A method and system for medical behavior compliance review based on AI model

Abstract

This invention relates to a method and system for medical behavior compliance review based on an AI model. The method includes: acquiring multi-source heterogeneous medical data and medical compliance data, and standardizing the medical data; constructing a medical dataset and a compliance dataset based on the standardized medical data and medical compliance data, respectively; based on event graph parsing, using a first intelligent agent to parse examination reports, identify treatment behaviors, and identify clinical decision-making logic from the standardized medical data, and outputting a patient treatment profile; and using a second intelligent agent to perform compliance review on the medical behavior to be reviewed based on the patient treatment profile and the compliance dataset. The compliance review includes compliance judgment and outputting the confidence level of the judgment result. This invention improves the automation and accuracy of medical behavior compliance review by using multiple intelligent agents, combined with patient treatment profiles and causal chains.

Description

Technical Field

[0001] This invention belongs to the field of medical information and artificial intelligence technology, specifically relating to a method and system for medical behavior compliance review based on an AI model. Background Technology

[0002] Compliance auditing of medical practices is a crucial link in ensuring the rational use of medical insurance funds and standardizing medical service behavior. With the rapid development of the healthcare industry and the continuous improvement of informatization, the compliance supervision of medical practices has become increasingly important. Traditional methods of medical practice auditing mainly rely on manual review, which is inefficient, highly subjective, and struggles to handle massive amounts of medical data and complex medical scenarios. Meanwhile, while some existing intelligent auditing systems can perform partial audits based on rules, they lack flexibility in rule application, understanding of complex medical practices, and completeness in end-to-end supervision. Therefore, a more intelligent, efficient, and comprehensive method for medical practice compliance auditing is needed to ensure the standardization, rationality, and legality of medical practices, improve the quality of medical services, and effectively supervise the rational use of medical resources.

[0003] Traditional compliance audits of medical practices primarily rely on manual spot checks and automated systems based on simple rules. Manual audits are inefficient, have limited coverage, are costly, and are susceptible to subjective biases. Existing automated systems, mostly based on structured data (such as billing items and drug codes) for post-hoc rule matching (such as drug contraindications and duplicate charges), lack depth and cannot understand complex clinical scenarios, unstructured medical records, the actual meaning of examination and test results, or the logical relationship between diagnosis and treatment. They struggle to effectively audit the rationality and necessity of treatment plans (such as the determination of surgical indications, the basis for using high-value consumables, and excessive testing) and their suitability to the patient's actual condition. Since these audits are largely post-hoc, they cannot effectively intervene before or during treatment, leading to difficulties in correcting violations, resulting in losses to medical insurance funds and waste of medical resources. Static rules are ill-suited to complex clinical realities and rapidly updating medical knowledge, easily generating numerous false positives or overlooking genuine violations. Summary of the Invention

[0004] To automate and improve the accuracy of medical practice compliance review, a first aspect of this invention provides an AI-based medical practice compliance review method, comprising: acquiring multi-source heterogeneous medical data and medical compliance data, and standardizing the medical data; constructing a medical dataset and a compliance dataset based on the standardized medical data and medical compliance data, respectively; based on event graph parsing, using a first intelligent agent to parse examination reports, identify treatment behaviors, and identify clinical decision-making logic from the standardized medical data, and outputting a patient treatment profile; and using a second intelligent agent to perform compliance review on the medical practice to be reviewed based on the patient treatment profile and the compliance dataset; wherein the compliance review includes compliance judgment and outputting the confidence level of the judgment result.

[0005] In some embodiments of the present invention, the step of parsing examination reports, identifying treatment behaviors, and recognizing clinical decision-making logic from the standardized medical data using a first intelligent agent based on event graph parsing, and outputting a patient treatment profile includes: parsing examination reports from the standardized medical data using a first intelligent agent based on a preset prompt word template; matching medical datasets to form a time-series chain using an enhanced retrieval matching method; and recognizing the causal chain of clinical decisions based on the event graph and the time-series chain.

[0006] Furthermore, the identification of causal chains for clinical decisions based on event graphs and temporal chains includes: identifying causal chains for clinical decisions through an attention network based on the event graph of clinical decisions.

[0007] In some embodiments of the present invention, the compliance review of the medical behavior to be reviewed by a second intelligent agent based on the patient's medical profile and the compliance dataset includes: retrieving a constraint set of the compliance dataset based on multiple diagnostic codes or multiple treatment codes in the patient's medical profile; the constraint set includes medical guidelines, payment rules, and operational constraints; inputting the matched constraint set and the medical behavior to be reviewed into the second intelligent agent to obtain a three-state decision and the confidence level of the three-state decision; the three-state decision includes compliant, non-compliant, and pending.

[0008] Furthermore, the step of inputting the matched constraint set and the medical behavior to be reviewed into the second intelligent agent to obtain the three-state decision and the confidence level of the three-state decision includes: calculating the compliance score of the three-state decision by the semantic similarity between the medical behavior to be reviewed and the matched constraint set, and the probability of the second intelligent agent judging compliance based on the matched constraint set.

[0009] In the above embodiments, the method further includes: reviewing real-time medical data through a second intelligent agent and issuing different warnings based on the stage of the medical behavior.

[0010] A second aspect of the present invention provides a medical behavior compliance review system based on an AI model, comprising: an acquisition module for acquiring multi-source heterogeneous medical data and medical compliance data, and standardizing the medical data; constructing a medical dataset and a compliance dataset based on the standardized medical data and medical compliance data, respectively; an output module for parsing examination reports, identifying treatment behaviors and clinical decision-making logic from the standardized medical data using a first intelligent agent based on event graph parsing, and outputting a patient treatment profile; and an review module for conducting compliance review of the medical behavior to be reviewed using a second intelligent agent based on the patient treatment profile and the compliance dataset; wherein the compliance review includes compliance judgment and outputting the confidence level of the judgment result.

[0011] A third aspect of the present invention provides an electronic device comprising: one or more processors; and a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the AI ​​model-based medical behavior compliance review method provided in the first aspect of the present invention.

[0012] In a fourth aspect, the present invention provides a computer-readable medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the AI-based medical behavior compliance review method provided in the first aspect of the present invention.

[0013] The beneficial effects of this invention are:

[0014] This invention utilizes clinical semantic parsing technology for unstructured medical texts, employing medical NLP models to extract "diagnosis and treatment logic chains" (such as causal chains of symptoms → examination → diagnosis → treatment) from medical records and examination reports, overcoming the limitations of traditional methods that can only process structured data. It constructs a dynamic, end-to-end triggering review mechanism, establishing differentiated triggering rules for "pre-event, during-event, and post-event" processes. A collaborative decision-making model combining a rule engine and AI inference is employed. Through a dual-engine architecture of a rule layer (matching policy clauses) and an AI inference layer (handling complex clinical scenarios), interpretable results are output, and violation determinations require correlation with specific medical record evidence and clinical guidelines. Attached Figure Description

[0015] Figure 1 This is a basic flowchart illustrating the AI-based medical behavior compliance review method in some embodiments of the present invention.

[0016] Figure 2 This is a schematic diagram illustrating the specific process of the AI-based medical behavior compliance review method in some embodiments of the present invention;

[0017] Figure 3This is a schematic diagram of the structure of an AI-based medical behavior compliance review system in some embodiments of the present invention;

[0018] Figure 4 This is a schematic diagram of the structure of an electronic device in some embodiments of the present invention. Detailed Implementation

[0019] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0020] refer to Figure 1 and Figure 2 In a first aspect, the present invention provides a medical behavior compliance review method based on an AI model, comprising: S100. acquiring multi-source heterogeneous medical data and medical compliance data, and standardizing the medical data; constructing a medical dataset and a compliance dataset based on the standardized medical data and medical compliance data, respectively; S200. based on event graph parsing, using a first intelligent agent to parse examination reports, identify treatment behaviors and clinical decision-making logic from the standardized medical data, and outputting a patient treatment profile; S300. based on the patient treatment profile and the compliance dataset, using a second intelligent agent to perform compliance review on the medical behavior to be reviewed; the compliance review includes compliance judgment and outputting the confidence level of the judgment result.

[0021] In step S100 of some embodiments of the present invention, multi-source heterogeneous medical data and medical compliance data are acquired, and the medical data is standardized; based on the standardized medical data and medical compliance data, a medical dataset and a compliance dataset are constructed respectively.

[0022] Specifically, the system accesses Hospital Information System (HIS), Electronic Medical Record System (EMR), Laboratory Information System (LIS), Picture Archiving and Communication System (PACS), and medical order system in real-time or near real-time through standardized interfaces. It accesses structured data, including patient basic information, diagnostic codes (ICD), surgical procedure codes (ICD-9-CM-3 / ICD-10-PCS), drug codes, billing items, and examination and test requests and results (numerical). It also accesses unstructured / semi-structured text data, including outpatient / inpatient medical records (chief complaint, present illness, past medical history, physical examination, progress notes, surgical records, discharge summaries), examination and test reports (descriptive sections), pathology reports, and imaging reports. The data access layer is responsible for cleaning, standardizing, and initially integrating the data to form a patient-centered, time-ordered medical dataset.

[0023] A compliance dataset (rule base) is constructed based on national medical insurance policies, clinical practice guidelines, drug instructions, and hospital regulations, creating a multi-dimensional compliance rule base. The knowledge base is structured based on eight tag systems: medical orders, medication, treatment, scales, features, diagnoses, symptoms, and laboratory tests, forming a standardized knowledge network for medical behavior analysis. A three-dimensional approach of "multi-source acquisition + semantic modeling + tag-based organization" is adopted: standardized terminology databases such as ICD-10 and SNOMED CT are automatically synchronized, and policy documents such as treatment guidelines are extracted. NLP technology is used to extract high-frequency terms from electronic medical records (e.g., "AMI" mapping to "acute myocardial infarction") to form localized knowledge. Ontology is used to define medical concepts and their associations, and frame representation is used to structure knowledge units, constructing a semantically related knowledge graph. A classification tree is built based on the eight tags, and cross-tag association attributes are configured to achieve orderly organization and efficient retrieval of knowledge.

[0024] In step S200 of some embodiments of the present invention, the step of parsing examination reports, identifying treatment behaviors, and identifying clinical decision-making logic from the standardized medical data based on event graph parsing, and outputting a patient treatment profile by a first intelligent agent, includes:

[0025] S201. Based on the preset prompt word template, the examination report is parsed from the standardized medical data by the first intelligent agent;

[0026] Specifically, Radiology-BERT, guided by preset prompt word templates, or a finely tuned LLM (Large Language Model), is used to extract and quantify the index of abnormal key sentences. For example, the application of natural language processing techniques in the parsing process, for the constructed three-level medical text parsing system, firstly, a word segmentation method combining dictionary and statistical learning is used to accurately segment and tag medical terms such as "acute myocardial infarction," laying the foundation for subsequent analysis. Then, a BiLSTM-CRF model is used to identify more than 10 types of entities, including diseases, drugs, and surgeries, achieving the separation and extraction of core entities and modifying information in complex structures. Finally, a word dependency network is constructed through syntactic analysis, and semantic role annotation clarifies the logical roles of each entity in medical behavior (such as agent, patient, time, etc.), achieving a deep semantic understanding of the diagnosis and treatment behavior and disease progression in the medical record.

[0027] It should be noted that for routine medical records within the hospital, AI semantic analysis technology is used to identify patient characteristics, abnormal examination and test results, diagnosis and treatment information, extract structured information from medical records, and parse medical natural language to integrate and build an AI model.

[0028] Domain knowledge fusion enhances parsing accuracy through a dual-pillar system: Firstly, it constructs a medical terminology knowledge base, integrating authoritative standards such as ICD-10 and SNOMED CT to form a localized terminology network containing multi-dimensional information including full names, abbreviations, and codes. This enables standardized mapping and ambiguity resolution of terms such as "myocardial infarction → acute myocardial infarction (I21)," ensuring consistency in terminology expression in medical records. Secondly, it embeds clinical guidelines and medication regulations into the parsing model, for example, intelligently identifying violations such as overdosing and contraindication. This system, through a dual mechanism of "terminology standardization + rule engineization," ensures that the parsing results conform to medical semantic norms and meet regulatory compliance requirements, providing underlying knowledge support for accurate analysis of medical practices.

[0029] Furthermore, causal chains are inferred by constructing an event graph based on medical time procedures.

[0030] The parsed results are structured, organizing the key information extracted according to a pre-defined structured data model to generate structured data containing fields such as patient basic information, chief complaint, present illness, past medical history, diagnosis, treatment plan, medication records, and examination and test results. Key information extraction and patient feature identification are then performed to extract key patient information from the text.

[0031] To achieve the identification and interpretation of abnormal test results, the following steps are taken: Identify key indicators in the report, extract their values, and analyze their clinical significance. This includes identifying key indicators in test results reports (such as abnormal findings in blood routine tests, biochemistry, and imaging descriptions), extracting their values, and analyzing their clinical significance (e.g., "significantly elevated white blood cell count" suggests infection, "ground-glass opacities in the lungs" suggests inflammation or a space-occupying lesion). Diagnostic information extraction and standardization are also crucial. This involves identifying the physician's diagnostic conclusions and mapping them to standard diagnostic codes to understand the certainty and priority of the diagnosis.

[0032] Therefore, in S202, based on the enhanced retrieval and matching method, the first agent matches the medical dataset to form a time-series chain; based on the event graph and the time-series chain, the causal chain of clinical decisions is identified. The medical dataset is matched with the compliance dataset to form a time-series chain. B= { b 1 ,…,b n}

[0033] Then there is the identification of medical treatment behaviors, which identifies the medical treatment measures planned or implemented by doctors and maps them to standard operation terms and codes. Mining of medical treatment basis and logical relationships. Analyze how the symptoms, signs, and examination results described in the text support the diagnosis, and how the diagnosis guides the formulation of treatment plans. Identify the key clinical decision-making logics. Output: Structure the extracted key information, associate it with the specific medical treatment events and time points of the patient, and form a patient medical treatment portrait. The output portrait is represented as P = {symptoms, examination reports, diagnostic path sequence, time sequence chain, cost} or P = <demographics, comorbidity, key-labs, staging, medical treatment path sequence, cost>.

[0034] Specifically, in order to improve the accuracy of matching results, the rule data is atomized, and an atomic clause = <Antecedent, Consequent, RefID>; DAG (Directed Acyclic Graph) generation: When the patient portrait P arrives, the combination engine retrieves in the graph database Neo4j according to the diagnosis / stage / department and satisfies: , the nodes are automatically spliced into Rule-DAG; The DAG topology is output according to topologicalsort for the symbol machine (LLM) to reason.

[0035] Furthermore, based on the event graph and time sequence chain, identifying the causal chain of clinical decisions includes: Based on the event graph of clinical decisions, identifying the causal chain of clinical decisions through the attention network.

[0036] Specifically, based on the aforementioned defined input, the input tensor is represented as:

[0037] X = x text , x lab , x img , x proc ,

[0038] where x text , x lab , x img , x proc respectively correspond to the input text reports, experimental records, images, diagnostic records, etc.

[0039] The learning process of the second intelligent agent is represented as:

[0040] ,

[0041] Event graph attention network for decision chain reasoning:

[0042] ,

[0043] in W,d These are the weights and distance parameters of the attention network;

[0044] The image vector is represented as:

[0045] ,

[0046] Where i represents the ordinal number of the patient or sample, and j represents the ordinal number of the data related to the patient's corresponding medical behavior;

[0047] The BERT-base has 12 layers for parsing medical records; the number of event nodes n defaults to 32, covering common diagnostic and treatment actions; the graph attention threshold τ is 0.15, and edges below the threshold are pruned.

[0048] It is understandable that the above steps achieve AI-based deep understanding and structured extraction of medical texts.

[0049] In step S300 of some embodiments of the present invention, the compliance review of the medical behavior to be reviewed by the second intelligent agent based on the patient's medical profile and compliance dataset includes:

[0050] S301. Based on multiple diagnostic codes or multiple treatment codes in the patient's medical profile, retrieve the constraint set of the compliant dataset; the constraint set includes medical guidelines, payment rules, and operational constraints;

[0051] Specifically, patient data is initially matched against static rules in a rule base. For complex, ambiguous, or rule-limited cases, the AI ​​model leverages its learned medical knowledge and pattern recognition capabilities to reason within the specific clinical context of the patient. For example, it determines whether the current diagnosis is sufficient to support the requested / executed treatment intensity, whether test results truly indicate the need for such treatment, and whether the patient's comorbidities or allergies constitute a risk or contraindication to the specific treatment.

[0052] Multi-model cross-validation is employed, integrating various AI models to analyze and validate medical behavior data from multiple angles and dimensions. This improves the accuracy and reliability of regulatory analysis results and reduces the limitations and risk of misjudgment associated with single models. In the cross-validation mechanism, each model independently processes the same data source, and then achieves accurate judgment through evidence chain fusion. The rule engine model pre-sets quantity consistency rules (the quantity of consumables on the billing list ≤ the quantity prescribed by the doctor and ≥ the quantity used in the surgical record) and specification matching rules (the specifications of consumables prescribed by the doctor must be consistent with the implant parameters in the surgical record). For example, in the "knee replacement surgery" scenario, it automatically compares whether the specifications of the "customized femoral prosthesis" in the doctor's order, the "prosthesis implantation size record" in the surgical record, and the "high-value consumable code" on the billing list completely match.

[0053] S302. Input the matched constraint set and the medical behavior to be reviewed into the second intelligent agent to obtain a three-state decision and the confidence level of the three-state decision; the three-state decision includes compliance, non-compliance and pending.

[0054] For items marked "pending (questionable)" or "violation", the model will generate interpretable evidence pointing to specific rule clauses or logical contradictions in clinical reasoning (such as "the patient has no history of diabetes, blood sugar is normal, and there is insufficient necessity to apply for glycated hemoglobin testing").

[0055] Specifically, the behavior to be audited, b, and the rule constraint, c, are input into the logic rule engine and LLM interpreter. A three-state decision {OK, Warning, Violation} is generated, along with a confidence level, s. The structured result (b, status, s, explanation) is written to the compliance audit system or triggers manual review.

[0056] Furthermore, the step of inputting the matched constraint set and the medical behavior to be reviewed into the second intelligent agent to obtain the three-state decision and the confidence level of the three-state decision includes: calculating the compliance score of the three-state decision by the semantic similarity between the medical behavior to be reviewed and the matched constraint set, and the probability of the second intelligent agent judging compliance based on the matched constraint set.

[0057] Specifically, the compliance score is expressed as:

[0058] ,

[0059] The matching constraint (RuleMatch) takes values ​​∈ {0, 0.5, 1} (complete mismatch / partial / complete match), and LLM_prob represents the softmax probability of the large language model generating a "compliant" token.

[0060] The confidence mapping S is represented as:

[0061] ,

[0062] σ ( x () represents the activation function. Typical values: λ 1 = 0.6, λ 2 = 0.4, γ = 12, μ = 0.5.

[0063] It is understandable that a rule base is constructed, based on national medical insurance policies, clinical practice guidelines, drug instructions, hospital regulations, etc., to build a multi-dimensional compliance rule base. The AI ​​reasoning and review model inputs the extracted structured patient diagnosis and treatment profiles into the AI ​​review model. Model input includes patient characteristics, diagnosis, examination and test results, planned / implemented treatment behaviors, key time-series information, and treatment logic fragments mined from text. Model processing involves initially matching patient data with static rules in the rule base. For complex, ambiguous, or rule-overwhelming cases, the AI ​​model utilizes its learned medical knowledge and pattern recognition capabilities, combined with the patient's specific clinical context, to perform reasoning. Risk scoring and violation determination: The model outputs a compliance judgment result for each medical behavior to be reviewed and provides a confidence score. For questionable or violation items, the model generates interpretable evidence pointing to specific rule clauses or inferred clinical logical contradictions.

[0064] In step S500 of some embodiments of the present invention, the method further includes: reviewing real-time medical data through a second intelligent agent and issuing different warnings based on the stage of the medical behavior.

[0065] Specifically, pre-approval involves acquiring data in real time, combining historical data with current patient information, and triggering an AI-powered review model for immediate analysis. The model performs rapid reasoning and rule matching based on existing information and the intended operation. Real-time feedback of the review data and early warnings are issued to prevent non-compliant medical practices. In-hospital review combines real-time data generated during hospitalization; the AI ​​model dynamically assesses whether new risks have emerged and issues warnings. Post-hospital review involves a comprehensive AI review of all relevant medical actions and records after the treatment is completed, before cost settlement, or before medical insurance application. In-depth analysis and cross-validation are performed using the most complete treatment information data to identify potentially overlooked or difficult-to-detect violations before or during treatment. Statistical analysis is used to deeply analyze and mine data from the medical behavior analysis database.

[0066] Specifically, based on the above steps, data feedback and model optimization are implemented, including: inputting review results as important feedback data; establishing a continuous learning mechanism for the model, using manually labeled review result samples to regularly retrain and fine-tune the AI ​​model, continuously improving its accuracy, coverage, and interpretability; and dynamically maintaining and expanding the rule base based on policy updates, changes in clinical practice, and potential new problem patterns discovered by the AI ​​model.

[0067] In a specific embodiment of the present invention, the following steps are included: S1, a multi-source heterogeneous medical data access and fusion module; S2, AI-based deep understanding of medical text and structured information extraction; S3, a dynamic compliance rule engine and AI inference model; S4, full-process review triggering and execution; S5, data feedback and model optimization.

[0068] Example 2

[0069] refer to Figure 3 In a second aspect, the present invention provides a medical behavior compliance review system 1 based on an AI model, comprising: an acquisition module 11, configured to acquire multi-source heterogeneous medical data and medical compliance data, and standardize the medical data; constructing a medical dataset and a compliance dataset based on the standardized medical data and medical compliance data, respectively; an output module 12, configured to, based on event graph parsing, parse examination reports, identify treatment behaviors and clinical decision-making logic from the standardized medical data through a first intelligent agent, and output a patient treatment profile; and an review module 13, configured to, based on the patient treatment profile and the compliance dataset, perform compliance review on the medical behavior to be reviewed through a second intelligent agent; the compliance review includes compliance judgment and outputting the confidence level of the judgment result.

[0070] Furthermore, the output module 12 includes: a parsing unit, used to parse examination reports from the standardized medical data through a first intelligent agent based on a preset prompt word template; a matching unit, used to match the medical dataset with the first intelligent agent based on an enhanced retrieval matching method to form a time-series chain; and an identification unit, used to identify the causal chain of clinical decisions based on the event graph and the time-series chain.

[0071] Example 3

[0072] refer to Figure 4 A third aspect of the present invention provides an electronic device comprising: one or more processors; and a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method of the first aspect of the present invention.

[0073] Electronic device 500 may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 501, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 502 or a program loaded from storage device 508 into random access memory (RAM) 503. The RAM 503 also stores various programs and data required for the operation of electronic device 500. The processing unit 501, ROM 502, and RAM 503 are interconnected via bus 504. An input / output (I / O) interface 505 is also connected to bus 504.

[0074] Typically, the following devices can be connected to I / O interface 505: input devices 506 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 507 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 508 including, for example, hard disks; and communication devices 509. Communication device 509 allows electronic device 500 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 4 An electronic device 500 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively. Figure 4 Each box shown can represent a device or multiple devices as needed.

[0075] Specifically, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 509, or installed from a storage device 508, or installed from a ROM 502. When the computer program is executed by a processing device 501, it performs the functions defined in the methods of embodiments of this disclosure. It should be noted that the computer-readable medium described in embodiments of this disclosure can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In embodiments of this disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In embodiments of this disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. Program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0076] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device. The aforementioned computer-readable medium carries one or more computer programs, which, when executed by the electronic device, cause the electronic device to:

[0077] Computer program code for performing the operations of embodiments of this disclosure can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages—such as Java, Smalltalk, C++, and Python—and conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0078] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0079] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A medical behavior compliance review method based on an AI model, characterized in that, include: Acquire multi-source heterogeneous medical data and medical compliance data, and standardize the medical data; Based on standardized medical data and medical compliance data, medical datasets and compliance datasets are constructed respectively. Based on event graph parsing, the first intelligent agent parses examination reports, identifies treatment behaviors and clinical decision-making logic from the standardized medical data, and outputs a patient treatment profile: Based on a preset prompt word template, the first intelligent agent parses examination reports from the standardized medical data. Based on the enhanced retrieval and matching method, the first agent matches the medical dataset to form a time-series chain; based on the event graph and the time-series chain, the causal chain of clinical decision-making is identified. Based on patient medical profiles and compliance datasets, a second intelligent agent performs compliance audits on the medical behaviors to be reviewed: Based on multiple diagnostic codes or multiple treatment codes in the patient medical profile, a constraint set from the compliance dataset is retrieved; the constraint set includes treatment guidelines, payment rules, and operational constraints; the matched constraint set and the medical behaviors to be reviewed are input into the second intelligent agent to obtain a three-state decision and the confidence level of the three-state decision; the three-state decision includes compliant, non-compliant, and pending; the compliance audit includes a compliance judgment and outputting the confidence level of the judgment result.

2. The medical behavior compliance review method based on an AI model according to claim 1, characterized in that, The identification of causal chains for clinical decisions based on event graphs and time sequences includes: Based on the event graph of clinical decision-making, the causal chain of clinical decisions is identified through attention networks.

3. The medical behavior compliance review method based on an AI model according to claim 1, characterized in that, The process of inputting the matched constraint set and the medical behavior to be reviewed into the second intelligent agent to obtain a three-state decision and the confidence level of the three-state decision includes: The compliance score of the three-state decision is calculated by using the semantic similarity between the medical behavior to be reviewed and the set of constraints after matching, as well as the probability of the second agent judging compliance based on the set of constraints after matching. Based on the compliance score, the confidence level of the three-state decision is calculated using an activation function.

4. The medical behavior compliance review method based on an AI model according to claim 1, characterized in that, Also includes: A second intelligent agent reviews real-time medical data and issues different warnings based on the stage of the medical action.

5. A medical behavior compliance review system based on an AI model, characterized in that, include: The acquisition module is used to acquire multi-source heterogeneous medical data and medical compliance data, and to standardize the medical data. Based on standardized medical data and medical compliance data, medical datasets and compliance datasets are constructed respectively. The output module, based on event graph parsing, uses a first intelligent agent to parse examination reports from the standardized medical data, identify treatment behaviors and clinical decision-making logic, and output a patient treatment profile: based on a preset prompt word template, the first intelligent agent parses examination reports from the standardized medical data; Based on the enhanced retrieval and matching method, the first agent matches the medical dataset to form a time-series chain; based on the event graph and the time-series chain, the causal chain of clinical decision-making is identified. The review module is used to conduct compliance review of the medical behavior to be reviewed based on the patient's medical profile and compliance dataset, through a second intelligent agent: Based on multiple diagnostic codes or multiple treatment codes in the patient's medical profile, it retrieves the constraint set of the compliance dataset; the constraint set includes treatment guidelines, payment rules, and operational constraints; the matched constraint set and the medical behavior to be reviewed are input into the second intelligent agent to obtain a three-state decision and the confidence level of the three-state decision; the three-state decision includes compliant, non-compliant, and pending; the compliance review includes compliance judgment and outputting the confidence level of the judgment result.

6. An electronic device, comprising: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the AI-based medical behavior compliance review method as described in any one of claims 1 to 4.

7. A computer-readable medium having a computer program stored thereon, wherein, When the computer program is executed by the processor, it implements the AI-based medical behavior compliance review method as described in any one of claims 1 to 4.

Images