.

[0157] The intelligent agent health analysis subsystem comprises a health analysis agent, a health level assessment agent, and a disease progression analysis agent. The intelligent agent health analysis subsystem reads data sequences of various detection indicators from the structured medical data analysis subsystem. It also reads a key semantic vector library from the health memory management subsystem and obtains a predetermined number of key semantic vectors.

[0158] The intelligent agent health analysis subsystem will input the data sequences of various detection indicators and a predetermined number of key semantic vectors into the health analysis intelligent agent and the health level assessment intelligent agent, respectively. The intelligent agent health analysis subsystem will input the data sequences of various detection indicators into the disease progression analysis intelligent agent.

[0159] The health analysis agent determines the importance of each detection indicator based on a set number of key semantic vectors, in order to identify key detection indicators in each detection indicator data sequence and obtain the analysis results of key detection indicators.

[0160] The health level assessment agent determines the user's health level based on a set number of key semantic vectors and data sequences of various detection indicators.

[0161] The disease progression analysis agent determines the user's disease progression trend based on the magnitude of changes and abnormal patterns in the data sequences of various detection indicators.

[0162] The analysis output layer reads the key detection indicators and their parsing results output by the health analysis agent, the user's health level output by the health level assessment agent, and the disease progression trend output by the disease progression analysis agent, and constructs the user's health profile.

[0163] The health analysis agent, health level assessment agent, and disease progression analysis agent of this invention work collaboratively simultaneously, enabling parallel analysis of a user's health status from multiple professional dimensions, including health level, key detection indicators and their interpretation, and disease progression tracking. Each agent focuses on its respective professional field, avoiding the problem of a single model being unable to cover both comprehensiveness and depth. The concurrent execution of multiple agents also significantly improves analysis efficiency.

[0164] The structured medical data analysis subsystem of this invention enables the traceability, calculability, and comparability of test indicators. Through a multi-level indexing architecture, intelligent structured extraction, and a standardized mechanism for test indicator names, the subsystem unifies the storage and management of test indicators scattered across different test reports from users (patients), achieving cross-time and cross-institutional indicator correlation and time-series tracking. In disease progression analysis, it can clearly display the historical trend of each indicator, helping to identify the direction of disease development.

[0165] The health memory management subsystem of this invention effectively solves the problems of contextual constraints and attention decay in large language models. By extracting summaries and retaining key semantic information from various health data, the complete report content is compressed into refined key semantic data (memory entries), significantly reducing the amount of text input to the large language model. An aggregation deduplication mechanism further eliminates redundant information, and an intelligent filtering mechanism ensures that the most valuable information is retained within a limited context window. This method can reduce the amount of input text by more than 50% while maintaining or even improving the analysis quality.

[0166] This invention supports incremental updates, improving the efficiency and scalability of the user health record construction system. Both the health memory management subsystem and the structured medical data analysis subsystem support incremental updates. When a user adds or deletes health data, these subsystems do not need to reprocess all previous health data; they only need to update the changed portions, significantly improving the operational efficiency and scalability of the user health record construction system.

[0167] The user health record construction system provided in this invention performs multi-level structured processing and indexing on at least one type of user health data to obtain a sequence of detection indicator data showing the changes of various detection indicators over time. It extracts key semantic information from various health data, generates key semantic vectors, and stores them in a key semantic vector library. Based on the key semantic vector library, it determines the importance of each detection indicator to identify key detection indicators in each detection indicator data sequence and obtains the analytical results of these key indicators. Based on the key semantic vector library and the data sequences of each detection indicator, it determines the user's health level. Based on the amplitude and abnormal change patterns of the detection indicator data sequences, it determines the user's disease progression trend. Based on the key detection indicators, the analytical results of the key detection indicators, the health level, and the disease progression trend, it constructs the user's health record. This invention generates a sequence of detection indicator data and a key semantic vector library based on at least one type of health data, achieving the integration of multi-source heterogeneous health data. This allows for the extraction of temporal change information and contextual semantic weights of detection indicators, which helps improve the accuracy of determining the user's health record. This invention performs differentiated analysis on the detection indicator data sequence and key semantic vector library from multiple dimensions, thereby obtaining key detection indicators, the analysis results of key detection indicators, health level and disease progression trend, realizing in-depth mining of user health data and improving the comprehensiveness and accuracy of user health records.

[0168] All relevant content of each step in the above method embodiment can be referenced from the functional descriptions of the corresponding structured medical data analysis subsystem, health memory management subsystem, health analysis intelligent agent, health level assessment intelligent agent, disease progression analysis intelligent agent, and analysis output layer, and will not be repeated here.

[0169] Figure 7 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 7As shown, the electronic device may include: a processor 710, a communications interface 720, a memory 730, and a communication bus 740, wherein the processor 710, the communications interface 720, and the memory 730 communicate with each other through the communication bus 740. The processor 710 can call logical instructions in the memory 730 to execute a method for constructing a user health record. This method includes: performing multi-level structured processing and indexing on at least one type of user health data to obtain a sequence of user detection indicators that change over time; extracting key semantic information from various health data to generate key semantic vectors and storing them in a key semantic vector library; determining the importance of each detection indicator based on the key semantic vector library to identify key detection indicators in each detection indicator data sequence and obtaining the analysis results of the key detection indicators; determining the user's health level based on the key semantic vector library and the data sequences of each detection indicator; determining the user's disease progression trend based on the amplitude and abnormal change patterns of the detection indicator data sequences; and constructing a user health record based on the key detection indicators, the analysis results of the key detection indicators, the health level, and the disease progression trend.

[0170] Furthermore, the logical instructions in the aforementioned memory 730 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0171] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements a method for constructing a user health profile provided by the methods described above. This method includes: performing multi-level structured processing and indexing on at least one type of user health data to obtain a sequence of user detection indicators that change over time; extracting key semantic information from various health data to generate key semantic vectors and storing them in a key semantic vector library; determining the importance of each detection indicator based on the key semantic vector library to identify key detection indicators in the data sequence of each detection indicator and obtaining the analysis results of the key detection indicators; determining the user's health level based on the key semantic vector library and the data sequence of each detection indicator; determining the user's disease progression trend based on the magnitude and abnormal change patterns of the detection indicator data sequence; and constructing a user health profile based on the key detection indicators, the analysis results of the key detection indicators, the health level, and the disease progression trend.

[0172] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0173] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0174] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing a user health record, characterized in that, include: Perform multi-level structured processing and index construction on at least one type of user's health data to obtain the data sequence of the user's various detection indicators changing over time; Key semantic information is extracted from various health data, generating key semantic vectors and storing them in a key semantic vector library; Based on the key semantic vector library, the importance of each detection indicator is determined, so as to identify the key detection indicators in each detection indicator data sequence and obtain the parsing results of the key detection indicators. Based on the key semantic vector library and the data sequences of each detection indicator, the user's health level is determined. Based on the magnitude and abnormal change patterns of the detection indicator data sequence, the disease progression trend of the user is determined; Based on the key detection indicators, the analysis results of the key detection indicators, the health level, and the disease progression trend, a user health profile is constructed.

2. The method for constructing a user health record according to claim 1, characterized in that, The process of performing multi-level structured processing and indexing on at least one type of user health data to obtain a time-varying sequence of various detection indicators for the user includes: Based on different extraction templates, structured information is extracted from the unstructured text of various health data to obtain structured detection indicator data; the different extraction templates include examination report extraction templates and medical record extraction templates; The structured detection index data is split into detection index datasets for each of the detection indicators; The names of each detection index are aligned and the units are unified to merge the detection index datasets with the same name, thus obtaining the preprocessed detection index datasets. Based on the detection time, the data of each detection indicator in the preprocessed detection indicator dataset is sorted by time to obtain the detection indicator data sequence.

3. The method for constructing a user health record according to claim 1, characterized in that, The step of extracting key semantic information from various health data and generating key semantic vectors includes: Obtain the structured correlation information of each of the detection indicators; Based on the structured association information, the key semantic information of various health data is extracted to obtain each key semantic data. Based on the hash value of each key semantic data, duplicate key semantic data are removed to obtain each deduplicated key semantic data. Based on the deduplicated key semantic data, generate each key semantic vector.

4. The method for constructing a user health record according to claim 1, characterized in that, The step of determining the importance of each detection indicator based on the key semantic vector library, in order to identify key detection indicators in each detection indicator data sequence, includes: Obtain a set number of key semantic vectors from the key semantic vector library; The importance of each detection indicator is identified based on the set number of key semantic vectors. The score for each detection indicator is calculated based on the importance of each detection indicator and the anomalies in the data sequence of each detection indicator. Based on the ranking results of the scores, the key detection indicators are identified.

5. The method for constructing a user health record according to claim 1, characterized in that, The step of determining the user's disease progression trend based on the variation amplitude and abnormal variation pattern of the detection indicator data sequence includes: Based on the change amplitude of the detection index data sequence, identify detection indicators with significant changes; Obtain the abnormal change patterns of the detection indicator data sequence; Based on the significant change detection index, the abnormal change pattern of the detection index, and the deviation value of each detection index data in the detection index data sequence, user prompt words are generated; Input the user prompt words into the large language model and obtain the disease progression trend output by the large language model; The large language model retrieves medical literature databases based on the user-provided prompts and obtains the disease progression trend based on the retrieval results.

6. The method for constructing a user health record according to claim 1, characterized in that, The process of determining a user's health level based on the key semantic vector library and the data sequences of each detection indicator includes: Obtain the sequence of abnormal states of each detection indicator data in the detection indicator data sequence relative to the normal detection indicator range; Based on the abnormal state sequence, determine the type of abnormal change in the detection index; Based on a set number of key semantic vectors, the reference value for identifying the abnormal change types of each of the detection indicators; The user's health level is determined based on the type of abnormal change and its reference value.

7. The method for constructing a user health record according to claim 4, characterized in that, The specified number of key semantic vectors are determined based on the following method: The individual key semantic vectors are aggregated into multiple key semantic vector groups according to the detection indicators; Within the key semantic vector group, multiple valid key semantic vectors are selected based on the importance score of each key semantic vector; the importance score of the key semantic vector is determined based on the anomaly index weight, timeliness weight, urgency weight, and detection index diversity constraints of the key semantic vector. Based on the effective key semantic vectors of each of the key semantic vector groups, the set number of key semantic vectors are obtained.

8. A system for constructing user health records, characterized in that, include: Structured medical data analysis subsystem, health memory management subsystem, health analysis intelligent agent, health level assessment intelligent agent, disease progression analysis intelligent agent, and analysis output layer: The structured medical data analysis subsystem is used to perform multi-level structured processing and index construction on at least one type of health data of the user, and to obtain the data sequence of the user's various detection indicators over time. The health memory management subsystem is used to extract key semantic information from various health data, generate key semantic vectors, and store them in a key semantic vector library. The health analysis intelligent agent is used to determine the importance of each detection indicator based on the key semantic vector library, so as to identify key detection indicators in each detection indicator data sequence and obtain the parsing results of key detection indicators. The health level assessment agent is used to determine the user's health level based on the key semantic vector library and the data sequences of each detection indicator; The disease progression analysis intelligent agent is used to determine the user's disease progression trend based on the change amplitude and abnormal change pattern of the detection indicator data sequence; The analysis output layer is used to construct a user health profile based on the key detection indicators, the analysis results of the key detection indicators, the health level, and the disease progression trend.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the method for constructing a user health record as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method for constructing a user health record as described in any one of claims 1 to 7.

Images