Methods, equipment, and media for generating auxiliary health risk analysis reports
By encapsulating abnormal evidence sets, combinations, and multi-time point indicator change trends into promoters, and utilizing deterministic candidate recall and triage models, the problem of inconsistent interpretation of abnormalities in medical health risk analysis in existing technologies is solved, achieving clarity in task assignment and visibility of conflicts, and generating reliable auxiliary health risk analysis reports.
Patent Information
- Application Number
- CN202611549967.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-08
- Publication Date
- 2026-11-17
AI Technical Summary
In existing technologies, medical health risk analysis suffers from problems such as isolated interpretation of single anomalies and repeated interpretation of cross-system anomalies. It lacks an integrated association between anomaly combinations and triage reasons, the model triage lacks verifiable roster boundaries, the communication boundaries of intelligent agents are unclear, conflict handling is easily masked by consensus, it is difficult to distinguish whether colleagues' messages have actually arrived and been adopted, the system has difficulty in stably identifying completion status, and it affects the final review and release judgment.
By associating abnormal evidence sets, abnormal combinations, and multi-time point indicator change trends, a promoter is encapsulated. Deterministic candidate recall is used to obtain reference candidate departments, which are then input into the triage model to obtain departments and items of concern. Specialty agents independently analyze the data, and the results are aggregated by the general agent to generate an auxiliary health risk analysis report.
It improves the integrity of medical context, ensures the clarity and traceability of task assignment, enables the system to clearly display conflicts, and generates degradeable auxiliary health risk analysis reports, thereby improving the accuracy and reliability of the analysis.
Smart Images

Figure CN123120181A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of computer technology, and in particular to a technology for generating auxiliary health risk analysis reports. Background Technology
[0002] In existing technologies, agent collaboration is typically organized based on doctor conversations, medical record texts, individual test items, or general medical questions. Task assignment granularity is often coarse or scattered, leading to isolated interpretations of single anomalies, duplicate interpretations of cross-system anomalies, or unclaimed anomalies. Furthermore, it often focuses only on anomalies at the current time point, failing to consider changes in the same indicator for the same object from normal to anomaly at multiple time points as task triggering criteria alongside the current anomaly evidence. There is a lack of integrated association between anomaly combinations and triage rationale. While existing solutions can set expert roles or collaborative structures, input evidence, participating departments, reasons for department selection, and unanswered questions may be scattered across different stages, making it difficult to trace "why this department was chosen" and "where the analysis for this department begins." Model triage lacks verifiable roster boundaries. Relying entirely on the model's free selection of agents may result in absent, duplicate, or unexplained departments; hard assignment based solely on the first label of the terminology may lose cross-system relationships. The model needs to see the entire batch of inputs, while the program verifies the validity of the department selection results and records the reasons for exclusion from selected departments. Agent communication boundaries are unclear. Free point-to-point dialogues struggle to centrally control message quantity, length, response relationships, and timeout status. They also struggle to distinguish whether colleague messages have actually arrived, been adopted, and whether their source was verified upon adoption. Conflict handling is easily masked by "consensus." In medical health risk analysis, inconsistencies in judgments across different specialties are indicative; if the system replaces conflict coexistence with forced consensus, write-overwrite, or single summaries, it may obscure differences requiring further physician confirmation. Furthermore, there is a lack of programmably identifiable collaborative states. If departmental results and messages are presented solely in natural language, the system struggles to consistently identify completed, partially completed, timed-out, and failed states. It also struggles to distinguish between unanswered questions, insufficient evidence, and failed source verification, thus affecting final review and decision-making. Summary of the Invention
[0003] One object of this application is to provide a method, apparatus, medium, and program product for generating auxiliary health risk analysis reports.
[0004] According to one aspect of this application, a method for generating an auxiliary health risk analysis report is provided, the method comprising:
[0005] Multiple items from the abnormal evidence set, abnormal combination, and multi-time point indicator change trend are associated and encapsulated into one or more promoters. The abnormal evidence set, abnormal combination, and multi-time point indicator change trend are obtained based on the physical examination report file. The promoter includes at least one of the following: member abnormal item, trend event, combination pattern type, and co-occurrence abnormal summary.
[0006] The reference candidate departments corresponding to the promoter are obtained through deterministic candidate recall, and the combined features corresponding to the promoter are extracted based on the reference candidate departments;
[0007] Input the promoter, the reference candidate department, the combined features, and the directory of departments in service into the triage model to obtain at least one department output by the triage model and the corresponding departmental focus items for the at least one department;
[0008] Based on the promoter and the undergraduate focus items, establish the analysis task corresponding to the at least one department, and submit the analysis task to the specialty agent corresponding to the at least one department;
[0009] The general inspection agent aggregates the results and statuses of the specialist agents regarding the analysis task to generate an auxiliary health risk analysis report corresponding to the physical examination report file.
[0010] According to one aspect of this application, a computer device for generating auxiliary health risk analysis reports is provided, comprising a memory, a processor, and a computer program stored in the memory, characterized in that the processor executes the computer program to implement the steps of any of the methods described above.
[0011] According to one aspect of this application, a computer-readable storage medium is provided having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements the steps of any of the methods described above.
[0012] According to one aspect of this application, a computer program product is provided, comprising a computer program, characterized in that, when executed by a processor, the computer program implements the steps of any of the methods described above.
[0013] According to one aspect of this application, a system for generating auxiliary health risk analysis reports is provided, the system comprising:
[0014] The first module is used to associate and encapsulate multiple items from the abnormal evidence set, abnormal combination, and multi-time point indicator change trend into one or more promoters. The abnormal evidence set, the abnormal combination, and the multi-time point indicator change trend are obtained based on the physical examination report file. The promoter includes at least one of the following: member abnormal item, trend event, combination pattern type, and co-occurrence abnormal summary.
[0015] The second module is used to obtain reference candidate departments corresponding to the promoter through deterministic candidate recall, and extract the combined features corresponding to the promoter based on the reference candidate departments;
[0016] The third module is used to input the promoter, the reference candidate department, the combined features, and the directory of departments in service into the triage model to obtain at least one department output by the triage model and the corresponding departmental focus item of the at least one department;
[0017] The fourth module is used to establish analysis tasks corresponding to the at least one department based on the promoter and the undergraduate focus items, and submit the analysis tasks to the specialist agent corresponding to the at least one department.
[0018] The fifth module is used to aggregate the results status of the specialist agents regarding the analysis task through the general inspection agent, and generate an auxiliary health risk analysis report corresponding to the physical examination report file.
[0019] According to one aspect of this application, a computer device for generating auxiliary health risk analysis reports is provided, the device comprising:
[0020] The module is used to associate and encapsulate multiple items from the abnormal evidence set, abnormal combination, and multi-time point indicator change trend into one or more promoters. The abnormal evidence set, abnormal combination, and multi-time point indicator change trend are obtained based on the physical examination report file. The promoter includes at least one of the following: member abnormal item, trend event, combination pattern type, and co-occurrence abnormal summary.
[0021] The first and second modules are used to obtain reference candidate departments corresponding to the promoter through deterministic candidate recall, and extract the combined features corresponding to the promoter based on the reference candidate departments;
[0022] The first and third modules are used to input the promoter, the reference candidate departments, the combined features, and the directory of departments in service into the triage model to obtain at least one department output by the triage model and the corresponding departmental focus items of the at least one department;
[0023] The first four modules are used to establish analysis tasks corresponding to the at least one department based on the promoter and the undergraduate focus items, and submit the analysis tasks to the specialized intelligent agents corresponding to the at least one department.
[0024] The first and fifth modules are used to aggregate the results and status of the specialist agents regarding the analysis task through the general inspection agent, and generate an auxiliary health risk analysis report corresponding to the physical examination report file.
[0025] Compared with existing technologies, this application uses externally formed sets of abnormal evidence, abnormal combinations, and trends of indicator changes at multiple time points as starters at the same input level, improving the completeness of the medical context for task assignment. The system no longer assigns tasks based solely on a single abnormality, current report, or entire case, but instead allows numerical values, deviation direction, severity, abnormality source, co-occurrence relationship, and longitudinal changes in the same indicator for the same object from normal to abnormal to participate in the analysis of task assignment. Through controlled collaboration between specialist agents and general agents, an auxiliary health risk analysis report is generated that is traceable in terms of assignment and completion status, has clear boundaries, visible conflicts, and can be degraded. Attached Figure Description
[0026] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:
[0027] Figure 1 This diagram illustrates a method for generating an auxiliary health risk analysis report according to an embodiment of this application;
[0028] Figure 2 This diagram illustrates a method for generating an auxiliary health risk analysis report according to one embodiment of the present application;
[0029] Figure 3 This diagram illustrates a system block diagram for generating an auxiliary health risk analysis report according to one embodiment of the present application;
[0030] Figure 4 This diagram illustrates a system architecture for generating an auxiliary health risk analysis report according to one embodiment of the present application.
[0031] Figure 5 This diagram illustrates a computer device structure for generating an assisted health risk analysis report according to one embodiment of the present application.
[0032] Figure 6 Exemplary systems that can be used to implement the various embodiments described in this application are shown.
[0033] The same or similar reference numerals in the accompanying drawings represent the same or similar parts. Detailed Implementation
[0034] The present application will now be described in further detail with reference to the accompanying drawings.
[0035] In a typical configuration of this application, the terminal, the device of the service network, and the trusted party all include one or more processors (e.g., a central processing unit (CPU)), input / output interfaces, network interfaces, and memory.
[0036] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash memory. Memory is an example of computer-readable media.
[0037] Computer-readable media, including both permanent and non-permanent, removable and non-removable media, can store information using any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PCM), programmable random access memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.
[0038] The devices referred to in this application include, but are not limited to, user equipment, network equipment, or devices composed of user equipment and network equipment integrated through a network. The user equipment includes, but is not limited to, any mobile electronic product capable of human-computer interaction (e.g., via a touchpad), such as smartphones and tablets. These mobile electronic products can use any operating system, such as Android or iOS. The network equipment includes an electronic device capable of automatically performing numerical calculations and information processing according to pre-set or stored instructions. Its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), and embedded devices. The network equipment includes, but is not limited to, computers, network hosts, single network servers, multiple network server clusters, or clouds composed of multiple servers. Here, a cloud consists of a large number of computers or network servers based on cloud computing, where cloud computing is a type of distributed computing, consisting of a virtual supercomputer composed of a group of loosely coupled computer clusters. The network includes, but is not limited to, the Internet, wide area network, metropolitan area network, local area network, VPN network, wireless ad hoc network, etc. Preferably, the device can also be a program running on the user equipment, network device, or a device formed by integrating user equipment and network device, network device, touch terminal, or network device and touch terminal through a network.
[0039] Of course, those skilled in the art should understand that the above-described devices are merely examples, and other existing or future devices that are applicable to this application should also be included within the scope of protection of this application, and are hereby incorporated by reference.
[0040] In the description of this application, "multiple" means two or more, unless otherwise expressly and specifically defined.
[0041] Figure 1This document illustrates a flowchart of a method for generating an auxiliary health risk analysis report according to an embodiment of this application. The method includes steps S11, S12, S13, S14, and S15. In step S11, a computer device associates and encapsulates multiple items from anomaly evidence sets, anomaly combinations, and multi-time-point indicator change trends into one or more promoters. The anomaly evidence sets, anomaly combinations, and multi-time-point indicator change trends are obtained based on a medical examination report file. The promoter includes at least one of member anomalies, trend events, combination pattern types, and co-occurrence anomaly summaries. In step S12, the computer device obtains reference candidate departments corresponding to the promoters through deterministic candidate recall and extracts combination features corresponding to the promoters based on the reference candidate departments. In step S13, the computer device... The system inputs the promoter, the reference candidate department, the combined features, and the directory of departments currently in service into the triage model to obtain at least one department output by the triage model and the corresponding departmental focus items for the at least one department; in step S14, the computer device establishes an analysis task corresponding to the at least one department based on the promoter and the departmental focus items, and submits the analysis task to the specialist agent corresponding to the at least one department; in step S15, the computer device aggregates the result status of the specialist agent regarding the analysis task through the general examination agent, and generates an auxiliary health risk analysis report corresponding to the physical examination report file.
[0042] In step S11, the computer device associates and encapsulates multiple items from the abnormal evidence set, abnormal combination, and multi-time point indicator change trend into one or more promoters. The abnormal evidence set, abnormal combination, and multi-time point indicator change trend are obtained based on the physical examination report file. The promoter includes at least one of the following: member abnormal item, trend event, combination pattern type, and co-occurrence abnormal summary.
[0043] In some embodiments, the input is not the medical examination report file itself, but rather multiple core inputs obtained from the medical examination report file at the same input level: a structured set of abnormal evidence already generated by the upstream module, abnormal combinations formed by organ systems, clinical problems, or co-occurrence relationships, and the trend of changes in the same indicator at multiple time points for the same object. Each piece of structured abnormal evidence includes at least the name or standard code of the abnormal item, the current result, the unit, the reference range or judgment basis, the direction of deviation, the source of the abnormality, the severity, the confidence level, the source location, and the identifier of the object to which it belongs; each abnormal combination includes at least the member abnormal items, the combination label, the source range, and the quality mark; each trend of changes in the indicator at multiple time points includes at least the object identifier, the name or standard code of the indicator, the previous time point, the next time point, the previous and next observations or qualitative conclusions, the previous and next reference ranges or judgment basis, the previous state, the next state, the type of state change, the direction of change, the time interval, and the source location, wherein the type of state change includes normal to abnormal, persistent abnormal, near the edge of abnormality, and abnormal to normal, etc.: the previous state is normal and the next state is normal. Events with an abnormal state are marked as normal-to-abnormal trend events and given the highest priority. Events with abnormal states before and after are marked as persistently abnormal and further differentiated into aggravated, stable, improved, or fluctuating states. Events with at least three valid numerical points that are normal throughout, progressively and strictly increasing, and whose relative position in the reference interval at the last point exceeds the configured threshold are marked as approaching the upper limit. Events with an abnormal state in the previous period and a normal state in the next period are marked as abnormal-to-normal trend events. The attention priority is recorded separately as normal-to-abnormal P1, aggravated abnormal P2, approaching the upper limit P3, and improved abnormal P4. Categories such as abnormal stability, fluctuation, transient abnormality, normal stability, and undetermined are also retained. Situations that are still within the abnormal range but are improving towards the reference interval can also be classified as abnormal improvement. In some embodiments, this application does not limit the upstream generation method for obtaining abnormal evidence sets, abnormal combinations, and multi-time point indicator change trends based on medical examination report documents; the three are at the same core input level.
[0044] In some embodiments, anomaly evidence sets undergo field integrity checks, requiring at least the inclusion of anomaly identifier, name, current result or qualitative conclusion, deviation direction, anomaly source, severity or priority, confidence level, source location, and object identifier. In some embodiments, anomaly combinations undergo field integrity checks, requiring at least the inclusion of member anomalies, combination labels, source range, and quality markers. In some embodiments, multi-time-point indicator change trends undergo field integrity checks, requiring at least the inclusion of object identifier, indicator identifier, previous and subsequent time points, previous and subsequent states, previous and subsequent observations or qualitative conclusions, previous and subsequent reference ranges or judgment criteria, state change type, time interval, and source location, with normal-to-anomaly events being a key focus input. In some embodiments, if key fields are missing from anomaly evidence sets, anomaly combinations, or multi-time-point indicator change trends, missing markers and data limitations are retained, and subsequent analysis and result status reflect these limitations; missing information is not added as known facts.
[0045] In some embodiments, multiple items from the abnormal evidence set, abnormal combination, and multi-time point indicator change trends are associated and encapsulated into one or more promoters. These promoters further associate the triage, task, and traceability information required for collaboration. The promoter is the logical starting point for collaborative analysis and can be carried by interconnected data objects; it is not required to be merged into a program class with a fixed name. In some embodiments, the promoter is a task package with a source, grouping the abnormal evidence set, abnormal combination, and multi-time point indicator change trends of the same user together. After triage, the selected department, selection reason, focus item, exclusion record, and departmental status (i.e., the result status of the specialist agent corresponding to each department) are associated. The promoter can be carried by several interconnected data objects, without requiring a program class with the same name. In some embodiments, the fields in the promoter are filled in step by step. During encapsulation, at least one of the following is filled in first: member abnormal item, trend event, combination mode type, and co-occurrence abnormal summary. Other fields such as selected department, focus item, exclusion reason, running identifier, and departmental status are filled in later. Undecided decisions are not filled in as final values in advance. In some embodiments, the promoter includes at least one of the following: member anomalies, trend events, combination pattern types, and co-occurrence anomaly summaries. Member anomalies include the code, name, current value, unit, reference range or judgment basis, deviation direction, anomaly source, severity, confidence level, and source location for each anomaly. Trend events include the time points before and after, the states before and after, the observations before and after, the type of state change, the direction of change, the time interval, the trend classification, and the priority of attention for the same indicator of the same object. Combination pattern types refer to the way in which a group of promoters are composed, including but not limited to single-system multiple anomalies, cross-system co-occurrence, severe anomaly priority, normal to anomaly trend, continuous anomaly trend, trend close to the edge of anomaly, anomaly to normal trend, low confidence pending coordination, and miss-the-word combination, etc. Co-occurrence anomaly summaries are used to summarize the names, deviation directions, and co-occurrence relationships of related anomalies to help departments identify key points. The summaries do not replace accessible original evidence, nor do they mask other indicators according to the primary responsibility or copying identity.In some embodiments, the promoter also includes a promoter identifier (uniquely identifying the promoter), an object and task identifier (corresponding to the subject, assessment task, report batch, or session identifier), a reference candidate department (a list of possible relevant departments obtained from a thesaurus, rules, external mapping, or model pre-classification (i.e., a list of departments that may be relevant before triage according to rules, for reference only by the triage model; the final selection of departments is based on the on-the-job directory), a reason for selection (recording the reasons for the association between the selected department and the combination pattern type, system label, severity, co-occurrence relationship, and longitudinal state change; combination pattern refers to the label of how a group of abnormalities are grouped, and the value is shown in the combination pattern type row; co-occurrence relationship refers to the simultaneous occurrence of multiple abnormalities in the same subject under the same examination background; longitudinal state change refers to the state change of the same indicator between different time points; simultaneous occurrence does not prove a common etiology), selected departments and focus items (recording the departments participating in this analysis and their focus; a complete attending physician analysis is performed on each selected department), and non-selected departments. The following data is required: exclusion reasons (recording all departments not included in the on-duty directory and their exclusion reasons, mutually exclusive with the included records and jointly covering the on-duty directory), operation identifier and roster verification status (related to this operation, the on-duty directory version, the participating roster (i.e., the roster containing the selected departments, i.e., the names of the departments participating in the analysis task), and verification results; model self-reported confidence is not used to replace program verification), unanswered questions (risk issues, evidence gaps, or conflict points that the selected departments should prioritize, without restricting their access to relevant case data), department-by-department execution and result status (related to the task submission, execution, and result status of each selected department; a complete roster or a submitted task does not mean that every anomaly has been analyzed), budget fields (including but not limited to department analysis budget, message quantity and length limits, checkpoint visibility, concurrency limit, retry limit, and deadline), and decision log identifiers (corresponding to at least one of the following: triage, roster verification, task submission, message transmission, source verification, conflict, and timeout or failure records). In some embodiments, the promoter serves three functions: input organization, triage reason tracing, and analysis starting point. Both triage reasons and departmental concerns can be traced back to the original member anomalies and trend events, avoiding a disconnect between the department selection explanation and the actual analysis data.
[0046] In step S12, the computer device obtains the reference candidate department corresponding to the promoter through deterministic candidate recall, and extracts the combined features corresponding to the promoter based on the reference candidate department.
[0047] In some embodiments, determinism refers to the process completed by pre-written program rules without calling the model, with the same input yielding the same result twice. It involves reading the labels (such as system labels, specialty labels, risk area labels, etc.) of member anomalies or trend events in the promoter, and using the external mapping result corresponding to the label as the reference candidate department corresponding to the promoter. Here, the labels are obtained by the program through querying versioned vocabularies, naming rules, keywords, or configuration mappings, and are not generated by the model. The labels are merely triage clues and do not directly prove the disease or final departmental affiliation. External mapping refers to the correspondence between labels and departments given by reference sources other than internal vocabularies (such as terminology tables, coding tables, or external interfaces). The department corresponding to the label of a member anomaly or trend event in the promoter is used as the reference candidate department corresponding to the promoter (i.e., a department worth considering before triage). For example, mapping indicator names to standard codes, and then mapping the codes to the system or department's terminology standards, coding tables, or external interfaces. In some embodiments, for member anomalies or trend events in the promoter, if the terminology is not matched, the original information is retained and explicitly marked. A terminology mismatch means the label of the member anomaly or trend event cannot be found in the versioned terminology, possibly due to different syntax or the terminology not being included; this does not indicate the indicator is normal. A terminology mismatch also means the indicator name cannot be found in the versioned terminology, possibly due to a new item, a non-standard alias, or the terminology not being included; this does not indicate the indicator is normal. Inputs with mismatches retain their original name, observation, direction, and source, and are explicitly marked. The triage model can still select departments from the available list based on the entire batch of cases. In some embodiments, the reference candidate departments are used to assist triage and do not constitute a hard boundary for the final department selection. In some embodiments, truncating to a single label is not allowed. Since a member anomaly or trend event may be associated with multiple system labels simultaneously (e.g., homocysteine is associated with cardiovascular and metabolic systems), not truncating means all these labels are retained and participate in candidate recall, preventing cross-departmental clues from being discarded at the entry point simply because only the first one is selected.
[0048] In some embodiments, the number of anomalies associated with each reference candidate department, severity distribution, number of low-confidence items, cross-system co-occurrence pairs, multi-system labels, trend classification and attention priority distribution, involved indicators, direction of change, time interval, hit status of the combination pattern library (the combination pattern library refers to a pre-configured list of common anomaly combinations or organizational rules, such as classic combinations like simultaneous abnormalities in blood glucose, blood lipids, and blood pressure, which is equivalent to writing a table of experienced doctors' experience that "these things often go wrong together," and a hit only indicates that the set of clues is worth analyzing together) and rule triggers that need attention (triggers are inputs or states that require the system to take action or provide explicit prompts after a certain preset rule is met, such as severity reaching a configured threshold, multiple sources having inconsistent attributions to the same indicator, budget reaching the limit, or the source not being found; whether a hit is achieved is determined by the program by looking up a table, not by the model on the spot; for example, triggers refer to combinations of features such as hitting pre-configured attention rules, anomalies or trend events that need to be explicitly prompted in triage).
[0049] In step S13, the computer device inputs the promoter, the reference candidate department, the combined features, and the directory of departments in service into the triage model to obtain at least one department output by the triage model and the corresponding departmental focus items for the at least one department.
[0050] In some embodiments, the promoter, reference candidate departments, combined features, and the directory of departments in operation are input into the triage model (referring to a large model that selects participating departments from the directory of departments in operation after seeing the entire batch of cases corresponding to the promoter at once). The triage model will select at least one department from the directory of departments in operation and output the corresponding departmental focus items for each of the at least one department. The triage model makes a department selection judgment after seeing the entire batch of cases in the promoter. It does not call the model separately for each item abnormality. When the output verification fails, it will retry according to the preset limit, and does not limit the process to calling the triage model only once.
[0051] In step S14, the computer device establishes an analysis task corresponding to the at least one department based on the promoter and the undergraduate focus item, and submits the analysis task to the specialist agent corresponding to the at least one department.
[0052] In some embodiments, for each of the at least one departments output by the triage model, an independent analysis task is established for that department based on the promoter and the corresponding departmental focus. At least one of the following information is injected into the analysis task: promoter, departmental focus, scope of available tools, budget, and output contract. Then, the analysis task is submitted to the corresponding specialty agent.
[0053] In some embodiments, each department's corresponding specialty agent independently completes a full analysis task, specifying the risk direction, supporting evidence, matters requiring coordination from the overall inspection agent, and uncertainties. Departmental focus items are used to prioritize the analysis task and do not indicate that all anomaly analyses have been completed item by item. In some embodiments, the tool scope, knowledge scope, and context scope of each specialty agent are limited by configuration. For example, the boundaries of specialty agents can be forcibly limited through a search-layer category whitelist or tool whitelist, rather than relying solely on prompts.
[0054] In some embodiments, each specialty agent executes a complete analysis process based on case data, specialty focus items, independent tool configurations, budget ledgers, and output records (the budget ledger refers to a budget record maintained separately for the department's analysis tasks, recording the used and remaining amounts of investigation steps, tool calls, and message quantities) in the starter. Specialty focus items guide the analysis focus and are not used to mask other original indicators of the case. The analysis first conducts an investigation with a limit on the number of steps and time, then plans the mechanism chain (a mechanism chain refers to an evidence-supported inference chain that starts from an abnormal indicator, goes through several inference steps, and points to a certain risk or mechanism; each step in the chain is a loop; loops without source support and only connected by inference are called inference loops) and dependent items (dependent items refer to risk, profile, or statement items that are only valid under the premise of a certain mechanism chain; when the chain exits the current conclusion, these items lose their legitimate dependency and must be withdrawn or downgraded simultaneously while retaining the reason), and the external system (the external system refers to an independent verification context that is separate from the department agent and does not read its initial draft) to perform analysis without reading the initial draft and chain-by-chain verification respectively. Based on the results, a constrained reflection is required. Reflection allows for the retention, deletion, downgrading, or wording modification, but not the addition of new chains or profiles (profiles refer to the structured description layer of the examinee's state based on the results of each chain; statements refer to the textual expression layer for the report), upgrading of supporting states, or alteration of the basis and targets through wording changes. Reflection refers to the process by which, after the specialized intelligent agent has planned its initial analysis draft (mechanism chains and their dependent items) and before batch evidence collection, it compares the draft with two independent analysis results returned by the external system and revises the draft accordingly. The independent analysis results are as follows: (1) The external system independently draws analysis conclusions based solely on the age group, gender, examination date, abnormal indicators and their historical values of the examinee without reading the initial draft, i.e., "analysis without reading the initial draft", and does not provide the external system with identity information such as name and original report text; (2) After reading the mechanism chain truncated to the breakpoint, the external system searches for evidence for each mechanism chain and gives a judgment of support, rebuttal or no evidence found, i.e. "chain-by-chain verification", reflecting that only one round is executed in one analysis task, without looping. Constrained reflection refers to pre-defining the revision actions that can be performed during reflection. The program verifies each revision given by the model one by one before deciding whether to adopt it. The direction of revision can only be tightened, not expanded. Specifically, for mechanism chains, only retention, deletion, downgrading, or wording changes are allowed. For profile entries, only retention, deletion, or wording changes are allowed. The program rejects the following revisions: revisions that point to mechanism chains or profile entries that do not exist in the initial draft (i.e., no new chains or profiles can be added); revisions that change the inference loop to have source support or improve the support status in other ways; revisions that change the basis or member indicators of the mechanism chain when changing the wording, or revisions that change the target, the mechanism chain referred to, the conclusion, or the evidence of the profile entry.If adopting all deletion actions would leave no mechanism chain, then all deletion actions are revoked. Other revisions are processed as usual. Non-compliant revisions are discarded item by item, not entirely. Each adoption or discard is recorded in the decision log. External analysis results with a confidence level below a preset lower limit (e.g., 0.60) are not used as a basis for reflection. Reflection is skipped if external analysis fails or times out, both results are unavailable, or the reflection model output cannot be parsed. The initial draft proceeds to subsequent steps as is; reflection does not cause the analysis task to fail. After revision, the program recalculates breakpoints and truncates mechanism chains. Simultaneously, it deletes profile entries whose attached mechanism chains have been deleted or whose loops fall after breakpoints. Statements whose attached profile entries no longer exist are also deleted. Risk directions only retain the portion corresponding to the still existing mechanism chains. Downgrading refers to not deleting an entry, but lowering its support status or conclusion level so that it no longer appears in the current conclusion with its original certainty, while retaining the original content and the reason for downgrading. Downgrading includes, but is not limited to: (1) changing the ring in the mechanism chain that was originally marked as having source support to an inference ring and clearing the basis of the ring; if two consecutive inference rings appear as a result, the chain is cut off again according to the breakpoint rule; (2) marking the entire mechanism chain as downgraded and removing it from the current conclusion area, while recording the reason, such as the review judgment contradicting the evidence or no supporting evidence after retrieval; (3) changing the conclusion of the statement attached to the downgraded mechanism chain to insufficient evidence; (4) writing the downgraded content as an insufficient evidence item in the report, instead of stating it as a definite conclusion. The difference between downgrading and deletion is that the downgraded entry and its reason are still preserved in the history record and can be presented in the insufficient evidence item or the item requiring doctor review in the report.
[0055] In some embodiments, the specialist agent then collects evidence in batches according to the identified issues, verifies the positive and negative sources (i.e., the sources of literature or tools that constitute support and opposition respectively, both of which must be retrieved), and reviews the conclusions in a separate context (a separate context refers to a new model context that does not carry the previous reasoning records of the main body, but only re-evaluates the support of each conclusion based on the evidence collection results); at most one round of targeted supplementary inspection is conducted when the budget is available and the conditions are met (supplementary inspection refers to the system performing another round of limited evidence collection for specific issues that still lack evidence, not requiring the inspected party to conduct supplementary inspections). In some embodiments, when two consecutive inference loops appear for the first time, the latter is used as the breakpoint, and only the usable part before the breakpoint is retained. In some embodiments, when the review determines that the chain is contradictory, or when there is no legitimate supporting evidence cited by the evaluation results for the chain and its risk, profile, and statement layers, the chain exits the current conclusion area. In some embodiments, entries that lose their legitimate attachment are processed simultaneously and the reasons are retained. The results are delivered to the department only after the literature review and profile backfilling are completed. A definitive conclusion is not forcibly formed when evidence is insufficient. Literature review refers to the process by which the specialized agent judges the relationship between each document retrieved during the evidence collection phase and the statement being examined, and then determines a conclusion for that statement based on this judgment. For documents without abstracts, the research subject is recorded as unknown and treated as a non-human study. For documents with abstracts, the model determines the research subject (human, animal, in vitro, or unknown) and stance (support, support with reservations, opposition, or irrelevant) according to the statement's content. When judged as "support with reservations," the original limiting sentence in the abstract must be extracted. The program then provides conclusions based on pre-defined deterministic rules, using only literature whose subjects are individuals and whose positions are not irrelevant: If there is neither support nor opposition, the evidence is insufficient; if there is only opposition, the evidence is refuted; if there is only support, those with "support but reservations" are considered controversial, otherwise supported; when both support and opposition exist, the proportion of opposition is calculated as (number of opposing articles - number of supporting articles) ÷ (number of opposing articles + number of supporting articles). If it exceeds a pre-defined threshold (e.g., 0.20), the evidence is refuted; otherwise, it is considered controversial. Conclusions are given by program rules, not directly by the model. Literature evaluation also includes assessing the credibility of population risk analysis results from external systems; results with a confidence level below a pre-defined lower limit (e.g., 0.60) are not accepted. Profile backfilling refers to the step where, after literature evaluation, the program writes the conclusions of each statement and its cited literature back to the corresponding profile entry. The profiling entries, during the planning phase, have specified the referred mechanism chain, referred loop, and operating principle, but the conclusion field is empty, indicating that it has not yet been verified. During backfilling, if multiple statements are made under the same profiling entry, the most unfavorable conclusion will be selected in the order of being refuted, controversial, lacking sufficient evidence, and supported, and will be used as the conclusion of that profiling entry. The reference numbers cited in these statements will be deduplicated and used as the basis for that profiling entry. Backfilling only modifies the conclusion and basis fields of the profiling entry; other content remains unchanged.Portrait entries without corresponding statements or incomplete evidence collection are retained as unverified and presented separately in the report; portrait entries whose conclusions are refuted are not included in the conclusion section of the report, but are listed separately along with the refutation evidence as excluded directions.
[0056] In step S15, the computer device aggregates the results status of the specialist agent regarding the analysis task through the general examination agent, and generates an auxiliary health risk analysis report corresponding to the physical examination report file.
[0057] In some embodiments, the overall inspection agent is the general aggregation unit for the result status of various specialty agents regarding the analysis task. It obtains the result status of each specialty agent regarding the analysis task after the specialty agent in each department completes the analysis task or closes according to the deadline (closing refers to the completion of the analysis task or stopping the reception of new content and fixing the result status according to the deadline), and aggregates the result status of each specialty agent. In some embodiments, the responsibilities of the overall inspection agent include cross-system correlation identification (referring to putting clues from different departments together to find possible common explanations and then verifying evidence; simultaneously, the occurrence of clues does not prove a common cause), sorting out duplicate statements in reports, marking conflict scopes, forming overall risk warnings, checking mutual constraints between recommendations (checking the limitations, contradictions, and data gaps between recommendations from different departments to form a consistent and readable explanation), explaining the impact of missing departments such as those not started or failed, and generating auxiliary health risk analysis reports that are traceable to users or doctors, have clear boundaries, visible conflicts, and can be downgraded. In some embodiments, the auxiliary health risk analysis report is a text report output by the general inspection agent and directed to users or doctors. The content includes, but is not limited to: (1) basic information: the examinee's gender, age group, examination date, the departments involved in this analysis and the result status of each department (completed, partially completed, timed out, failed or not started); (2) main positive results: the current value and previous observations of each abnormal item, and the relevant trend timeline listed in chronological order; (3) interpretation of abnormal indicators: explanation of possible risk directions, risk time scales and evidence for abnormal items or combinations, and each conclusion is marked with a citation number that can be parsed to the source; (4) health status analysis: cross-system correlation, sub- The report includes a summary of departmental results and an overall risk warning, retaining the original judgments of each department. Conflicting interpretations are listed in parallel as “different interpretation paths exist”; (5) Items with insufficient evidence: including items that have been withdrawn or downgraded and the reasons therefor; (6) Results of cell-related evidence investigations (if performed): only cite the sources obtained from the investigation and state the boundaries of the evidence, without presenting them as treatment recommendations; (7) Explanation of the impact of missing departments: the impact of departments that have exceeded the time limit, failed, or not started on the completeness of the conclusions; (8) Other suggestions and matters requiring physician review: unanswered questions, evidence gaps, and suggested supplementary examinations or information; (9) List of reference sources: each source is recorded with its title, source, year, and the location of the reference in the text. In some embodiments, the report does not use diagnostic or prescription terminology, and the numerical values are only taken from input data or deterministic calculation results; the source numbers may be renumbered into a reader-readable sequence number according to the order in which they first appear in the text before publication, and all checks are still performed on the original source numbers. In some embodiments, a complete record of the collaborative process (triage reasons, departmental analysis steps, inter-departmental messages, final review records, etc.) may be output as a report attachment or a separate file.
[0058] In some embodiments, the overall review agent does not delete the historical records of departmental conclusions, nor does it rewrite conflicting conclusions into a single, definitive conclusion. In some embodiments, the overall review agent first distinguishes whether the disagreement stems from differing wording or contradictory interpretations, then verifies whether the evidence from each party is true, relevant, and passes the verification. If the evidence is still insufficient, different interpretations are retained, and the facts are not determined by majority opinion. In some embodiments, for conflicts that cannot be resolved, the overall review agent retains them as issues requiring physician review or subsequent examination. For example, it first distinguishes whether the two departments have differing wording or contradictory interpretations. Differences in wording can be addressed, while contradictions in interpretation constitute a conflict. Then, it checks whether the evidence from each party is true, relevant, and passes departmental verification. If the evidence from one party is invalid, it withdraws or is downgraded to insufficient evidence according to the verification rules. If all evidence is valid but the evidence is still insufficient, different interpretations are retained, the disagreement is truthfully stated, and the case is submitted to the physician for review. The facts are not determined by majority opinion, and the historical records are only added to, not deleted. The original conclusions of either party are not deleted.
[0059] In some embodiments, associating and encapsulating multiple items from the abnormal evidence set, abnormal combination, and multi-time point indicator change trends into one or more promoters includes: normalizing the multiple items from the abnormal evidence set, abnormal combination, and multi-time point indicator change trends; associating and encapsulating the resulting unified representation into one or more promoters, wherein the unified representation includes standardized fields. In some embodiments, it is necessary to convert abnormal evidence sets, abnormal combinations, and multi-time point indicator change trends from different sources into a unified representation (i.e., a unified internal object), standardizing fields such as indicator identifier, abnormal direction, severity level, confidence range, source type, time point, state enumeration, and time interval representation.
[0060] In some embodiments, the unified representation result further includes quality tags. In some embodiments, quality tags are data limitation descriptions attached to anomalous evidence sets, anomalous combinations, and multi-time-point indicator change trends, used to indicate where the data is incomplete. Quality tags include, but are not limited to, missing key fields, insufficient confidence, incomplete source location, terminology miss, changes in reference intervals, or insufficient historical values. In some embodiments, tags are transmitted along with the data to the specialist agent and the general agent to determine which content can be compared and which needs to be supplemented. This does not indicate that the data is automatically downgraded or removed. The terminology refers to a pre-configured indicator terminology with version identifiers. Each term corresponds to a test or inspection indicator, recording the standard name of the indicator (such as Chinese name, English name, and abbreviation), as well as various spellings of its name appearing in reports from different institutions (such as standard spelling, colloquial spelling, English spelling, historical spelling, and spellings from different testing systems). It may also record similar indicators that are easily confused but should not be merged, their system labels (one indicator may correspond to one or more system labels), and the version of the labeling rules. The program searches the thesaurus by indicator name. Upon a match, it obtains the standard name and system label of the indicator for deterministic candidate recall. The thesaurus can also include a supplementary table, which adds system labels for indicators not included in the thesaurus based on rules such as indicator name, inspection package, or keywords. A "thesaurus not found" means that no corresponding entry can be found for an indicator name in either the thesaurus or the supplementary table. In this case, the original name, observed value, deviation direction, and source of the indicator are retained and marked; this does not indicate that the indicator is normal.
[0061] In some embodiments, obtaining at least one department output by the triage model and the corresponding departmental focus items for that at least one department includes: obtaining at least one department output by the triage model, the corresponding departmental focus items and reasons for selection for that at least one department, and departments not selected and their corresponding reasons for exclusion. In some embodiments, the triage model may also output the reasons for selection for the at least one selected department and the reasons for exclusion for the departments not selected.
[0062] In some embodiments, the step of establishing the analysis task corresponding to the at least one department based on the promoter and the undergraduate focus items includes: verifying the participation list of the at least one department based on the undergraduate focus items, the reasons for selection, the non-selected departments, and the reasons for exclusion; if the verification is successful, establishing the analysis task corresponding to the at least one department based on the promoter and the undergraduate focus items. In some embodiments, the participation list verification mainly includes checking that the list of selected departments is not empty, that the departments are legal and have no duplicates, that the selection and exclusion are mutually exclusive and that the total covers the directory of on-duty departments, and checking the mandatory fields such as reasons and undergraduate focus items. Among them, the legality of a department means that each department (including the selected departments and the departments in the exclusion table) that appears in the output of the triage agent meets two conditions at the same time: (1) its department identifier can be resolved to one of the department identifiers pre-registered by the system, that is, it is not a model-fabricated or incorrectly formatted identifier; (2) the department belongs to the directory of on-duty departments in this operation, that is, it is in the enabled state. Departments that have been deactivated (e.g., department slots that have been deactivated under a certain directory configuration) and non-existent departments are not legal. In some embodiments, if an invalid department is found, the triage agent's output fails the verification and is retried within a preset number of attempts according to the rules. In some embodiments, if the participation list verification fails, it is retried within a preset number of attempts and a deadline. If it still fails, the triage failure is recorded, and no analysis task is submitted to the department. The department is not automatically reassigned by the rules or the overall review agent. In this application, the triage agent combines abnormal evidence sets, abnormal combinations, and multi-time point indicator change trends to select multiple participating departments from the on-duty directory and provides reasons for selection, departmental focus items, and reasons for exclusion of non-selected departments. It checks the legality and completeness of the selected and excluded lists. If an invalid list is found, only a limited number of retries are allowed. If it still fails, the triage is terminated and a record is kept to avoid mistaking automatic reassignment or a complete list as the completion of the analysis.
[0063] In some embodiments, the step of inputting the promoter, the reference candidate department, the combined features, and the directory of departments currently in service into the triage model to obtain at least one department output by the triage model and the corresponding departmental focus items of the at least one department includes: inputting the promoter, the reference candidate department, the combined features, the directory of departments currently in service, and a preset team size upper limit into the triage model to obtain at least one department output by the triage model and the corresponding departmental focus items of the at least one department, wherein the number of departments corresponding to the at least one department is less than or equal to the preset team size upper limit.
[0064] In some embodiments, submitting the analysis task to the specialist agent corresponding to the at least one department includes: submitting the analysis task to the specialist agent corresponding to the at least one department through an execution queue, wherein the execution queue has a preset concurrency limit, which is used to control the number of analysis tasks running at the same time. In some embodiments, all analysis tasks are submitted to the specialist agent through the execution queue, the execution queue has a preset concurrency limit, and the concurrency limit only controls the number of analysis tasks running at the same time.
[0065] In some embodiments, the method further includes: for each specialist agent in the at least one department, at the model call checkpoint corresponding to that specialist agent, sending the collaborative message generated by the specialist agent to specialist agents in other departments via the central scheduling module, wherein the collaborative message includes discovery type messages. In some embodiments, the central scheduling module is the sole relay point for communication among multiple specialist agents. Each department undertakes a complete analysis task, and direct connections between departments are not allowed, nor is it allowed to bypass the central scheduling module to read the unfinished content of other departments. In some embodiments, the central scheduling module is responsible for executing message budgets, concurrency limits, deadlines, and exception records. Departments send or receive collaborative messages through the model call checkpoints of the existing models of the specialist agents (the checkpoint refers to the point in time before each model call by the specialist agent during the analysis process, at which the central scheduling module writes the message to be injected into the prompt word of this call according to the upper limit of visibility), without setting a separate fixed round of discussion phases. In some embodiments, the content of a collaborative message includes at least one of the following: message identifier, object identifier, task identifier, associated initiator, sending department, receiving object, current stage, message type, parent message identifier, relevant indicators, source number, creation time, and processing status. In some embodiments, the collaborative message includes a discovery type message, which refers to clues related to a case and their source numbers (the source number refers to a number uniformly assigned to each source returned by the tool during this run; the message only references the number, and the receiving department must verify that it matches the same source in the tool's actual return before it can be adopted). The central scheduling module of this application can improve the controllability of collaboration. Discoveries, problems, and responses between departments are all relayed through the central scheduling module, which uniformly checks the message length, quantity, response relationship, and checkpoint visibility range, and records the message publication, actual injection, and source verification status.
[0066] In some embodiments, the collaboration message also includes a question type message. In some embodiments, the question type message refers to a question that requires attention or response from other selected departments.
[0067] In some embodiments, the collaborative message further includes a response type message regarding question-type messages received by the specialty agent from other specialty agents sent via the central scheduling module. In some embodiments, the response type message refers to a response to existing question-type messages from other departments, and the response type message needs to be associated with a message identifier of the question-type message it responded to (i.e., the parent message of the response type message). In some embodiments, the response type message is associated with existing question-type messages through the parent message identifier and must not respond to questions raised by the department itself. In some embodiments, the department explains its judgment, basis, and uncertainties regarding the question-type messages in the response type message. Disagreements and unresolved matters are reserved for review by the chief inspection agent. The program saves this information according to the association between questions and responses and the structured results, without requiring mandatory consensus as the condition for the department's task to end. In some embodiments, the collaborative message is published when the department's task has not yet closed, there is still a budget, and the deadline has not been exceeded, and is injected at the upper limit of visibility at the receiving department's existing model call checkpoints. In some embodiments, discovery-type messages, question-type messages, and response-type messages are used to provide clues, raise questions, and respond to existing issues, respectively. A message is only recorded as injected after it actually reaches the receiving department's prompt. When the model declares adoption (adoption means the receiving department uses the content of a message in its analysis or conclusion, declared by the model in its output and registered by the program), the message source number must be non-empty and all can be verified in the actual tool returned by the receiving department; otherwise, it is recorded as pending verification. The message number itself is not considered original evidence, and source verification does not equate to a confirmed medical conclusion. In some embodiments, after a department's corresponding specialist agent completes its analysis task, its message reading, writing, and adoption registration are frozen. Late messages do not restart the process. Unresolved issues are addressed to the chief agent along with the message and the status of the analysis task results, without being discussed again in the current closed task. This application improves the identifiability of communication through three types of lightweight messages: discovery, question, and response. Responses must be associated with existing issues, and the source in the message must be verified against the actual tool returned by the receiving department. The message number itself is not considered independent evidence, and successful source location does not equate to the medical interpretation being verified. In this application, messages, tool steps, and time budgets jointly define the scope of collaboration. It does not require all departments to wait for all questions to be answered. Once a department's task is closed, message reading, writing, and acceptance will cease. Late messages will not restart the task. Unanswered questions will be treated as gaps and entered into the general inspection agent for verification.
[0068] In some embodiments, the total number of collaborative messages generated by the specialist agent corresponding to each department is less than or equal to a preset threshold. In some embodiments, the total number of collaborative messages that the specialist agent corresponding to each department can generate in this analysis task is limited, and its total number must be less than or equal to the preset threshold. In some embodiments, discovery type messages, problem type messages, and response type messages are all included in the total message volume of each department, and responses do not receive an additional unlimited sending quota. In some embodiments, if the number of departments selected this time is N and the message limit per department is M, then the total number of inter-department collaborative messages in this time will not exceed N×M.
[0069] In some embodiments, the result status includes the task result obtained by the specialist agent after completing the analysis task.
[0070] In some embodiments, the result state also includes the current state fixed by the specialized agent after the cutoff condition is met. In some embodiments, "cutoff" refers to the completion of the analysis task or the cessation of receiving new content according to the cutoff condition and the fixing of the current state.
[0071] In some embodiments, the aggregation of the result status of the specialist agents regarding the analysis task by the overall inspection agent to generate an auxiliary health risk analysis report includes: performing a subject-by-subject check on the result status of the specialist agents regarding the analysis task by the overall inspection agent, aggregating the target result statuses that do not contain preset hard errors, and generating an auxiliary health risk analysis report. In some embodiments, the overall inspection agent performs a subject-by-subject check on the result status of each specialist agent regarding its analysis task, checking the source of data, the existence of members and targets (targets refer to indicators, organs, pathways, or molecular objects pointed to by the mechanism chain), the resolvability of citations, the wording of diagnoses or prescriptions, the direction of anomalies and trends, the risk time scale, and the unit. Here, a member refers to a member indicator of the mechanism chain, that is, the indicator corresponding to the anomaly or trend event that the mechanism chain uses as its starting point and as its basis in each loop. Checking the existence of members means that the member indicators of each mechanism chain must be found in the current input anomaly, the sequence of previous observations, or the trend event, and cannot be indicators that do not exist in the input data. In this context, the existence of the target point verification means that the mechanism chain pointed to by the profile entry must exist, the sequence number of the pointed-to loop must be within that chain, and the profile entry to which the statement is attached must exist. In some embodiments, the overall inspection agent aggregates target result states that do not have preset hard errors in the result states to generate an auxiliary health risk analysis report. Hard errors refer to the failure of any of the aforementioned verification items, and the specific items are specified by a pre-set configuration. In some embodiments, the overall inspection agent reorganizes and repeats the descriptions of the target result states, but does not merge the original mechanism chains and results of each department, and still needs to verify the values, references, and output structure. In some embodiments, the final report (i.e., the auxiliary health risk analysis report) will only be issued if at least one department passes the hard checks.
[0072] In some embodiments, the step of performing a department-by-department check on the result status of the specialist agents regarding the analysis task through the overall inspection agent, and aggregating the target result statuses without pre-set hard errors to generate an auxiliary health risk analysis report, includes: performing a department-by-department check on the result status of the specialist agents regarding the analysis task through the overall inspection agent, and performing a cell-related evidence investigation on the result statuses, aggregating the target result statuses without pre-set hard errors and with clear evidence boundaries to generate an auxiliary health risk analysis report. In some embodiments, the overall inspection agent also performs a cell-related evidence investigation on the result status of each specialist agent regarding the analysis task, with the source stored independently, without rewriting the original departmental conclusions, and the relevant content is only included in the final report (i.e., the auxiliary health risk analysis report) when there is a corresponding source and clear evidence boundaries, and the mechanism hypothesis is not presented as a treatment recommendation. In some embodiments, cell-related evidence investigation refers to the evidence investigation that the overall inspection agent performs uniformly on all departments once after checking each department and before summarizing, using the read-only view of the result status of each department as input, to find out whether there is a direction in the pathological link pointed to by the mechanism chain of each department that cell therapy (such as mesenchymal stem cells, exosomes, hematopoietic stem cells, natural killer cells and other cell types available in outpatient clinics) may intervene, and what evidence is currently available for that direction. The steps are as follows: (1) The program combines the hypothesis, member indicators, statements of each link and risk direction of each mechanism chain into text, matches it with the pre-configured disease terminology, and obtains candidate directions in the form of "disease × cell type"; only one direction is retained for the same combination of disease and cell type, and the number of directions is set to an upper limit (e.g., 5); the terminology entries can be set to match only the indicator name, match only when the indicator is high, etc., to avoid the statement in differential diagnosis or the low indicator triggering the direction by mistake. (2) Generate a search query for each direction and search in sources such as public literature databases, local stem cell literature databases, and clinical trial registry databases. The search results are filtered for errata or retraction, exclusion of treatment type (e.g., CAR-T, gene therapy, etc.), and human research. (3) The model reviews the literature retrieved round by round to see if it truly supports the use of the cell type for the disease. If there is no supporting literature, a new search query is given to enter the next round until supporting literature is found, no new search query is found, the maximum number of rounds or the maximum time limit is reached. (4) Mark the evidence level for each direction, such as approved indication, under investigation (with matching clinical trials in the registry database), human research, only mechanism exploration, or insufficient evidence. The evidence level is only marked and the direction is not deleted based on it. (5) Optionally, an independent exploration agent can be assigned to each direction, which can choose to search for literature, call tools, or use quantitative analysis in this case to supplement evidence, and then write an explanatory paragraph that only cites the source number obtained.The investigation results and sources should be kept independently and should not be rewritten from the department's original conclusions. In the final report, cell-related content can only cite the source numbers obtained in this investigation and specify the boundaries of the evidence; it should not be presented as a treatment recommendation.
[0073] In some embodiments, multiple specialized agents employ nine system-level directories (i.e., nine system directories): nervous system, immune system, cardiovascular system, metabolic system, musculoskeletal system, skin or epidermal system, reproductive and endocrine system, respiratory system, and digestive system. In some embodiments, the nine system directories, indicator mappings, and cross-system attribution rules (cross-system attribution rules refer to a pre-configured lookup table recording the related systems and common co-occurrence patterns when the same indicator is tagged with multiple systems) are saved version by version. The indicator mapping refers to a pre-configured lookup table recording the correspondence between each indicator and one or more systems in the nine system directories; for example, mapping fasting blood glucose to the metabolic system, and mapping homocysteine to both the cardiovascular and metabolic systems. The program assigns system tags to anomalies based on this table and then uses it for deterministic candidate recall. It is saved version by version along with the cross-system attribution rules. In some embodiments, the nine system directories are a preferred configuration; however, the invention is not limited to hospital departments, health management risk domains, chronic disease topics, or other specialized directories. In some embodiments, another department directory is configured as follows: cardiology, endocrinology and metabolism, urology and nephrology, gastroenterology (including hepatobiliary), respiratory medicine, ENT, gynecology (including breast), orthopedics (including rehabilitation and sports medicine). This nine-department directory is configured separately from the above-mentioned nine physiological system directories.
[0074] In some embodiments, the execution of the above method further includes calling a shared state module that only adds and does not delete. The shared state is used to store the initiator, participant roster, messages, the result status of the patent agent corresponding to each department, conclusions, conflict annotations, anomaly records, and aggregated results. The historical record layer of this state adopts an add-only strategy; any unit can only add conclusions, annotations, or statuses, and cannot delete or rewrite content already written by other units. If an error is found in the conclusion or subsequent evidence changes, only a correction annotation or a new version conclusion can be added, and this is associated with the version field and decision log. In one implementation, the shared state uses an append-only event table. The write interface only provides operations for adding events, not for updating or deleting existing events. Correction events store the identifier of the corrected event, and conflict events simultaneously store the identifiers of conflicting conclusions. During querying, the current version can be generated through a materialized view, but the original event remains in the append-only event table. This application retains conflict facts through an add-only historical shared state, ensuring that conclusions from different specialties are not silently overwritten, and that conflicts are explicitly annotated and their evidence is retained, facilitating review by doctors or subsequent processes.
[0075] In some embodiments, the historical record and the currently available conclusion view are maintained separately. Reflection or verification can retract or downgrade the current entry, and simultaneously address the risks, profiles, and statements that have lost their legitimate basis. However, the original version and the reasons for the disposal should be retained. Not deleting the historical record does not mean that invalid entries must continue to be published as valid conclusions. The currently available conclusion view refers to the set of conclusions that are still valid, filtered from the historical record (which only adds entries) according to the latest version and status of each entry. For example, it could be the current version generated from an append-only event table through a materialized view. The historical record stores all versions, change actions, and reasons for changes for each entry. The currently available conclusion view only includes entries whose latest versions have not been retracted. Downgraded entries appear in the view in their downgraded state or are transferred to insufficient evidence items. The conclusion section of the report is generated based on the currently available conclusion view. Retracted entries are only retained in the historical record and can be described in the report as insufficient evidence items or excluded directions. In one implementation, the retracted mechanism chain is removed from the current conclusion set and recorded in the downgrade chain list, along with the reason for retraction. Both are stored together in the record of this operation. Among them, verification refers to various verifications that can change the current state of an item, in addition to reflection, including: (1) Independent review within a specialist intelligent body: reviewing each conclusion in a separate context and solely based on the evidence collection results, and removing mechanism chains that are judged to be contrary to the evidence or have no legal supporting evidence from the current conclusion area; (2) Rule verification of program execution: for example, determining the breakpoint and cutting off the mechanism chain according to two consecutive inference loops, changing the conclusion marked as having support but without providing any supporting evidence to having no support, and removing items whose citations cannot be resolved to the source; (3) Department-by-department verification of the general inspection intelligent body (i.e., the general practice stage A verification in the disclosure document): checking items such as the source of data, the existence of members and targets, the citations being resolvable, and the wording of diagnoses or prescriptions, etc. After a departmental result with hard errors is returned at most once under restricted conditions, the program will withdraw the problematic mechanism chain or change the relevant statement to insufficient evidence. Restricted returns only allow withdrawal or downgrading, and do not allow rewriting of the conclusion text. The current entry refers to the entry that is currently in the available conclusion view when performing a reflection or verification. This includes the mechanism chain and its links, as well as the risk directions, profile entries, and statements attached to the mechanism chain. Retracted entries are no longer current entries, but their historical versions are still retained. A legitimate attachment means that the attachment relationship between the attached entry and its attached object is still valid: the mechanism chain to which the attached entry is attached is still in the currently available conclusion view, the link it points to is still in the available portion before the chain breakpoint, and the profile entry to which the statement is attached still exists. For example, a profile entry points to one or more links of a mechanism chain through a mechanism chain identifier and a link number; a statement is attached to a profile entry through a profile entry identifier; and a risk direction is attached to the mechanism chain from which it originates.When a mechanism chain is withdrawn or downgraded and then truncated so that the loop pointed to by the profile entry falls after the breakpoint, the corresponding dependent entry loses its legitimate dependency. The program simultaneously withdraws or downgrades these entries and records the reason: the profile entry is withdrawn along with the mechanism chain it is attached to, the statement is withdrawn along with the profile entry it is attached to, and the risk direction only retains the part corresponding to the still existing mechanism chain; if all profile entries are withdrawn, the reason for withdrawal is stated, without leaving any blanks without explanation.
[0076] In some embodiments, complete results from each department are stored independently, and target and source numbers are isolated by department; source deduplication retains the original number, the department using it, and the reference location (source deduplication refers to identifying the same source referenced by different departments as a single record during the final review), and the mechanism chains of different departments are not merged due to identical chain members. In some embodiments, when multiple departments give different judgments on the same promoter or the same anomaly, the system retains multiple conclusions and their judgments, basis, time, and sources, without overwriting later writes with earlier writes, and without forcing the generation of superficial consensus.
[0077] In some embodiments, the execution of the above method further includes a call budget, timeout degradation, and decision log module. The budget module sets the triage retry limit, department concurrency, total external call concurrency, total number of messages per department, number of visible messages at checkpoints, tool call budget, number of investigation steps, department task deadline, and overall checkup aggregation timeout. All selected department tasks are submitted; the concurrency limit only determines the number of tasks executed simultaneously, and other tasks are queued and not discarded due to full concurrency.
[0078] Once the message budget is exhausted, no new collaborative messages will be published. Existing valid results and budget exhaustion statuses will still be submitted to the chief inspector for verification. When a department's task times out, existing materials will be retained, and the timeout and missing items will be clearly recorded. Departments whose tasks have not started by the time they are closed will be recorded as not started and will not be counted as completed. Unanswered questions will be submitted to the chief inspector's AI agent based on their status; closed department tasks will not be extended while waiting for a response. If the triage output is invalid or the model is unavailable, retrying will be performed within the configured limit. If it still fails, a triage failure will be recorded, and a reassignment roster will not be automatically generated. When the chief inspector times out, completed content, incomplete items, and inspection records will be saved.
[0079] The decision log records triage selection and exclusion, roster verification and retries, task submission and queuing, message publishing and injection and source verification, departmental result status, conflict annotation, budget exhaustion, timeout sealing, and overall agent verification and publishing judgment. Each log entry includes at least an input snapshot identifier, trigger rule, original judgment, selected action, reason, operator or module, timestamp, and associated initiator identifier. An input snapshot is a read-only record that captures the input content and conditions upon which a decision is based at the moment it is made. The input snapshot identifier is an identifier pointing to the record, such as a record number or a summary value calculated from the input content. The content of the input snapshot varies depending on the decision type. For example, the input snapshot for triage decisions may include initiator content, reference candidate departments, combined features, and the version of the on-duty department directory; the input snapshot for roster verification may include the original output of the triage model and various verification results; the input snapshot for message injection may include the list of messages visible at that time and the maximum visibility limit; the input snapshot for budget exhaustion or timeout processing may include the used amount, limit, and time at the time of triggering. Each run can also record the program version used, configuration switches, and model parameters. The criterion for judging whether the input snapshot is complete is whether the decision can be replayed based on the snapshot, so that the decision log can be used for auditing, playback, and quality statistics. This application improves the feasibility of the project through timeout degradation and decision logs. When a department fails, the budget is exhausted, or the model is unavailable, the existing content, completion status, and degradation reasons are retained. The final auxiliary report must still meet the release conditions of the overall inspection agent. When there is no department result status without preset hard errors, only auditable records are retained, and the final report is not released.
[0080] For example, a health assessment system has generated a set of structured abnormal evidence, a "metabolic abnormality combination," and a "hepatobiliary abnormality combination," and provides the trend of the same indicator at multiple time points for the same subject, such as fasting blood glucose being within the reference range at a previous examination time point and exceeding the reference range at a later examination time point. The system receives these three types of core inputs, performs standardization and quality labeling, and encapsulates relevant abnormalities, combinations, and normal-to-abnormal trend events as promoters. Deterministic candidate recall identifies departments such as metabolism, cardiovascular, and gastroenterology; combination features retain the co-occurrence of multiple metabolic indicators, severity distribution, cross-system co-occurrence relationships, and the direction and time interval of fasting blood glucose turning from normal to abnormal. The triage agent reviews the entire batch of cases and the directory of departments on duty, selects relevant departments, and provides reasons for selection and departmental concerns for each department, while providing reasons for exclusion for departments not selected. After verifying the roster, a complete analysis task for all selected departments is submitted. The metabolism department, in its independent analysis, finds the need to verify the impact on cardiovascular risk and raises questions to the cardiovascular department via the central scheduling module. Upon receiving a question at the checkpoint, the cardiology department responds that there is currently a lack of blood pressure or past medical history information, making a definitive judgment impossible. This response is recorded along with the source verification results. Each department continues its own analysis, without indefinitely extending the task by waiting for responses to all questions. After the department closes its task, the overall intelligence system aggregates the results, completion status, and unanswered questions from all departments, retaining metabolic risk interpretations and gaps in cardiovascular evidence, without forcibly merging them into a definitive cardiovascular risk conclusion. Once verification is completed and the release conditions are met, an auxiliary health risk analysis report with source location and supplementary information prompts is output.
[0081] For example, if a structured outlier does not match the versioned vocabulary, the name, observation, direction of deviation, source, and co-occurrence anomalies of the outlier are retained, and the vocabulary is marked as out of range. If the reference candidate is empty, the original indicator is not deleted, and the triage agent is not restricted from selecting relevant departments from the on-call list based on the entire batch of cases. The triage agent output must list the selected departments and their focus items, as well as the departments not selected and the reasons for exclusion. If the roster contains illegal departments, omissions, or inconsistencies, a retry is required within a predetermined limit; if the retry is still invalid, the triage is recorded as failed and the current process is terminated, without automatic department replacement by rules or the overall review. The decision log retains the vocabulary outlier marker, input snapshot, triage output, roster verification, and retry results. When the roster is approved, the selected departments analyze the case according to the complete process, retaining relevant evidence gaps in the results. The approval of the roster itself does not indicate that the outlier has been analyzed completely.
[0082] For example, in a complex health risk analysis, multiple selected departments may wish to exchange findings or raise questions. The central scheduling module checks the number of messages, message length, and visible checkpoints for each department according to its configuration. Once the limit is reached, excess messages are rejected and the reason is recorded. Departments continue to complete their independent analysis within the remaining tool steps and time budget. Running out of message budget does not automatically equate to task failure. If a department fails to complete the task before the time limit, the timeout and existing content are recorded, and the task is closed. After closure, message reading, writing, and acceptance by that department are stopped. Late responses do not restart the task. The overall review agent collects completed results, timeout status, and unanswered questions, explains relevant evidence gaps, and outputs the final auxiliary health risk analysis report.
[0083] For example, one participating department might believe that existing evidence favors endocrine risk for a particular promoter, while another might support short-term lifestyle influences. Both departments complete their analyses and provide structured results and supporting evidence, or they can explain their differences through findings, questions, or responses. The shared state preserves the original results from both parties and records the objects, scope, and basis of the conflict; if invalid entries are withdrawn during subsequent verification, their historical records are still retained. When the overall review agent aggregates data, it does not forcibly eliminate the conflict but instead describes it as "different interpretive paths exist," providing suggested reconciliation results, such as suggesting supplementary medical history, follow-up examination time points, or relevant indicators. This approach preserves the differences in multi-specialty judgments while avoiding packaging uncertainties as definitive conclusions.
[0084] In some embodiments, the system sets up nine system-level sub-agents, corresponding to the nervous system, immune system, cardiovascular system, metabolic system, musculoskeletal system, skin or epidermal system, reproductive and endocrine system, respiratory system, and digestive system, respectively, with one sub-agent for each system. A versioned mapping table records the correspondence between anomalies and one or more systems, and cross-system attribution rules record common co-occurrence patterns. The cross-system attribution rules are a pre-configured lookup table that records the relevant systems and common co-occurrence patterns when an indicator is simultaneously tagged with multiple systems. When a case involves both metabolic and cardiovascular issues, the triage agent can select the corresponding system-level department and provide the relevant departmental focus items for each, both of which undergo complete analysis. Items not selected must be listed in the exclusion list with reasons; if the list is invalid, there is a limited number of retries, and the process terminates if it still fails. This nine-system directory is only a subordinate configuration and does not require all of them to be selected each time; it can also be replaced with a hospital specialty directory or an institution-defined risk domain.
[0085] In some embodiments, the upstream system provides structured states of the same object and the same indicator at different time points. The system receives or generates a change event in which the indicator changes from normal at a previous time point to abnormal at a subsequent time point, and encapsulates this multi-time-point indicator change trend as a core input at the same input level as structured abnormal evidence and abnormal combinations in the promoter. This trend event includes at least the object identifier, indicator identifier, previous and subsequent time points, previous and subsequent observations or qualitative conclusions, previous and subsequent reference ranges or judgment criteria, previous and subsequent states, state change type, time interval, change direction, and source location. The normal-to-abnormal trend event is retained in the combined features and used for triage, department selection, departmental focus items, and unanswered questions. When the overall review agent aggregates, it retains the timeline of the event, departmental conclusions, and evidence gaps, and prompts doctors to review it in conjunction with the original data. If historical data is missing, the trend state is recorded as undecidable, but subsequent processes can still be executed based on the other two types of core inputs.
[0086] In some embodiments, upstream or extended modules can perform subject association, time alignment, unit unification, and source tagging on health data from different sources, and submit the unified anomalous evidence to the input interface. Upon receipt, the input interface processes the anomalous evidence as structured anomalous evidence, rather than directly unifying the multi-source data. If there are conflicts between the multi-source data, the input interface can use the conflict status as a quality marker or source marker for the anomalous evidence; subsequent initiator encapsulation and overall inspection agent aggregation can incorporate this marker into the unanswered questions and conflict annotations.
[0087] Figure 2 This diagram illustrates a method for generating an auxiliary health risk analysis report according to one embodiment of the present application.
[0088] like Figure 2As shown, the system receives three types of core inputs: structured sets of abnormal evidence, combinations of abnormalities, or trends in indicators at multiple time points are received via the input interface. The system records the object identifier, task identifier, source of evidence, source location, and input batch. These three types of objects are at the same core input level and can enter the process individually or in combination. For trends in indicators at multiple time points, the system prioritizes receiving events where the same indicator for the same object changes from normal at one time point to abnormal at another. The system standardizes and marks quality status: it uniformly represents the indicator name, standard code, abnormal direction, severity, confidence level, preceding and following time points, preceding and following states, time interval, and source type for abnormal items and trend events. Quality marks are retained when fields are missing, have low confidence, or the source is incomplete. These marks are transmitted with the data to the triage agent and the overall review agent to explain the limitations of the corresponding conclusions. When the terminology is not matched or cross-system attribution is uncertain, the corresponding triage clues and matters requiring attention are retained; the input is not directly discarded. Encapsulate anomalous evidence combination promoters: Associate the three core inputs into one or more promoters, supplementing triage and collaboration control information (triage information corresponds to reference candidate departments, reasons for selection, selected departments and concerns, non-selected departments and reasons for exclusion, operation identifiers, and roster verification status; collaboration control information corresponds to unanswered questions, departmental execution and result status, budget fields, and decision log identifiers), and retain the source relationship of the original evidence. The promoters associate member anomalies and trend events, trend priorities, co-occurrence summaries, combination patterns, candidate clues, unanswered questions, and budgets, and subsequently associate selected departments, reasons for selection, concerns, exclusion records, operation identifiers, and departmental result statuses. Deterministic candidate recall: Recall all reference candidates for each anomaly and trend event from versioned vocabularies, external mappings, specialty directories, and cross-system rules (cross-system rules refer to pre-configured rules specifying which related departments should be recalled as reference candidates when an anomaly occurs), without truncating tags due to field order. Inputs that do not match the vocabulary list are retained and marked, and the triage model can still select departments from the on-duty directory based on the entire batch of data. Combined features are extracted: the number of outliers, severity distribution, system label distribution, number of low-confidence items, co-occurrence pairs, trend classification and priority distribution, involved indicators, direction of change, time interval, and combination pattern hit rate are statistically analyzed, and rule triggers that require attention are retained. Dynamic triage of the entire batch of input: The triage model reads the entire batch of input, candidate clues, combined features, and the on-duty directory, selects 1 to K participating departments, outputs the reasons for selection and the department's focus items, and explains the reasons for exclusion of the remaining on-duty departments. Roster verification and limited retries: The model checks that selected departments are not empty, have the maximum number of departments, are legal and unique, have complete reasons, and that selection and exclusion are mutually exclusive and cover the on-duty directory. If verification fails, retries are only made within a preset number of attempts and a deadline; if it still fails, the triage failure is recorded and the process ends, without automatic reassignment or initiation of subsequent department tasks.Departmental Task Submission and Concurrent Scheduling: Each selected department submits a complete analysis task, providing case data, departmental focus areas, tool scope, budget, and output contract (the output contract refers to the structured format requirements for the department's results, including fields, status enumerations, and source number formats; the program uses this to verify whether the department's output can be identified). Concurrency limits control the queuing order and the number of simultaneous runs, ensuring no reduction in the number of selected departments. Complete Departmental Analysis: Each selected department performs a complete analysis, including budget-constrained investigations, analysis planning, source verification, and constrained reflection, resulting in structured results. Departments conduct analyses based on approved case data and real tool feedback, without bypassing central scheduling to read unpublished drafts from other departments. Checkpoint Bounded Message Exchange: During departmental analysis, checkpoints within existing models exchange three types of messages via central scheduling: findings, problems, and responses. The program limits the number, length, visibility, and response relationships of messages; sources are only adopted according to rules after verification by real tools. The system does not wait for all questions to be answered; unanswered matters are left to the chief inspector. After a department's task ends, its communication is frozen, and late messages do not restart the task. The system collects results by department and retains historical status: it collects the result status of each selected department, adding to historical records without deletion. Currently valid conclusions can be withdrawn or downgraded according to constrained rules, and dependent items are cleaned up simultaneously. Different interpretations and their basis for each department are retained for verification by the chief inspector's AI. Chief Inspector Verification and Aggregation: After a department's task closes, the chief inspector reads the participant roster, the result status of each department, source mapping (source mapping refers to the corresponding record of the source number and its content, the department using it, and the reference location), messages, and change records. It performs a maximum of one restricted return and review within the specified scope for each department and can perform team-level evidence investigations within the specified scope. The chief inspector retains the original chain and disagreements of each department and summarizes cross-system associations, risk warnings, unanswered matters, and evidence gaps. It outputs an auxiliary health risk analysis report: At least one department's result status passes a hard check, the report text is not empty, and the process has not been canceled. The report includes participating departments and the status of each department's results, a timeline of relevant trends (a longitudinal overview of the status changes of each indicator in chronological order), main risk directions, departmental results summaries, conflicts and insufficient evidence items, the impact of timeouts or failures, and matters requiring physician review. It writes to the decision log and demotion record: triage input, roster verification, limited retries, task submission, message publishing and injection, source verification, conflicts, budget exhaustion, timeouts or failures, final review, and publication judgment are written to the decision log, supporting auditing, playback, and quality statistics. The above steps are executed according to dependencies. Departmental tasks are submitted only after roster verification is passed. During departmental analysis, checkpoint messages can be exchanged intermittently. Results are compiled and finalized after the task is closed. Departments only exchange findings, problems, and responses; there is no separate discussion round where all departments wait. Unanswered issues are submitted to the final review agent for explanation along with the results.
[0089] Figure 3This diagram illustrates a system block diagram for generating an auxiliary health risk analysis report according to one embodiment of the present application.
[0090] like Figure 3 As shown, the core input module receives structured anomaly evidence sets, anomaly combinations, or multi-time-point indicator change trends of the same indicator for the same object submitted by external systems. The input normalization and promoter encapsulation module normalizes these data and converts them into a unified internal object. This unified internal object is then encapsulated into a promoter, retaining the original source, quality constraints, combination relationships, and unanswered questions. The triage and roster verification module selects departments from the on-the-job directory, recording the reasons for selection, departmental focus items, and exclusion reasons. The central scheduling module submits all selected tasks to the corresponding specialist agents in each department. Tasks are queued according to the concurrency limit, with execution budget and timeout control implemented. Each selected department operates independently. The entire treatment process is executed. Each specialty agent exchanges messages about discovery type, problem type, and response type through the central scheduling module. Messages are read at existing call checkpoints. There is a message limit for each department. Messages are frozen after the department ends. The results of each department are collected and archived by the final inspection module. Stage A verification (department-by-department verification) is performed. You can enter the optional stage B (cell-related evidence investigation). Finally, the results are summarized. Only after the inspection is released will the final auxiliary report be generated. Otherwise, the reasons for failure, missing or canceled results and inspection records are retained. The entire process is traced and the source snapshot is added. Historical versions are retained. Invalid conclusions can be withdrawn. Unresolved disagreements are retained.
[0091] Figure 4 This application illustrates a system for generating auxiliary health risk analysis reports (i.e., an auxiliary health risk analysis report system) according to one embodiment of the present application. The system includes a first module, a second module, a third module, a fourth module, and a fifth module.
[0092] The first module is used to associate and encapsulate multiple factors from the abnormal evidence set, abnormal combination, and multi-time point indicator change trend into one or more promoters. The abnormal evidence set, abnormal combination, and multi-time point indicator change trend are obtained based on medical examination report files. Each promoter includes at least one of the following: member anomaly item, trend event, combination pattern type, and co-occurrence anomaly summary. Related operations are... Figure 1 The embodiments shown are the same or similar, so they will not be described again, but are included here by reference.
[0093] The second module is used to obtain reference candidate departments corresponding to the promoter through deterministic candidate recall, and extract combined features corresponding to the promoter based on the reference candidate departments. Here, the related operations are... Figure 1 The embodiments shown are the same or similar, so they will not be described again, but are included here by reference.
[0094] The third module is used to input the promoter, the reference candidate departments, the combined features, and the directory of departments currently in operation into the triage model, and obtain at least one department output by the triage model and the corresponding departmental focus items for the at least one department. Here, the related operations are... Figure 1 The embodiments shown are the same or similar, so they will not be described again, but are included here by reference.
[0095] The fourth module is used to establish analysis tasks corresponding to the at least one department based on the promoter and the undergraduate focus items, and submit the analysis tasks to the specialty agent corresponding to the at least one department. Here, the related operations are... Figure 1 The embodiments shown are the same or similar, so they will not be described again, but are included here by reference.
[0096] The fifth module is used to aggregate the results and statuses of the specialist agents regarding the analysis task through the general examination agent, and generate an auxiliary health risk analysis report corresponding to the physical examination report file. Here, the related operations are... Figure 1 The embodiments shown are the same or similar, so they will not be described again, but are included here by reference.
[0097] Figure 5 The diagram illustrates a computer device structure for generating an auxiliary health risk analysis report according to an embodiment of this application. The device includes module 11, module 12, module 13, module 14, and module 15.
[0098] Module 11 is used to associate and encapsulate multiple items from the abnormal evidence set, abnormal combination, and multi-time point indicator change trend into one or more promoters. The abnormal evidence set, abnormal combination, and multi-time point indicator change trend are obtained based on the physical examination report file. The promoter includes at least one of the following: member abnormal item, trend event, combination pattern type, and co-occurrence abnormal summary.
[0099] In some embodiments, the input is not the medical examination report file itself, but rather multiple core inputs obtained from the medical examination report file at the same input level: a structured set of abnormal evidence already generated by the upstream module, abnormal combinations formed by organ systems, clinical problems, or co-occurrence relationships, and the trend of changes in the same indicator at multiple time points for the same object. Each piece of structured abnormal evidence includes at least the name or standard code of the abnormal item, the current result, the unit, the reference range or judgment basis, the direction of deviation, the source of the abnormality, the severity, the confidence level, the source location, and the identifier of the object to which it belongs; each abnormal combination includes at least the member abnormal items, the combination label, the source range, and the quality mark; each trend of changes in the indicator at multiple time points includes at least the object identifier, the name or standard code of the indicator, the previous time point, the next time point, the previous and next observations or qualitative conclusions, the previous and next reference ranges or judgment basis, the previous state, the next state, the type of state change, the direction of change, the time interval, and the source location, wherein the type of state change includes normal to abnormal, persistent abnormal, near the edge of abnormality, and abnormal to normal, etc.: the previous state is normal and the next state is normal. Events with an abnormal state are marked as normal-to-abnormal trend events and given the highest priority. Events with abnormal states before and after are marked as persistently abnormal and further differentiated into aggravated, stable, improved, or fluctuating states. Events with at least three valid numerical points that are normal throughout, progressively and strictly increasing, and whose relative position in the reference interval at the last point exceeds the configured threshold are marked as approaching the upper limit. Events with an abnormal state in the previous period and a normal state in the next period are marked as abnormal-to-normal trend events. The attention priority is recorded separately as normal-to-abnormal P1, aggravated abnormal P2, approaching the upper limit P3, and improved abnormal P4. Categories such as abnormal stability, fluctuation, transient abnormality, normal stability, and undetermined are also retained. Situations that are still within the abnormal range but are improving towards the reference interval can also be classified as abnormal improvement. In some embodiments, this application does not limit the upstream generation method for obtaining abnormal evidence sets, abnormal combinations, and multi-time point indicator change trends based on medical examination report documents; the three are at the same core input level.
[0100] In some embodiments, anomaly evidence sets undergo field integrity checks, requiring at least the inclusion of anomaly identifier, name, current result or qualitative conclusion, deviation direction, anomaly source, severity or priority, confidence level, source location, and object identifier. In some embodiments, anomaly combinations undergo field integrity checks, requiring at least the inclusion of member anomalies, combination labels, source range, and quality markers. In some embodiments, multi-time-point indicator change trends undergo field integrity checks, requiring at least the inclusion of object identifier, indicator identifier, previous and subsequent time points, previous and subsequent states, previous and subsequent observations or qualitative conclusions, previous and subsequent reference ranges or judgment criteria, state change type, time interval, and source location, with normal-to-anomaly events being a key focus input. In some embodiments, if key fields are missing from anomaly evidence sets, anomaly combinations, or multi-time-point indicator change trends, missing markers and data limitations are retained, and subsequent analysis and result status reflect these limitations; missing information is not added as known facts.
[0101] In some embodiments, multiple items from the abnormal evidence set, abnormal combination, and multi-time point indicator change trends are associated and encapsulated into one or more promoters. These promoters further associate the triage, task, and traceability information required for collaboration. The promoter is the logical starting point for collaborative analysis and can be carried by interconnected data objects; it is not required to be merged into a program class with a fixed name. In some embodiments, the promoter is a task package with a source, grouping the abnormal evidence set, abnormal combination, and multi-time point indicator change trends of the same user together. After triage, the selected department, selection reason, focus item, exclusion record, and departmental status (i.e., the result status of the specialist agent corresponding to each department) are associated. The promoter can be carried by several interconnected data objects, without requiring a program class with the same name. In some embodiments, the fields in the promoter are filled in step by step. During encapsulation, at least one of the following is filled in first: member abnormal item, trend event, combination mode type, and co-occurrence abnormal summary. Other fields such as selected department, focus item, exclusion reason, running identifier, and departmental status are filled in later. Undecided decisions are not filled in as final values in advance. In some embodiments, the promoter includes at least one of the following: member anomalies, trend events, combination pattern types, and co-occurrence anomaly summaries. Member anomalies include the code, name, current value, unit, reference range or judgment basis, deviation direction, anomaly source, severity, confidence level, and source location for each anomaly. Trend events include the time points before and after, the states before and after, the observations before and after, the type of state change, the direction of change, the time interval, the trend classification, and the priority of attention for the same indicator of the same object. Combination pattern types refer to the way in which a group of promoters are composed, including but not limited to single-system multiple anomalies, cross-system co-occurrence, severe anomaly priority, normal to anomaly trend, continuous anomaly trend, trend close to the edge of anomaly, anomaly to normal trend, low confidence pending coordination, and miss-the-word combination, etc. Co-occurrence anomaly summaries are used to summarize the names, deviation directions, and co-occurrence relationships of related anomalies to help departments identify key points. The summaries do not replace accessible original evidence, nor do they mask other indicators according to the primary responsibility or copying identity.In some embodiments, the promoter also includes a promoter identifier (uniquely identifying the promoter), an object and task identifier (corresponding to the subject, assessment task, report batch, or session identifier), a reference candidate department (a list of possible relevant departments obtained from a thesaurus, rules, external mapping, or model pre-classification (i.e., a list of departments that may be relevant before triage according to rules, for reference only by the triage model; the final selection of departments is based on the on-the-job directory), a reason for selection (recording the reasons for the association between the selected department and the combination pattern type, system label, severity, co-occurrence relationship, and longitudinal state change; combination pattern refers to the label of how a group of abnormalities are grouped, and the value is shown in the combination pattern type row; co-occurrence relationship refers to the simultaneous occurrence of multiple abnormalities in the same subject under the same examination background; longitudinal state change refers to the state change of the same indicator between different time points; simultaneous occurrence does not prove a common etiology), selected departments and focus items (recording the departments participating in this analysis and their focus; a complete attending physician analysis is performed on each selected department), and non-selected departments. The following data is required: exclusion reasons (recording all departments not included in the on-duty directory and their exclusion reasons, mutually exclusive with the included records and jointly covering the on-duty directory), operation identifier and roster verification status (related to this operation, the on-duty directory version, the participating roster (i.e., the roster containing the selected departments, i.e., the names of the departments participating in the analysis task), and verification results; model self-reported confidence is not used to replace program verification), unanswered questions (risk issues, evidence gaps, or conflict points that the selected departments should prioritize, without restricting their access to relevant case data), department-by-department execution and result status (related to the task submission, execution, and result status of each selected department; a complete roster or a submitted task does not mean that every anomaly has been analyzed), budget fields (including but not limited to department analysis budget, message quantity and length limits, checkpoint visibility, concurrency limit, retry limit, and deadline), and decision log identifiers (corresponding to at least one of the following: triage, roster verification, task submission, message transmission, source verification, conflict, and timeout or failure records). In some embodiments, the promoter serves three functions: input organization, triage reason tracing, and analysis starting point. Both triage reasons and departmental concerns can be traced back to the original member anomalies and trend events, avoiding a disconnect between the department selection explanation and the actual analysis data.
[0102] Module 12 is used to obtain reference candidate departments corresponding to the promoter through deterministic candidate recall, and extract the combined features corresponding to the promoter based on the reference candidate departments.
[0103] In some embodiments, determinism refers to the process completed by pre-written program rules without calling the model, with the same input yielding the same result twice. It involves reading the labels (such as system labels, specialty labels, risk area labels, etc.) of member anomalies or trend events in the promoter, and using the external mapping result corresponding to the label as the reference candidate department corresponding to the promoter. Here, the labels are obtained by the program through querying versioned vocabularies, naming rules, keywords, or configuration mappings, and are not generated by the model. The labels are merely triage clues and do not directly prove the disease or final departmental affiliation. External mapping refers to the correspondence between labels and departments given by reference sources other than internal vocabularies (such as terminology tables, coding tables, or external interfaces). The department corresponding to the label of a member anomaly or trend event in the promoter is used as the reference candidate department corresponding to the promoter (i.e., a department worth considering before triage). For example, mapping indicator names to standard codes, and then mapping the codes to the system or department's terminology standards, coding tables, or external interfaces. In some embodiments, for member anomalies or trend events in the promoter, if the terminology is not matched, the original information is retained and explicitly marked. A terminology mismatch means the label of the member anomaly or trend event cannot be found in the versioned terminology, possibly due to different syntax or the terminology not being included; this does not indicate the indicator is normal. A terminology mismatch also means the indicator name cannot be found in the versioned terminology, possibly due to a new item, a non-standard alias, or the terminology not being included; this does not indicate the indicator is normal. Inputs with mismatches retain their original name, observation, direction, and source, and are explicitly marked. The triage model can still select departments from the available list based on the entire batch of cases. In some embodiments, the reference candidate departments are used to assist triage and do not constitute a hard boundary for the final department selection. In some embodiments, truncating to a single label is not allowed. Since a member anomaly or trend event may be associated with multiple system labels simultaneously (e.g., homocysteine is associated with cardiovascular and metabolic systems), not truncating means all these labels are retained and participate in candidate recall, preventing cross-departmental clues from being discarded at the entry point simply because only the first one is selected.
[0104] In some embodiments, the number of anomalies associated with each reference candidate department, severity distribution, number of low-confidence items, cross-system co-occurrence pairs, multi-system labels, trend classification and attention priority distribution, involved indicators, direction of change, time interval, hit status of the combination pattern library (the combination pattern library refers to a pre-configured list of common anomaly combinations or organizational rules, such as classic combinations like simultaneous abnormalities in blood glucose, blood lipids, and blood pressure, which is equivalent to writing a table of experienced doctors' experience that "these things often go wrong together," and a hit only indicates that the set of clues is worth analyzing together) and rule triggers that need attention (triggers are inputs or states that require the system to take action or provide explicit prompts after a certain preset rule is met, such as severity reaching a configured threshold, multiple sources having inconsistent attributions to the same indicator, budget reaching the limit, or the source not being found; whether a hit is achieved is determined by the program by looking up a table, not by the model on the spot; for example, triggers refer to combinations of features such as hitting pre-configured attention rules, anomalies or trend events that need to be explicitly prompted in triage).
[0105] Module 13 is used to input the promoter, the reference candidate department, the combined features, and the directory of departments in service into the triage model to obtain at least one department output by the triage model and the corresponding departmental focus item of the at least one department.
[0106] In some embodiments, the promoter, reference candidate departments, combined features, and the directory of departments in operation are input into the triage model (referring to a large model that selects participating departments from the directory of departments in operation after seeing the entire batch of cases corresponding to the promoter at once). The triage model will select at least one department from the directory of departments in operation and output the corresponding departmental focus items for each of the at least one department. The triage model makes a department selection judgment after seeing the entire batch of cases in the promoter. It does not call the model separately for each item anomaly. When the output verification fails, it will retry according to the preset upper limit. This process is not limited to calling the triage model only once.
[0107] Module 14 is used to establish analysis tasks corresponding to the at least one department based on the promoter and the undergraduate focus items, and submit the analysis tasks to the specialist agent corresponding to the at least one department.
[0108] In some embodiments, for each of the at least one departments output by the triage model, an independent analysis task is established for that department based on the promoter and the corresponding departmental focus. At least one of the following information is injected into the analysis task: promoter, departmental focus, scope of available tools, budget, and output contract. Then, the analysis task is submitted to the corresponding specialty agent.
[0109] In some embodiments, each department's corresponding specialty agent independently completes a full analysis task, specifying the risk direction, supporting evidence, matters requiring coordination from the overall inspection agent, and uncertainties. Departmental focus items are used to prioritize the analysis task and do not indicate that all anomaly analyses have been completed item by item. In some embodiments, the tool scope, knowledge scope, and context scope of each specialty agent are limited by configuration. For example, the boundaries of specialty agents can be forcibly limited through a search-layer category whitelist or tool whitelist, rather than relying solely on prompts.
[0110] In some embodiments, each specialty agent executes a complete analysis process based on the case data, specialty focus items, independent tool configuration, budget ledger, and output records in the promoter (the budget ledger refers to the budget record maintained separately for the analysis task of the department, recording the used and remaining amounts of the investigation steps, tool calls, and message quantity). The specialty focus items guide the analysis focus and are not used to mask other raw indicators of the case. The analysis first conducts an investigation with a limit on the number of steps and time, and then plans the mechanism chain in segments (the mechanism chain refers to an evidence-supported inference chain that starts from abnormal indicators, goes through several inference steps, and points to a certain risk or mechanism; each step in the chain is a loop, without source support, relying solely on inference). The connected loops are called inference loops, and the dependent entries (dependent entries refer to risk, profile, or statement entries that are only established on the premise of a certain mechanism chain; when the chain exits the current conclusion, these entries lose their legitimate dependence and must be withdrawn or downgraded simultaneously while retaining the reasons) and external systems (external systems refer to independent verification contexts that are separate from the departmental intelligent agents and do not read their initial drafts) respectively perform analysis and chain-by-chain verification without looking at the initial drafts, and make constrained reflections based on the results. Reflection allows for the retention, deletion, downgrading, or modification of wording, but it is not allowed to add new chains or profiles (profiles refer to the structured description layer of the examinee's state formed based on the results of each chain, and statements refer to the textual expression layer oriented towards the report), upgrade the supporting state, or change the basis and target by modifying wording. In some embodiments, the specialist agent then collects evidence in batches according to the identified issues, verifies the positive and negative sources (i.e., the sources of literature or tools that constitute support and opposition respectively, both of which must be retrieved), and reviews the conclusions in a separate context (a separate context refers to a new model context that does not carry the previous reasoning records of the main body, and only re-evaluates the support of each conclusion based on the evidence collection results); at most one round of targeted supplementary inspection is conducted when the budget is available and the conditions are met (supplementary inspection refers to the system performing another limited-scope evidence collection for specific issues that still lack evidence, not requiring the examinee to conduct supplementary inspections). In some embodiments, when two consecutive inference loops appear for the first time, the latter is used as the breakpoint, and only the usable part before the breakpoint is retained. In some embodiments, when the review is judged to be contradictory, or when there is no legitimate supporting evidence cited by the evaluation results for the chain and its risk, profile, and statement layers, the chain exits the current conclusion area. In some embodiments, entries that lose their legitimate attachments are processed simultaneously and the reasons are retained. The literature evaluation and profile backfilling are still completed before the results of this department are delivered. When the evidence is insufficient, a definite conclusion is not forcibly formed.
[0111] Module 15 is used to aggregate the results and status of the specialist agents regarding the analysis task through the general inspection agent, and generate an auxiliary health risk analysis report corresponding to the physical examination report file.
[0112] In some embodiments, the overall review agent is the overall aggregation unit of team results. It obtains the result status of each specialty agent regarding the analysis task after each specialty agent completes its analysis task or closes the loop according to the deadline (closing the loop means completing the analysis task or stopping receiving new content and fixing the result status according to the deadline), and aggregates the result statuses of each specialty agent. In some embodiments, the responsibilities of the overall review agent include cross-system correlation identification (combining clues from different departments to find possible common explanations and then verifying evidence; simultaneously, the presence of clues does not prove a common cause), organizing duplicate statements in reports, marking conflict scopes, forming overall risk warnings, checking mutual constraints between recommendations (checking the limitations, contradictions, and data gaps between recommendations from different departments to form a consistent and readable explanation), explaining the impact of missing departments such as those not started or failed, and generating auxiliary health risk analysis reports that are traceable to users or doctors, with clear boundaries, visible conflicts, and degradable levels.
[0113] In some embodiments, the overall review agent does not delete the historical records of departmental conclusions, nor does it rewrite conflicting conclusions into a single, definitive conclusion. In some embodiments, the overall review agent first distinguishes whether the disagreement stems from differing wording or contradictory interpretations, then verifies whether the evidence from each party is true, relevant, and passes the verification. If the evidence is still insufficient, different interpretations are retained, and the facts are not determined by majority opinion. In some embodiments, for conflicts that cannot be resolved, the overall review agent retains them as issues requiring physician review or subsequent examination. For example, it first distinguishes whether the two departments have differing wording or contradictory interpretations. Differences in wording can be addressed, while contradictions in interpretation constitute a conflict. Then, it checks whether the evidence from each party is true, relevant, and passes departmental verification. If the evidence from one party is invalid, it withdraws or is downgraded to insufficient evidence according to the verification rules. If all evidence is valid but the evidence is still insufficient, different interpretations are retained, the disagreement is truthfully stated, and the case is submitted to the physician for review. The facts are not determined by majority opinion, and the historical records are only added to, not deleted. The original conclusions of either party are not deleted.
[0114] In some embodiments, associating and encapsulating multiple items from the abnormal evidence set, abnormal combination, and multi-time point indicator change trends into one or more promoters includes: normalizing multiple items from the abnormal evidence set, abnormal combination, and multi-time point indicator change trends; associating and encapsulating the resulting unified representation into one or more promoters, wherein the unified representation includes standardized fields. Here, the related operations are... Figure 1The embodiments shown are the same or similar, so they will not be described again, but are included here by reference.
[0115] In some embodiments, the unified representation result further includes quality tags. Here, the related operations are... Figure 1 The embodiments shown are the same or similar, so they will not be described again, but are included here by reference.
[0116] In some embodiments, obtaining at least one department output by the triage model and the corresponding departmental focus items for that at least one department includes: obtaining at least one department output by the triage model, the corresponding departmental focus items and reasons for selection for that at least one department, and departments not selected and their corresponding reasons for exclusion. Here, the related operations are... Figure 1 The embodiments shown are the same or similar, so they will not be described again, but are included here by reference.
[0117] In some embodiments, establishing the analysis task corresponding to the at least one department based on the promoter and the undergraduate focus items includes: verifying the participation list of the at least one department based on the undergraduate focus items, the reasons for selection, the non-selected departments, and the reasons for exclusion; if the verification passes, establishing the analysis task corresponding to the at least one department based on the promoter and the undergraduate focus items. Here, the related operations are... Figure 1 The embodiments shown are the same or similar, so they will not be described again, but are included here by reference.
[0118] In some embodiments, the step of inputting the promoter, the reference candidate departments, the combined features, and the directory of departments currently in operation into the triage model to obtain at least one department output by the triage model and the corresponding departmental focus items for the at least one department includes: inputting the promoter, the reference candidate departments, the combined features, the directory of departments currently in operation, and a preset team size upper limit into the triage model to obtain at least one department output by the triage model and the corresponding departmental focus items for the at least one department, wherein the number of departments corresponding to the at least one department is less than or equal to the preset team size upper limit. Here, the related operations are... Figure 1 The embodiments shown are the same or similar, so they will not be described again, but are included here by reference.
[0119] In some embodiments, submitting the analysis task to the specialist agent corresponding to the at least one department includes: submitting the analysis task to the specialist agent corresponding to the at least one department through an execution queue, wherein the execution queue has a preset concurrency limit, the concurrency limit being used to control the number of analysis tasks running simultaneously. Here, related operations are... Figure 1 The embodiments shown are the same or similar, so they will not be described again, but are included here by reference.
[0120] In some embodiments, the method further includes: for each specialty agent corresponding to the at least one department, at the model invocation checkpoint corresponding to that specialty agent, sending the collaborative message generated by that specialty agent to specialty agents corresponding to other departments via a central scheduling module, wherein the collaborative message includes a discovery type message. Here, the related operations are... Figure 1 The embodiments shown are the same or similar, so they will not be described again, but are included here by reference.
[0121] In some embodiments, the collaboration message also includes a problem type message. Here, related operations are... Figure 1 The embodiments shown are the same or similar, so they will not be described again, but are included here by reference.
[0122] In some embodiments, the collaborative message also includes a response type message regarding question-type messages received by the specialized agent from other specialized agents sent via the central scheduling module. Here, related operations are... Figure 1 The embodiments shown are the same or similar, so they will not be described again, but are included here by reference.
[0123] In some embodiments, the total number of collaborative messages generated by the specialist agent corresponding to each department is less than or equal to a preset threshold. Here, the related operations are... Figure 1 The embodiments shown are the same or similar, so they will not be described again, but are included here by reference.
[0124] In some embodiments, the result status includes the task result obtained by the specialized intelligent agent after completing the analysis task. Here, related operations are... Figure 1 The embodiments shown are the same or similar, so they will not be described again, but are included here by reference.
[0125] In some embodiments, the result state also includes the current state fixed by the specialized intelligent agent after the cutoff condition is met. Here, the related operations are... Figure 1 The embodiments shown are the same or similar, so they will not be described again, but are included here by reference.
[0126] In some embodiments, the step of aggregating the result status of the specialist agents regarding the analysis task through the overall inspection agent to generate an auxiliary health risk analysis report includes: performing a subject-by-subject check on the result status of the specialist agents regarding the analysis task through the overall inspection agent, aggregating target result statuses that do not contain preset hard errors, and generating an auxiliary health risk analysis report. Here, the related operations are... Figure 1 The embodiments shown are the same or similar, so they will not be described again, but are included here by reference.
[0127] In some embodiments, the step of performing a subject-by-subject check on the result status of the specialist agents regarding the analysis task through the overall inspection agent, aggregating target result statuses without preset hard errors, and generating an auxiliary health risk analysis report includes: performing a subject-by-subject check on the result status of the specialist agents regarding the analysis task through the overall inspection agent, conducting a cell-related evidence investigation on the result status, aggregating target result statuses without preset hard errors and with clear evidence boundaries, and generating an auxiliary health risk analysis report. Here, the related operations are... Figure 1 The embodiments shown are the same or similar, so they will not be described again, but are included here by reference.
[0128] Figure 6 Exemplary systems that can be used to implement the various embodiments described in this application are shown; such as Figure 6 As shown in some embodiments, system 300 can function as any of the devices described in each of the embodiments. In some embodiments, system 300 may include one or more computer-readable media having instructions (e.g., system memory or NVM / storage device 320) and one or more processors (e.g., one or more processors 305) coupled to the one or more computer-readable media and configured to execute the instructions to implement the module and thus perform the actions described in this application.
[0129] In one embodiment, the system control module 310 may include any suitable interface controller to provide any suitable interface to at least one of the processors 305 and / or any suitable device or component communicating with the system control module 310.
[0130] The system control module 310 may include a memory controller module 330 to provide an interface to the system memory 315. The memory controller module 330 may be a hardware module, a software module, and / or a firmware module.
[0131] System memory 315 can be used, for example, to load and store data and / or instructions for system 300. In one embodiment, system memory 315 may include any suitable volatile memory, such as suitable DRAM. In some embodiments, system memory 315 may include double data rate type quad synchronous dynamic random access memory (DDR4 SDRAM).
[0132] In one embodiment, the system control module 310 may include one or more input / output (I / O) controllers to provide interfaces to the NVM / storage device 320 and (one or more) communication interfaces 325.
[0133] For example, NVM / storage device 320 may be used to store data and / or instructions. NVM / storage device 320 may include any suitable non-volatile memory (e.g., flash memory) and / or may include any suitable (one or more) non-volatile storage devices (e.g., one or more hard disk drive (HDD), one or more optical disc (CD) drives, and / or one or more digital universal optical disc (DVD) drives).
[0134] NVM / storage device 320 may include storage resources that are physically part of a device on which system 300 is mounted, or that can be accessed by the device without necessarily being part of it. For example, NVM / storage device 320 may be accessed via a network through one or more communication interfaces 325.
[0135] One or more communication interfaces 325 may provide the system 300 with an interface to communicate over one or more networks and / or with any other suitable device. The system 300 may wirelessly communicate with one or more components of a wireless network in accordance with any of one or more wireless network standards and / or protocols.
[0136] In one embodiment, at least one of the processors 305 may be logically packaged with one or more controllers of the system control module 310 (e.g., memory controller module 330). In one embodiment, at least one of the processors 305 may be logically packaged with one or more controllers of the system control module 310 to form a system-in-package (SiP). In one embodiment, at least one of the processors 305 may be integrated with the logic of one or more controllers of the system control module 310 on the same die. In one embodiment, at least one of the processors 305 may be integrated with the logic of one or more controllers of the system control module 310 on the same die to form a system-on-a-chip (SoC).
[0137] In various embodiments, system 300 may be, but is not limited to, a server, workstation, desktop computing device, or mobile computing device (e.g., laptop computing device, handheld computing device, tablet computer, netbook, etc.). In various embodiments, system 300 may have more or fewer components and / or different architectures. For example, in some embodiments, system 300 includes one or more cameras, a keyboard, a liquid crystal display (LCD) screen (including a touchscreen display), a non-volatile memory port, multiple antennas, a graphics chip, an application-specific integrated circuit (ASIC), and a speaker.
[0138] In addition to the methods and devices described in the above embodiments, this application also provides a computer-readable storage medium storing computer code that, when executed, performs the method described in any of the preceding embodiments.
[0139] This application also provides a computer program product that, when executed by a computer device, performs the method described in any of the preceding claims.
[0140] This application also provides a computer device, the computer device comprising:
[0141] One or more processors;
[0142] Memory, used to store one or more computer programs;
[0143] When the one or more computer programs are executed by the one or more processors, the one or more processors cause the one or more processors to perform the method as described in any of the preceding methods.
[0144] It should be noted that this application can be implemented in software and / or a combination of software and hardware, for example, using an application-specific integrated circuit (ASIC), a general-purpose computer, or any other similar hardware device. In one embodiment, the software program of this application can be executed by a processor to implement the steps or functions described above. Similarly, the software program of this application (including related data structures) can be stored in a computer-readable recording medium, such as RAM memory, a magnetic or optical drive, a floppy disk, or similar devices. Furthermore, some steps or functions of this application can be implemented in hardware, for example, as circuitry that cooperates with a processor to perform the various steps or functions.
[0145] Furthermore, a portion of this application can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to this application through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.
[0146] Communication media include media through which communication signals containing, for example, computer-readable instructions, data structures, program modules, or other data are transmitted from one system to another. Communication media can include guided transmission media (such as cables and wires (e.g., optical fibers, coaxial cables, etc.)) and wireless (unguided transmission) media capable of propagating energy waves, such as sound, electromagnetic, RF, microwave, and infrared. Computer-readable instructions, data structures, program modules, or other data can be embodied as modulated data signals in, for example, wireless media (such as carrier waves or similar mechanisms embodied as part of spread spectrum technology). The term "modulated data signal" refers to a signal whose one or more characteristics are altered or set in a manner that encodes information in the signal. Modulation can be analog, digital, or a hybrid modulation technique.
[0147] By way of example and not limitation, computer-readable storage media may include volatile and non-volatile, removable and non-removable media implemented by any method or technology for storing information such as computer-readable instructions, data structures, program modules or other data. For example, computer-readable storage media include, but are not limited to, volatile memories such as random access memory (RAM, DRAM, SRAM); and non-volatile memories such as flash memory, various read-only memories (ROM, PROM, EPROM, EEPROM), magnetic and ferromagnetic / ferroelectric memories (MRAM, FeRAM); and magnetic and optical storage devices (hard disks, magnetic tapes, CDs, DVDs); or other media now known or hereafter developed capable of storing computer-readable information / data for use by a computer system.
[0148] Herein, one embodiment of this application includes an apparatus comprising a memory for storing computer program instructions and a processor for executing the program instructions, wherein when the computer program instructions are executed by the processor, the apparatus is triggered to run a method and / or technical solution based on the foregoing embodiments of this application.
[0149] It will be apparent to those skilled in the art that this application is not limited to the details of the exemplary embodiments described above, and that this application can be implemented in other specific forms without departing from the spirit or essential characteristics of this application. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of this application is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be embraced within this application. No reference numerals in the claims should be construed as limiting the scope of the claims. Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices recited in the apparatus claims may also be implemented by a single unit or device in software or hardware. The terms "first," "second," etc., are used to indicate names and do not indicate any particular order.
Claims
1. A method for generating auxiliary health risk analysis reports, wherein, The method includes: Multiple items from the abnormal evidence set, abnormal combination, and multi-time point indicator change trend are associated and encapsulated into one or more promoters. The abnormal evidence set, abnormal combination, and multi-time point indicator change trend are obtained based on the physical examination report file. The promoter includes at least one of the following: member abnormal item, trend event, combination pattern type, and co-occurrence abnormal summary. The reference candidate departments corresponding to the promoter are obtained through deterministic candidate recall, and the combined features corresponding to the promoter are extracted based on the reference candidate departments; Input the promoter, the reference candidate department, the combined features, and the directory of departments in service into the triage model to obtain at least one department output by the triage model and the corresponding departmental focus items for the at least one department; Based on the promoter and the undergraduate focus items, establish the analysis task corresponding to the at least one department, and submit the analysis task to the specialty agent corresponding to the at least one department; The general inspection agent aggregates the results and statuses of the specialist agents regarding the analysis task to generate an auxiliary health risk analysis report corresponding to the physical examination report file.
2. The method according to claim 1, wherein, The process of associating and encapsulating multiple factors from the abnormal evidence set, abnormal combinations, and multi-time point indicator change trends into one or more promoters includes: The abnormal evidence set, abnormal combination, and multiple time point indicator change trends are standardized. The unified representation results obtained after processing are associated and encapsulated into one or more promoters, wherein the unified representation results include standardized fields.
3. The method according to claim 2, wherein, The unified representation result also includes quality tags.
4. The method according to claim 1, wherein, The process of obtaining at least one department output by the triage model and the corresponding departmental focus items for that at least one department includes: Obtain at least one department output by the triage model, the corresponding departmental focus items and reasons for selection for the at least one department, and the non-selected departments and their corresponding reasons for exclusion.
5. The method described in section 4 requires a long time to be used, in which... The step of establishing the analysis task corresponding to the at least one department based on the promoter and the undergraduate focus items includes: The participation list of at least one department is verified based on the undergraduate focus items, the reasons for selection, the non-selected departments, and the reasons for exclusion. If the verification is successful, an analysis task corresponding to the at least one department is established based on the promoter and the undergraduate focus items.
6. The method according to claim 1, wherein, The step of inputting the promoter, the reference candidate departments, the combined features, and the directory of departments currently in operation into the triage model to obtain at least one department output by the triage model and the corresponding departmental focus items for the at least one department includes: The promoter, the reference candidate department, the combined features, the directory of departments in service, and the preset team size limit are input into the triage model to obtain at least one department output by the triage model and the corresponding undergraduate focus item of the at least one department, wherein the number of departments corresponding to the at least one department is less than or equal to the preset team size limit.
7. The method according to claim 1, wherein, Submitting the analysis task to the specialist agent corresponding to the at least one department includes: The analysis task is submitted to the specialist agent corresponding to the at least one department through an execution queue, wherein the execution queue has a preset concurrency limit, which is used to control the number of analysis tasks running at the same time.
8. The method according to claim 1, wherein, The method further includes: For each specialty agent in the at least one department, at the model call checkpoint corresponding to that specialty agent, the collaboration message generated by that specialty agent is sent to the specialty agents in other departments via the central scheduling module, wherein the collaboration message includes discovery type messages.
9. The method according to claim 8, wherein, The collaboration messages also include question type messages.
10. The method according to claim 9, wherein, The collaborative message also includes a response message to a question-type message received by the specialized agent from other specialized agents sent via the central scheduling module.
11. The method according to any one of claims 8 to 10, wherein, The total number of collaborative messages generated by the specialist intelligent agent corresponding to each department is less than or equal to a preset threshold.
12. The method according to claim 1, wherein, The result status includes the task results obtained by the specialized intelligent agent after completing the analysis task.
13. The method according to claim 12, wherein, The result state also includes the current state of the specialized intelligent agent after the cutoff condition is met.
14. The method according to claim 1, wherein, The process of aggregating the results and statuses of the specialist agents regarding the analysis task through the general inspection agent to generate an auxiliary health risk analysis report includes: The general inspection agent performs a subject-by-subject check on the results status of the specialist agents regarding the analysis task, aggregates the target result statuses that do not contain preset hard errors, and generates an auxiliary health risk analysis report.
15. The method according to claim 14, wherein, The process involves the general inspection agent performing a subject-by-subject check on the results status of the specialist agents regarding the analysis task, aggregating the target result statuses that do not contain preset hard errors, and generating an auxiliary health risk analysis report, including: The general inspection agent performs a subject-by-subject check on the results status of the specialist agents regarding the analysis task, and conducts a cell-related evidence investigation on the results status. The target results status that do not have preset hard errors and have clear evidence boundaries are aggregated to generate an auxiliary health risk analysis report.
16. A system for generating auxiliary health risk analysis reports, the system comprising: The first module is used to associate and encapsulate multiple items from the abnormal evidence set, abnormal combination, and multi-time point indicator change trend into one or more promoters. The abnormal evidence set, the abnormal combination, and the multi-time point indicator change trend are obtained based on the physical examination report file. The promoter includes at least one of the following: member abnormal item, trend event, combination pattern type, and co-occurrence abnormal summary. The second module is used to obtain reference candidate departments corresponding to the promoter through deterministic candidate recall, and extract the combined features corresponding to the promoter based on the reference candidate departments; The third module is used to input the promoter, the reference candidate department, the combined features, and the directory of departments in service into the triage model to obtain at least one department output by the triage model and the corresponding departmental focus item of the at least one department; The fourth module is used to establish analysis tasks corresponding to the at least one department based on the promoter and the undergraduate focus items, and submit the analysis tasks to the specialist agent corresponding to the at least one department. The fifth module is used to aggregate the results status of the specialist agents regarding the analysis task through the general inspection agent, and generate an auxiliary health risk analysis report corresponding to the physical examination report file.
17. A computer device for generating auxiliary health risk analysis reports, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method as described in any one of claims 1 to 16.
18. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method as described in any one of claims 1 to 16.
19. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the steps of the method as described in any one of claims 1 to 16.
