An interpretable coronary heart disease non-invasive screening method and system based on multi-agent decision

Abstract

The application provides an explainable coronary heart disease noninvasive screening method and system based on multi-agent decision, and the method comprises the following steps: encoding according to a preset field order based on noninvasive health assessment data to obtain an input vector, inputting the input vector into a grade score prediction model, outputting a score vector of each candidate risk grade, and selecting the grade corresponding to the maximum score vector as a target risk grade. Evidence description information is constructed according to the source information of the noninvasive health assessment data and the related information of the target risk grade, and the evidence description information is divided into multiple evidence units according to different data types. It is judged whether each evidence unit has the ability to support the target risk grade, and the evidence unit judgment result is obtained, and each evidence unit corresponds to a stability weight. The establishment judgment result is calculated according to the evidence unit judgment result, the stability weight and the establishment threshold, and it is judged whether the target risk grade is established at the evidence level according to the establishment judgment result, and the reliability of disease screening is improved.

Description

Technical Field

[0001] This invention belongs to the field of disease screening, and in particular relates to an interpretable non-invasive screening method and system for coronary heart disease based on multi-agent decision-making. Background Technology

[0002] Atherosclerotic cardiovascular disease (ASCVD) is a leading cause of serious adverse cardiovascular events such as myocardial infarction and stroke. Risk assessment plays a fundamental role in disease prevention, stratified management, and long-term health intervention. Currently, ASCVD risk assessment relies heavily on risk prediction models built from population statistical data. These models typically take limited traditional risk factors such as age, sex, blood pressure, blood lipids, blood glucose, and smoking history as input, outputting the probability or risk level of an individual's cardiovascular event within a certain future timeframe. These models simply aggregate data from different sources and directly provide a risk level, lacking a holistic mechanism to determine whether the current risk level is supported by evidence. This makes it difficult to meet the practical needs of non-invasive screening and chronic disease management scenarios for reliable and controllable risk assessment results. Therefore, improving the reliability of disease screening has become an urgent technical challenge. Summary of the Invention

[0003] The purpose of this invention is to design an interpretable non-invasive screening method and system for coronary heart disease based on multi-agent decision-making, which can improve the reliability of disease screening.

[0004] To achieve the above objectives, in a first aspect of the present invention, an interpretable non-invasive screening method for coronary artery disease based on multi-agent decision-making is provided, the method comprising: Obtain non-invasive health assessment data and its source information, and encode the non-invasive health assessment data according to a preset field order to obtain an input vector; The input vector is input into a preset level score prediction model, and the score vector of each candidate risk level is output. The candidate risk level corresponding to the largest score vector is selected as the target risk level. Evidence description information is constructed based on the source information and the relevant information of the target risk level, and the evidence description information is divided into multiple evidence units according to different data types; Based on preset judgment conditions, each of the evidence units is judged to determine whether it has the ability to support the target risk level, and the judgment result of the evidence unit is obtained; wherein, each of the evidence units corresponds to a stability weight. The validity determination result is calculated based on the determination result of the evidence unit, the stability weight, and the preset validity threshold. The validity determination result is then used to determine whether the target risk level is valid at the evidentiary level.

[0005] Furthermore, after determining whether the target risk level is valid at the evidentiary level based on the validity determination result, the method further includes: If the target risk level is determined to be valid at the evidentiary level based on the validity determination result, then the level constraint coefficient corresponding to the target risk level is obtained; The validity determination result is multiplied by the level constraint coefficient to obtain the output qualification decision result; Based on the output eligibility decision results, determine whether the target risk level is eligible for external output.

[0006] Furthermore, after multiplying the validity determination result with the level constraint coefficient to obtain the output qualification decision result, the method further includes: Obtain the output eligibility decision results from the previous evaluation cycle; The output qualification decision result is updated based on the preset adjustment coefficient, the output qualification decision result, and the previous output qualification decision result to obtain the updated output qualification decision result.

[0007] Furthermore, after determining whether the target risk level is qualified to provide external output based on the output qualification decision result, the method further includes: If the target risk level is deemed to have the qualification to be exported based on the output qualification decision result, then the validity determination result is multiplied by the output qualification decision result to obtain the explanatory emphasis mark; Based on the explanation of the emphasis mark and the preset sorting rules, evidence units are filtered from the evidence description information to obtain the target evidence unit; A corresponding interpretation card is generated for each target evidence unit, and all the interpretation cards constitute an interpretation card set. The target risk level is written into a preset conclusion field, the set of explanation cards is written into a preset explanation field, and the conclusion field and the explanation field are pushed to the target application according to the preset application output field.

[0008] Furthermore, after determining whether the target risk level is qualified to provide external output based on the output qualification decision result, the method further includes: If the target risk level is determined not to be qualified for external output based on the output qualification decision result, then the controlled state risk level is obtained based on the target risk level. The controlled state conclusion is calculated based on the output qualification decision result, the target risk level, and the controlled state risk level. A status interpretation card is generated based on the evidence description information. The controlled status risk level is written into a preset conclusion field, and the status interpretation card is written into a preset interpretation field. The conclusion field and the interpretation field are pushed to the target application terminal according to the preset application output field.

[0009] Further, the step of calculating the controlled state conclusion based on the output qualification decision result, the target risk level, and the controlled state risk level includes: The first data is obtained by multiplying the output qualification decision result by the target risk level; Subtract the preset value from the output qualification decision result, and then multiply by the controlled state risk level to obtain the second data; The controlled state conclusion is obtained by adding the first data and the second data.

[0010] Further, the validity determination result includes a valid result or a invalid result, and the calculation of the validity determination result based on the evidence unit determination result, the stability weight, and the preset validity threshold includes: The determination result of each evidence unit is multiplied by the corresponding stability weight to obtain multiple third data. Add all the third data together to obtain the fourth data. If the fourth data is greater than or equal to the validity threshold, the validity determination result is the validity result; otherwise, the validity determination result is the invalid result.

[0011] In a second aspect, the present invention provides an interpretable non-invasive screening system for coronary artery disease based on multi-agent decision-making, the system comprising: The acquisition unit is used to acquire non-invasive health assessment data and the source information of the non-invasive health assessment data, and to encode the non-invasive health assessment data according to a preset field order to obtain an input vector. The prediction unit is used to input the input vector into a preset level score prediction model, output the score vector of each candidate risk level, and select the candidate risk level corresponding to the largest score vector as the target risk level. A construction unit is used to construct evidence description information based on the source information and the relevant information of the target risk level, and to divide the evidence description information into multiple evidence units according to different data types; The judgment unit is used to determine whether each of the evidence units has the ability to support the target risk level according to preset judgment conditions, and to obtain the evidence unit judgment result; wherein, each of the evidence units corresponds to a stability weight. The calculation unit is used to calculate the validity determination result based on the determination result of the evidence unit, the stability weight and the preset validity threshold, and to determine whether the target risk level is valid at the evidence level based on the validity determination result.

[0012] In a third aspect of the invention, an electronic device is provided, the electronic device including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the method described in the first aspect above.

[0013] In a fourth aspect of the invention, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in the first aspect.

[0014] The beneficial technical effects of the present invention are at least as follows: To address the aforementioned issues, this invention provides an interpretable non-invasive coronary artery disease screening method and system based on multi-agent decision-making. Its core lies in the structured reconstruction of the generation, judgment, and output processes of risk assessment results. This invention does not merely focus on improving the accuracy of risk prediction, but rather addresses the crucial question of "whether a risk level is valid and whether it should be output" in practical applications. It constructs a continuous decision-making chain from the generation of the risk assessment object to the validity determination, then to output eligibility control and final application output. By binding and encapsulating the target risk level with its corresponding evidence structure, and introducing a validity determination mechanism for non-invasive screening scenarios, the effectiveness of the risk level no longer depends solely on the model calculation results themselves, but rather on whether the currently available data can form stable and consistent evidence support. Furthermore, this invention transforms the validity determination results into risk level output eligibility decisions, making the external presentation of the risk level a controlled engineering behavior rather than a direct result of assessment calculation. It also generates interpretable screening results and application output formats that match the output eligibility status. Through the above technical solutions, this invention effectively adapts to the application environment of non-invasive screening, multi-source data fusion, and dynamic follow-up without increasing the burden of invasive examinations. This improves the overall rationality, stability, and interpretability of ASCVD risk assessment results, thereby providing more reliable and sustainable technical support for clinical decision support and long-term health management. Attached Figure Description

[0015] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0016] Figure 1This is a flowchart of an interpretable non-invasive screening method for coronary heart disease based on multi-agent decision-making, provided in an embodiment of this application.

[0017] Figure 2 This is a flowchart of an interpretable non-invasive screening method for coronary heart disease based on multi-agent decision-making, provided in another embodiment of this application.

[0018] Figure 3 This is a flowchart of an interpretable non-invasive screening method for coronary heart disease based on multi-agent decision-making, provided in the third embodiment of this application.

[0019] Figure 4 This is a flowchart of an interpretable non-invasive screening method for coronary heart disease based on multi-agent decision-making, provided in the fourth embodiment of this application.

[0020] Figure 5 This is a schematic diagram of the structure of an interpretable non-invasive screening system for coronary heart disease based on multi-agent decision-making, provided in an embodiment of this application. Detailed Implementation

[0021] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0022] Please refer to Figure 1 , Figure 1 This is a flowchart of an interpretable non-invasive screening method for coronary artery disease based on multi-agent decision-making, provided in an embodiment of this application. Figure 1 The method may include, but is not limited to, steps S101 to S105.

[0023] Step S101: Obtain non-invasive health assessment data and source information of non-invasive health assessment data; encode the non-invasive health assessment data according to the preset field order to obtain the input vector; Step S102: Input the input vector into the preset level score prediction model, output the score vector of each candidate risk level, and select the candidate risk level corresponding to the largest score vector as the target risk level. Step S103: Construct evidence description information based on source information and relevant information on target risk level, and divide the evidence description information into multiple evidence units according to different data types; Step S104: Determine whether each piece of evidence has the ability to support the target risk level according to the preset judgment conditions, and obtain the judgment result of the evidence unit; wherein, each piece of evidence has a corresponding stability weight. Step S105: Calculate the validity determination result based on the evidence unit determination result, stability weight, and preset validity threshold, and determine whether the target risk level is valid at the evidence level based on the validity determination result.

[0024] In step S101 of some embodiments, the non-invasive health assessment data collection method adopts the stable interface of the existing business system: structured records are obtained through the information system interface of medical institutions and fall into structured fields (such as field values ​​in physical examination records and follow-up forms); lifestyle-related information is collected through the questionnaire interface and stored in discrete encoding forms such as options or frequencies; wearable device data is reported through the synchronization interface or mobile SDK provided by the device manufacturer, and the system generates trend summaries according to preset time windows (such as "recent change direction" and "fluctuation degree" represented by numerical labels).

[0025] Furthermore, the non-invasive health assessment data is encoded according to a preset field order to obtain an input vector. The sources of the coded data include information system field values, questionnaire option codes, and window summary codes of wearable data. All of these are represented using a uniform scale and then written into the input vector. .

[0026] In step S102 of some embodiments, the input vector is input to a preset risk level prediction model, and the score vectors of each candidate risk level are output. The candidate risk level corresponding to the largest score vector is selected as the target risk level. Specifically, the target risk level is calculated using a fixed-structure forward computation, which consists of two fully connected mapping layers plus nonlinear activation, outputting the score vectors of each candidate risk level. Then, the candidate risk level corresponding to the largest score vector is taken as the target risk level. The calculation relationship is as follows: ; in, Represents the input vector; , The weight matrix and bias vector of the first fully connected layer are written to the model file through offline configuration before system deployment and loaded by the inference service. They are used directly for matrix multiplication and addition operations at runtime. This represents the hidden representation of the first layer output. This represents an element-wise nonlinear mapping, specifically implemented by performing threshold truncation on each dimension to obtain a nonnegative representation. , The weight matrix and bias vector representing the second fully connected layer are also fixed before deployment and loaded by the inference service; This represents the score vector for each candidate risk level. Indicates the first The score vector of each candidate risk level. Indicates the target risk level, taking The candidate risk level corresponding to the largest score vector is taken as the target risk level.

[0027] In step S103 of some embodiments, after obtaining the target risk level... Simultaneously, generate evidence description information and... Binding and encapsulation. Evidence description information is constructed based on the source information of the non-invasive health assessment data and relevant information about the target risk level. Specifically, the generated input vector is recorded. The actual list of fields used and their coding methods (e.g., a field comes from a structured field in an information system, a field comes from a questionnaire option code, a field comes from a wearable window summary) are recorded, along with intermediate calculation summaries related to the target risk level determination (e.g., related to...). Corresponding score vector and in (Relative sorting position within). Ultimately, the target risk level will be determined. Together with the evidence description information, it is encapsulated as a risk assessment object, serving as input for the subsequent "establishment determination step," enabling subsequent steps to complete evidence verification and output control around the same object.

[0028] In step S104 of some embodiments, the risk assessment object includes a target risk level. And corresponding descriptive evidence information for each risk level. This descriptive evidence information has been generated through a structured organization of non-invasive health assessment data, and its content not only identifies which non-invasive information was involved... The calculation also records the state characteristics and encoding results of this information in the current evaluation period in the form of structured fields.

[0029] In the specific implementation process, the evidence is first divided into several evidentiary units with clear semantic boundaries based on the clearly defined structure in the evidence description information. The division of these evidentiary units is determined according to the type of evidence actually used as an independent criterion in generating the target risk level, typically categorized as structured field evidence from information systems, coded questionnaire options, wearable window summary evidence, and other evidence. The evidence is categorized according to relevant intermediate computational summaries and other types, allowing multiple fields within the same category to be merged into a single evidence unit or split into multiple evidence units according to configuration. Therefore, the number of evidence units N is determined by the number of evidence unit types actually used in this assessment. Each evidence unit corresponds to one type actually used to generate... The system utilizes various non-invasive information types, such as status codes from physical examination records, behavioral marker codes from follow-up records, or trend summary codes from wearable devices. For each of these evidence units, the system is configured with multiple decision agents, each corresponding to a specific evidence unit type. The internal logic of each agent employs a deterministic rule-based decision structure. Specifically, the decision agent reads the description field of the corresponding evidence unit from the risk assessment object and, based on pre-configured decision conditions, outputs whether the evidence unit, in its current state, supports the target risk level. The output is represented in discrete form and recorded as the judgment result of the evidence unit. It should be noted that the description field is a set of structured content that supports the judgment, used to fully express the status and validity of the evidence unit within the current evaluation period. For example, the description field of a structured field-type evidence unit in an information system typically includes a list of field identifiers, the corresponding coding results, a status flag indicating whether the preset judgment conditions are met, and an identifier indicating whether the field is enabled in this evaluation. Questionnaire-type evidence units include item identifiers, option codes, and validity flags; wearable-type evidence units include window summary identifiers, trend codes, and stability flags; and intermediate calculation summary-type evidence units include... The system includes structured information such as the corresponding score position and its ranking relationship in the score vector. When determining whether an evidence unit has the ability to support the target risk level based on the judgment conditions, the judgment agent reads the description field of the evidence unit and verifies it item by item according to the pre-configured rules corresponding to the type of evidence unit. For example, it verifies whether the encoding result falls within the valid range corresponding to the semantics of the target risk level, whether the state label meets the continuity or stability requirements, and whether the evidence unit is marked as a valid input and actually used in this evaluation. When the necessary set of conditions for this type of evidence unit is met, the system outputs a discrete judgment result that the evidence unit "has the ability to support the target risk level"; otherwise, it outputs a discrete judgment result that "does not have the ability to support the target risk level".

[0030] In step S105 of some embodiments, to ensure that the validity determination reflects not only the quantity of evidence but also its stability over time, a stability adjustment term designed for non-invasive screening scenarios is introduced when summarizing the output of the determination agent. Specifically, for each evidence unit, the evidence description information already includes a state change marker for that evidence over the most recent evaluation periods. The state change marker refers to structured identification information describing the state changes of a particular evidence unit over multiple consecutive evaluation periods, reflecting the consistency or volatility of the evidence unit over time, such as whether it maintains the same coding result in adjacent evaluation periods, whether frequent switching occurs, or whether it exhibits a continuous trend. This marker is maintained along with the evidence unit when generating the evidence description information. While outputting the determination result of the evidence unit, the determination agent simultaneously provides the stability weight of that evidence unit, reflecting the consistency of the evidence within a continuous observation window. The determination result of the evidence unit is combined with the stability marker to form a weighted evidence validity count, used for overall validity determination. That is, the validity determination result is calculated based on the evidence unit determination result, the stability weight, and a preset validity threshold. Specifically, the determination result of each evidence unit is multiplied by its corresponding stability weight to obtain multiple third data points; all third data points are summed to obtain a fourth data point. If the fourth data point is greater than or equal to the validity threshold, the validity determination result is valid; otherwise, the validity determination result is invalid. As shown in the following formula: ; in, The determination of validity indicates whether the target risk level is valid at the evidentiary level. The determination of validity includes a valid result or a invalid result. When the threshold is 1, the determination of validity is considered valid, indicating that the target risk level is valid at the evidentiary level. When the threshold is 0, the determination of validity is invalid, indicating that the target risk level is not valid at the evidentiary level. Indicates the first Each intelligent agent provides its judgment result for its corresponding evidence unit. When the evidence unit supports the target risk level... The value is 1 if the capability is met, and 0 otherwise. Indicates the relationship with the first The stability weight corresponding to each evidence unit is directly mapped from the state change marker in the evidence description information, and is used to reflect the stability of the evidence within a continuous non-invasive observation period. The stability weight is obtained from the state change marker through a pre-defined mapping rule. Specifically, the state change of the evidence unit within several consecutive evaluation periods is transformed into a comparable stability description. For example, a state change marker is generated based on whether the evidence unit maintains the same coding result in adjacent periods, whether there are frequent switching, or whether it shows a continuous and consistent trend of change. Then, the corresponding stability weight is found in the configuration table based on the marker. The smaller the state change and the more consistent the performance within consecutive periods, the higher the stability weight mapped, and vice versa. This indicates the number of evidence units involved in the determination of validity, and this number is directly determined by the structure of the evidence description information in the risk assessment object; Indicates the target risk level The corresponding establishment threshold is set during the system configuration phase based on the semantics of the risk level and remains consistent throughout operation. It should be noted that the purpose of determining whether a target risk level is established at the evidentiary level is to assess whether the risk level has sufficient and stable evidentiary support under current non-invasive data conditions, thereby determining whether it can serve as a reliable basis for subsequent output and application. When the establishment determination result... When = 1, it indicates that the evidence supporting the target risk level generally meets the preset requirements in terms of quantity and temporal stability, suggesting that the target risk level has an evidentiary basis and is usable, indicating that the target risk level is not derived from scattered or unstable evidence; when When the value is 0, it means that the current evidence is insufficient to support the direct presentation of this risk level to the public.

[0031] Through the above calculation process, the judgment results of multiple decision-making agents on different evidence units are fused within the same framework, so that the validity determination reflects both the scope of evidence coverage and the consistency of evidence over time. This design makes the validity determination more in line with the actual need in non-invasive screening scenarios where "long-term trends are more valuable than single fluctuations," thereby improving the stability and credibility of the validity determination in practical applications.

[0032] Steps S101 to S105 of this embodiment involve acquiring non-invasive health assessment data and its source information, encoding the data according to a preset field order, and obtaining an input vector. This input vector is then fed into a preset risk level prediction model, which outputs score vectors for each candidate risk level. The candidate risk level corresponding to the largest score vector is selected as the target risk level. Evidence description information is constructed based on the source information and relevant information of the target risk level. This evidence description information is divided into multiple evidence units according to different data types. Each evidence unit is judged based on preset judgment conditions to determine whether it supports the target risk level, resulting in an evidence unit judgment result. Each evidence unit has a corresponding stability weight. An validity judgment result is calculated based on the evidence unit judgment result, the stability weight, and a preset validity threshold. The validity judgment result determines whether the target risk level is valid at the evidence level, thus improving the reliability of disease screening.

[0033] In some embodiments, after step S105, an interpretable non-invasive screening method for coronary artery disease based on multi-agent decision-making may also include, but is not limited to, steps S201 to S203: Step S201: If the target risk level is determined to be valid at the evidentiary level based on the validity determination result, then obtain the level constraint coefficient corresponding to the target risk level. Step S202: Multiply the validity determination result by the level constraint coefficient to obtain the output qualification decision result; Step S203: Determine whether the target risk level is qualified to be exported based on the output qualification decision results.

[0034] In step S201 of some embodiments, when the validity determination result indicates that the target risk level is valid at the evidentiary level, the judgment of "whether the target risk level is valid at the evidentiary level" is further transformed into a risk level output qualification decision that the system can directly execute in practical applications. This makes whether the target risk level is presented externally no longer a natural endpoint of the calculation process, but a clearly controlled decision result.

[0035] In the specific implementation process, the first step is based on the target risk level. The semantic features are used to construct grade-related output constraints. Different risk levels correspond to different decision sensitivities in non-invasive screening scenarios. For example, higher risk levels usually have stronger intervention targeting in clinical practice, thus requiring stricter reliance on the validity of evidence when determining output eligibility. To address this, the system pre-configures a grade constraint coefficient for each possible risk level during the deployment phase to characterize the strictness of output under the validity condition for that level. This coefficient remains stable during operation and corresponds one-to-one with the risk level, thereby incorporating the semantic features of the risk level into the output eligibility decision without introducing additional data.

[0036] In some embodiments, steps S202 to S203 will determine the validity of the determination result. With target risk level The corresponding level constraint coefficients are input to and output to the qualification decision module. This module adopts a deterministic rule-based calculation structure, and its core calculation relationship can be expressed as: ; in, This indicates the output eligibility decision result, used to indicate whether the target risk level is eligible for external output; This indicates the result of the validity determination; Indicates the target risk level The corresponding risk level constraint coefficient is set during the system configuration phase based on the application semantics of the risk level. Through this calculation, the system maps "whether the evidence is valid" and "risk level semantic constraints" into an executable output eligibility determination result. It should be noted that the purpose of determining whether a target risk level is eligible for external output is to determine whether, under the current non-invasive data and evidence conditions, the risk level is suitable for presentation and use as a definitive conclusion. The target risk level generated in a non-invasive screening scenario is essentially a calculation result, while the output eligibility determination distinguishes between "risk levels that can be directly used as conclusions" and "risk levels that require controlled processing," thereby avoiding the direct application of risk levels to clinical decision-making or health management in cases of insufficient evidence, inadequate stability, or high semantic sensitivity.

[0037] The meaning of the output qualification decision result is uniformly agreed upon by the system during the design phase. When the output qualification decision result is in a valid state, it means that the target risk level can be presented externally as a clear risk level and used for subsequent applications under the current evidence conditions and level semantic constraints. When the output qualification decision result is in an invalid state, it means that the risk level is not suitable for direct external output and needs to be processed through controlled presentation or status prompts.

[0038] Furthermore, considering the periodic and continuous nature of data acquisition in non-invasive screening scenarios, a stable control term for the continuity of risk levels is introduced into the output eligibility decision results. The system retains the output eligibility decision results from the most recent assessment period, i.e., the previous output eligibility decision results, in the risk assessment objects. These previous output eligibility decision results serve as historical constraints in updating the output eligibility decision results in this step. By combining the currently calculated output eligibility decision results with the previous output eligibility decision results, consistency in risk level output behavior can be maintained throughout continuous assessment periods, avoiding frequent changes in output eligibility due to short-term evidence fluctuations. This combination relationship can be expressed as: ; in, This indicates the updated output eligibility decision result; This indicates the output eligibility decision results for the current assessment cycle; The previous output qualification decision result indicates the output qualification status of the previous assessment cycle associated with the current risk assessment object. This status is maintained by the system in the historical assessment records and is read and used when making the current output qualification decision. This represents a moderating coefficient used to balance the impact of current decisions with historical states. This coefficient is set during the system configuration phase to reflect the need for output stability in non-invasive screening scenarios. Through this design, output eligibility decisions can respond to changes in current evidence conditions while maintaining continuity over time.

[0039] After completing the above calculations, the system will The table is mapped to discrete output eligibility markers, and these markers are written into the risk assessment object. This allows the object to explicitly identify, in subsequent processes, whether its target risk level is eligible for external output within the current assessment period. Logically, the output eligibility decision result has only two states: eligible for external output or not eligible for external output. In the specific implementation, the results are calculated continuously... It will be mapped to discrete output eligibility flags for explicit judgment in subsequent steps, where a value of 1 indicates that it is qualified to output externally, and a value of 0 indicates that it is not qualified to output externally.

[0040] Steps S201 to S203, as illustrated in this embodiment, are used to clearly indicate whether the target risk level is qualified for external output under the current conditions of validity, semantic constraints of the level, and continuous control. Subsequent steps, when generating the final non-invasive screening result, can directly rely on... Choose the corresponding output format without having to perform judgment or calculation again.

[0041] Please refer to Figure 3In some embodiments, after step S203, an interpretable non-invasive screening method for coronary artery disease based on multi-agent decision-making may also include, but is not limited to, steps S301 to S304: Step S301: If the target risk level is deemed to have the qualification to be exported based on the output qualification decision result, the validity determination result is multiplied by the output qualification decision result to obtain the interpretation emphasis mark. Step S302: Based on the interpretation emphasis mark and preset sorting rules, filter evidence units from the evidence description information to obtain the target evidence unit; Step S303: Generate a corresponding interpretation card for each target evidence unit, and all interpretation cards constitute an interpretation card set; Step S304: Write the target risk level into the preset conclusion field, write the set of explanation cards into the preset explanation field, and push the conclusion field and explanation field to the target application terminal according to the preset application output field.

[0042] In step S301 of some embodiments, the focus is on outputting eligibility decision results. The goal is to transform the output eligibility decision into a final result that can be directly used in actual non-invasive screening applications. This step serves as the result implementation function in the overall process. Its processing object is no longer raw data or intermediate judgment signals, but a structured decision result that already includes the target risk level, validity judgment result, and output eligibility marker, thereby ensuring that the result generation process always maintains strict consistency with the preceding steps.

[0043] The outcome generation path includes a risk level presentation path and a controlled presentation path. In the specific processing, the first step is to base the output qualification decision on... The corresponding output eligibility flag selects the result generation path, and assembles the final screening result data structure accordingly. This data structure is organized in a unified conclusion body format, containing a conclusion field, an explanation field, and an application output field. The conclusion field expresses the final screening conclusion, the explanation field carries the evidence explanation content corresponding to the conclusion, and the application output field encapsulates the output payload for both the doctor and health management ends. To ensure consistency between the explanation content and the output eligibility status, an explanation strength flag is introduced when assembling the explanation field. This flag is determined by the validity judgment result. Together with the output eligibility decision results, they determine the depth and number of evidence units presented in the interpretation field. The calculation relationship of the interpretation strength marker is as follows: ; in, This indicates the interpretation strength flag, used to control the assembly strategy of the interpretation fields; This indicates the result of the validity determination; This indicates the output qualification decision result, used to indicate whether the current assessment cycle has entered the risk level presentation path.

[0044] In some embodiments, steps S302 to S304, the interpretation generation module is based on The value of is determined by selecting several evidence units from the evidence description information reference identifier set according to a preset sorting rule, and generating corresponding interpretation cards. Among these, the interpretation strength marker... The value of is discrete and is usually 0 or 1. This indicates that the assembly of complete risk level evidence is permitted. This indicates that only the explanatory content or minimal set of explanations related to the controlled state is assembled. Each explanation card consists of an evidence source type identifier, an evidence state description, and matching instructions related to the risk level. Its content is directly filled by structured entries in the evidence description information, thereby ensuring that the explanation result is consistent with the previous evidence judgment. It should be noted that, based on When selecting evidence units from the evidence description information reference identifier set, the evidence units are sorted using a preset sorting rule, such as sorting by the importance of the evidence unit, stability weight, or correlation with the target risk level. When selecting the top-ranked pieces of evidence to form a complete explanation, Only a small number of evidence units at the top of the sorting are selected for status description.

[0045] When the output qualification decision result indicates that the risk level presentation path is entered, the target risk level will be... The system writes the conclusion field and assembles a complete set of explanation cards in the explanation field. Simultaneously, it generates output payloads for different application endpoints in the application output field. The output payload for physicians is organized in structured message format, including a risk level field, a set of explanation cards, and structured fragments that can be stored in the medical record system. The output payload for health management is generated through template mapping, mapping the evidence descriptions in the explanation cards to colloquial text and attaching corresponding health management tips. The template mapping process uses a fixed template library, with template selection based on the evidence source type identifier, and the template content directly taken from the structured fields in the explanation cards. After assembling the conclusion field, explanation field, and application output field, the final ASCVD non-invasive screening result is presented as the output for this evaluation cycle and pushed to the target application endpoint through a preset interface or messaging mechanism. The ASCVD non-invasive screening result is a combination of the assembled conclusion and explanation fields. The conclusion field expresses the final screening conclusion or controlled status, and the explanation field provides evidence corresponding to that conclusion; together, they constitute the externally usable screening result content. The output can be directly used by the doctor's system for risk stratification management, or by the health management platform for continuous monitoring and management display, thus completing a closed loop from risk assessment, validity determination, output eligibility decision to practical application output.

[0046] In this embodiment, steps S301 to S304 involve, upon entering the risk level presentation path, interpreting the intensity markers... The value is selected from the evidence description information to generate the corresponding evidence unit and interpretation card, while the target risk level is also assigned. Including information in the conclusion field ensures that screening results maintain clarity while possessing interpretive depth commensurate with the level of evidence supporting them, thus avoiding situations where only the target risk level is given without supporting explanations. By assembling an interpretation set consisting of multiple interpretation cards in the interpretation field, it's possible to visually present which non-invasive evidence contributed to the formation of the target risk level and the current status of this evidence. This helps doctors or health management systems quickly understand the source of the risk level, improving the interpretability and credibility of screening results, while maintaining a consistent result structure for easy display and subsequent processing across different applications.

[0047] Please refer to Figure 4 In some embodiments, after step S203, an interpretable non-invasive screening method for coronary artery disease based on multi-agent decision-making may also include, but is not limited to, steps S401 to S403: Step S401: If the target risk level is determined not to be qualified for external output based on the output qualification decision result, then the controlled state risk level is obtained based on the target risk level. Step S402: Calculate the controlled state conclusion based on the output qualification decision result, target risk level, and controlled state risk level; Step S403: Generate a status interpretation card based on the evidence description information, write the controlled status risk level into the preset conclusion field, write the status interpretation card into the preset interpretation field, and push the conclusion field and interpretation field to the target application terminal according to the preset application output field.

[0048] In steps S401 to S403 of some embodiments, if the target risk level is determined not to be eligible for external output based on the output eligibility decision result, it indicates that the system enters the controlled presentation path. When entering the controlled presentation path, the core content of the conclusion field is not a clear risk level, but rather a controlled state conclusion is generated. The controlled state conclusion refers to the system's expression of the conclusion given when the target risk level is not eligible for external output; its meaning is "the current assessment is in a controlled state," indicating that no clear risk level will be output during this assessment period, but a state result that can still be used for management and subsequent actions will still be output. Structurally, this state result uses the same field system encoding as the target risk level conclusion. At this time, the explanation field is equipped with a state explanation card. This card organizes its content around the current assessment state and combines evidence description information to generate a state description and subsequent data preparation prompts. The state explanation card differs in content organization from the explanation card under the risk level presentation path. The former is organized around the explanation of the reasons for the controlled state and subsequent data preparation prompts, while the latter is organized around the evidence support for the target risk level. Both originate from evidence description information but have different presentation objectives. The generation of status interpretation cards involves selecting evidence unit entries related to "unmet output eligibility" from the evidence description information reference identifier set, organizing them into status description fields and data preparation prompt fields. For example, evidence units judged to have unmet supporting conditions or low stability are mapped to "current condition descriptions," and "subsequent data preparation prompts" are generated by matching preset action templates based on evidence unit type identifiers. Finally, the above structured fields are encapsulated into status interpretation cards and written into the interpretation fields. To maintain the consistency of the conclusion field structure, a unified conclusion assembly logic is used to write the risk level field, ensuring that output messages under different presentation paths remain consistent at the field level. The controlled state conclusion is calculated based on the output eligibility decision result, the target risk level, and the controlled state risk level. This is achieved by multiplying the output eligibility decision result by the target risk level to obtain the first data; subtracting the preset value from the output eligibility decision result and then multiplying by the controlled state risk level to obtain the second data; and finally, adding the first and second data to obtain the controlled state conclusion. The formula is as follows:

[0049] in, The conclusion for the controlled state indicates the value of the conclusion field in the final output; This indicates the output of the eligibility decision result; Indicates the target risk level; The controlled state risk level indicates the risk level value corresponding to the controlled screening state. This value is related to... The same risk level field system is used for coding, which is used to identify the current screening status under the controlled presentation path. This is pre-configured by the system during the deployment phase. Specifically, a risk level value representing the "controlled screening status" is reserved in the risk level field system and written into a configuration table or dictionary mapping table. When entering the controlled presentation path, the corresponding controlled status risk level value is read from this configuration table as... This assembly method ensures that the final output maintains structural uniformity while accurately distinguishing between risk level presentation and controlled presentation states semantically. The risk level presentation path uses a "target risk level + interpretation card set" approach, directly linking the presented risk conclusion with supporting evidence. This facilitates doctors' or management systems' understanding of the risk level's origin and its use in tiered decision-making. The controlled presentation path uses a "controlled state conclusion + state interpretation card" approach, enabling the system to provide usable state conclusions and subsequent guidance even when conditions for outputting a clear risk level are unavailable. This avoids outputting risk levels that might mislead decision-making. Furthermore, a unified field structure ensures consistent output messages across different states, facilitating front-end display and downstream system integration. This enhances the reliability, interpretability, and engineering usability of results output in non-invasive screening scenarios.

[0050] The controlled state risk level is written to the conclusion field, and the state interpretation card is written to the interpretation field. Based on the application output field, the conclusion and interpretation fields are pushed to the target application. That is, after assembling the conclusion, interpretation, and application output fields, the system will push the final ASCVD non-invasive screening result as the output for this assessment cycle to the target application via a preset interface or messaging mechanism. This output can be directly used by the doctor's system for risk stratification management, or by the health management platform for continuous monitoring and management display, thus completing a closed loop from risk assessment, validity determination, output eligibility decision to actual application output.

[0051] In steps S401 to S403 of this embodiment, when entering the controlled presentation path, the controlled state risk level is written into the conclusion field and a state interpretation card is assembled in the interpretation field. This allows clear and usable screening status information to be provided to external systems without outputting a specific risk level, enabling doctors or health management platforms to accurately understand that the current assessment result is under control and take corresponding management measures accordingly. At the same time, the state interpretation card explains the reasons for the control and the direction of subsequent data preparation, which helps to avoid being misled by directly outputting a risk level that is not supported by sufficient evidence. While maintaining the consistency of the structure of the conclusion field and the interpretation field, it improves the safety, interpretability, and engineering usability of non-invasive screening results in practical applications.

[0052] Please see Figure 5 This application also provides an interpretable non-invasive coronary artery disease screening system based on multi-agent decision-making, which can implement the above-mentioned interpretable non-invasive coronary artery disease screening method based on multi-agent decision-making. The system includes: The acquisition unit 501 is used to acquire non-invasive health assessment data and the source information of the non-invasive health assessment data, and to encode the non-invasive health assessment data according to the preset field order to obtain the input vector. The prediction unit 502 is used to input the input vector into the preset level score prediction model, output the score vector of each candidate risk level, and select the candidate risk level corresponding to the largest score vector as the target risk level. Construction unit 503 is used to construct evidence description information based on source information and relevant information on target risk level, and to divide the evidence description information into multiple evidence units according to different data types; The judgment unit 504 is used to judge whether each evidence unit has the ability to support the target risk level according to the preset judgment conditions, and to obtain the evidence unit judgment result; wherein, each evidence unit has a corresponding stability weight. The calculation unit 505 is used to calculate the validity determination result based on the evidence unit judgment result, stability weight and preset validity threshold, and to determine whether the target risk level is valid at the evidence level based on the validity determination result.

[0053] The specific implementation of this interpretable non-invasive coronary artery disease screening system based on multi-agent decision-making is basically the same as the specific implementation of the aforementioned interpretable non-invasive coronary artery disease screening method based on multi-agent decision-making, and will not be repeated here.

[0054] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. An interpretable non-invasive screening method for coronary heart disease based on multi-agent decision-making, characterized in that, The method includes: Obtain non-invasive health assessment data and its source information, and encode the non-invasive health assessment data according to a preset field order to obtain an input vector; The input vector is input into a preset level score prediction model, and the score vectors of each candidate risk level are output. The candidate risk level corresponding to the largest score vector is selected as the target risk level. Evidence description information is constructed based on the source information and the relevant information of the target risk level, and the evidence description information is divided into multiple evidence units according to different data types; Based on preset judgment conditions, each of the evidence units is judged to determine whether it has the ability to support the target risk level, and the judgment result of the evidence unit is obtained; wherein, each of the evidence units corresponds to a stability weight. The validity determination result is calculated based on the determination result of the evidence unit, the stability weight, and the preset validity threshold. The validity determination result is then used to determine whether the target risk level is valid at the evidentiary level.

2. The interpretable non-invasive screening method for coronary heart disease based on multi-agent decision-making as described in claim 1, characterized in that, After determining whether the target risk level is established at the evidentiary level based on the establishment determination result, the method further includes: If the target risk level is determined to be valid at the evidentiary level based on the validity determination result, then the level constraint coefficient corresponding to the target risk level is obtained; The validity determination result is multiplied by the level constraint coefficient to obtain the output qualification decision result; Based on the output eligibility decision results, determine whether the target risk level is eligible for external output.

3. The interpretable non-invasive screening method for coronary heart disease based on multi-agent decision-making as described in claim 2, characterized in that, After multiplying the validity determination result by the level constraint coefficient to obtain the output qualification decision result, the method further includes: Obtain the output eligibility decision results from the previous evaluation cycle; The output qualification decision result is updated based on the preset adjustment coefficient, the output qualification decision result, and the previous output qualification decision result to obtain the updated output qualification decision result.

4. The interpretable non-invasive screening method for coronary heart disease based on multi-agent decision-making as described in claim 2, characterized in that, After determining whether the target risk level is qualified to provide external services based on the output qualification decision result, the method further includes: If the target risk level is deemed to have the qualification to be exported based on the output qualification decision result, then the validity determination result is multiplied by the output qualification decision result to obtain the explanatory emphasis mark; Based on the explanation of the emphasis mark and the preset sorting rules, evidence units are filtered from the evidence description information to obtain the target evidence unit; A corresponding interpretation card is generated for each target evidence unit, and all the interpretation cards constitute an interpretation card set. The target risk level is written into a preset conclusion field, the set of explanation cards is written into a preset explanation field, and the conclusion field and the explanation field are pushed to the target application according to the preset application output field.

5. The interpretable non-invasive screening method for coronary heart disease based on multi-agent decision-making as described in claim 2, characterized in that, After determining whether the target risk level is qualified to provide external services based on the output qualification decision result, the method further includes: If the target risk level is determined not to be qualified for external output based on the output qualification decision result, then the controlled state risk level is obtained based on the target risk level. The controlled state conclusion is calculated based on the output qualification decision result, the target risk level, and the controlled state risk level. A status interpretation card is generated based on the evidence description information. The controlled status risk level is written into a preset conclusion field, and the status interpretation card is written into a preset interpretation field. The conclusion field and the interpretation field are pushed to the target application terminal according to the preset application output field.

6. The interpretable non-invasive screening method for coronary heart disease based on multi-agent decision-making as described in claim 5, characterized in that, The step of calculating the controlled state conclusion based on the output qualification decision result, the target risk level, and the controlled state risk level includes: The first data is obtained by multiplying the output qualification decision result by the target risk level; Subtract the preset value from the output qualification decision result, and then multiply by the controlled state risk level to obtain the second data; The controlled state conclusion is obtained by adding the first data and the second data.

7. The interpretable non-invasive screening method for coronary heart disease based on multi-agent decision-making as described in claim 1, characterized in that, The validity determination result includes a valid result or a invalid result. The calculation of the validity determination result based on the evidence unit determination result, the stability weight, and the preset validity threshold includes: The determination result of each evidence unit is multiplied by the corresponding stability weight to obtain multiple third data. Add all the third data together to obtain the fourth data. If the fourth data is greater than or equal to the validity threshold, the validity determination result is the validity result; otherwise, the validity determination result is the invalid result.

8. An interpretable non-invasive screening system for coronary heart disease based on multi-agent decision-making, characterized in that, The system includes: The acquisition unit is used to acquire non-invasive health assessment data and the source information of the non-invasive health assessment data, and to encode the non-invasive health assessment data according to a preset field order to obtain an input vector. The prediction unit is used to input the input vector into a preset level score prediction model, output the score vector of each candidate risk level, and select the candidate risk level corresponding to the largest score vector as the target risk level. A construction unit is used to construct evidence description information based on the source information and the relevant information of the target risk level, and to divide the evidence description information into multiple evidence units according to different data types; The judgment unit is used to determine whether each of the evidence units has the ability to support the target risk level according to preset judgment conditions, and to obtain the evidence unit judgment result; wherein, each of the evidence units corresponds to a stability weight. The calculation unit is used to calculate the validity determination result based on the determination result of the evidence unit, the stability weight and the preset validity threshold, and to determine whether the target risk level is valid at the evidence level based on the validity determination result.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the interpretable non-invasive screening method for coronary heart disease based on multi-agent decision-making as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the interpretable non-invasive screening method for coronary heart disease based on multi-agent decision-making as described in any one of claims 1 to 7.

Images