A medical behavior compliance-oriented deep semantic analysis method and system

Abstract

This invention relates to a deep semantic analysis method and system for medical practice compliance. The method includes: acquiring multi-source heterogeneous medical data and medical compliance data; standardizing the medical data and medical compliance data; training a first AI model and a second AI model to judge the compliance of illegal medical practices based on the standardized medical data and clinical semantic elements; acquiring patient medical data and multimodal appeal materials in real time; dynamically prioritizing the appeal materials using a multi-factor weighting method based on the trained first AI model and second AI model; identifying evidence of medical practices from the multimodal appeal materials based on the priority classification results using a third AI model; and structuring the evidence of medical practices. This invention, through the collaboration of a rule base and interpretable AI, achieves multi-level judgment and accountability for illegal medical practices, improving the accuracy and automation of medical practice compliance review.

Description

Technical Field

[0001] This invention belongs to the field of medical information and artificial intelligence technology, specifically relating to a deep semantic analysis method and system for medical behavior compliance; and more particularly to a deep semantic analysis method and system for medical behavior compliance in the context of medical insurance appeals. Background Technology

[0002] Accurate determination of the compliance of medical practices is the core basis for standardizing medical service practices, ensuring the safe and efficient operation of the medical insurance fund, and a key foundation for handling related appeals. Currently, the mainstream compliance review model still relies heavily on manual operation. Reviewers typically need to access and compare massive amounts of data across multiple heterogeneous medical information systems, which has significant limitations: low review efficiency, strong subjectivity in review results, and insufficient ability to handle complexity and massive amounts of data. Although some existing intelligent review systems have introduced automation technology and rely on preset rules for preliminary screening, their capabilities are also quite limited: the rule system lacks flexibility, has limited understanding of complex medical practices, and the review model is mainly ex post and incomplete.

[0003] The existing technology has three main shortcomings: First, traditional manual review relies on manual operation across systems (such as HIS and medical insurance platforms), which is inefficient and easily affected by subjective factors. Second, automated systems can only perform static rule matching based on structured billing data and lack the ability to parse unstructured clinical texts (such as medical records), resulting in an inability to understand the diagnosis and treatment logic, assess the rationality of the plan, and evaluate the necessity of medical treatment. Third, the full-process coverage is insufficient, focusing on post-event review while lacking in-process monitoring and pre-event warning. Furthermore, the appeal process relies on manually prepared materials, which lack professionalism and leads to a high failure rate for appeals. Summary of the Invention

[0004] To automate and improve the accuracy of medical practice compliance review, a first aspect of this invention provides a deep semantic analysis method for medical practice compliance, comprising: acquiring multi-source heterogeneous medical data and medical compliance data; standardizing the medical data and medical compliance data; extracting clinical semantic elements from the standardized medical data based on natural language processing methods; training a first AI model for identifying non-compliant medical practices based on the medical compliance data; training a second AI model for analyzing non-compliant medical practices and judging their compliance based on the standardized medical data and clinical semantic elements; acquiring patient medical data and multimodal appeal materials in real time; dynamically prioritizing the appeal materials using a multi-factor weighting method based on the trained first and second AI models; identifying medical practice evidence from the multimodal appeal materials using a third AI model based on the priority classification results; and structuring the medical practice evidence.

[0005] In some embodiments of the present invention, the first AI model for identifying illegal medical practices based on medical compliance data includes: training the first AI model for identifying illegal medical practices based on medical policies and regulations, clinical knowledge bases, and historical case databases.

[0006] In some embodiments of the present invention, the step of training a second AI model for parsing non-violation medical behavior and judging the compliance of non-violation medical behavior based on standardized medical data and clinical semantic elements includes: constructing a medical knowledge graph based on standardized medical data and clinical semantic elements; and constructing and training a clinical semantic understanding layer and a causal reasoning module through the medical knowledge graph.

[0007] In some embodiments of the present invention, the dynamic priority classification of the appeal materials based on a multi-factor weighted method using a first AI model and a second AI model that have been trained respectively includes: analyzing the violation markers, violation rates, and cost deviations of the appeal materials based on the first AI model that has been trained; determining the risk confidence level of the appeal materials based on the second AI model that has been trained; and calculating the dynamic priority score of the appeal materials using the entropy weight method based on the violation markers, violation rates, cost deviations, and risk confidence levels of the appeal materials.

[0008] Furthermore, the step of calculating the dynamic priority score of the appeal materials using the entropy weight method based on the violation markers, violation rates, cost deviations, and risk confidence levels of the appeal materials includes: constructing a multi-factor dynamic scoring matrix based on the violation markers, violation rates, cost deviations, and risk confidence levels of the appeal materials; determining the weight of each factor in the scoring matrix using the entropy weight method; and calculating the dynamic priority score of the appeal materials based on the score and weight of each factor.

[0009] In the above embodiments, the step of identifying medical behavior evidence from multimodal complaint materials using a third AI model includes: detecting clinical entity boundaries, analyzing test reports, and locating key objects in the reports using a third AI model.

[0010] A second aspect of the present invention provides a deep semantic analysis system for medical behavior compliance, comprising: an acquisition module for acquiring multi-source heterogeneous medical data and medical compliance data; standardizing the medical data and medical compliance data; extracting clinical semantic elements from the standardized medical data based on natural language processing methods; a training module for training a first AI model for identifying illegal medical behaviors based on medical compliance data; and training a second AI model for analyzing non-illegal medical behaviors and judging the compliance of non-illegal medical behaviors based on the standardized medical data and clinical semantic elements; a classification module for acquiring patient medical data and multimodal appeal materials in real time; dynamically prioritizing the appeal materials using a multi-factor weighting method based on the trained first AI model and second AI model; and an identification module for identifying medical behavior evidence from the multimodal appeal materials based on the priority classification results using a third AI model; and structuring the medical behavior evidence.

[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 deep semantic analysis method for medical behavior compliance 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 deep semantic analysis method for medical behavior compliance provided in the first aspect of the present invention.

[0013] The beneficial effects of this invention are:

[0014] (1) By integrating multi-source heterogeneous data through a “unified data lake”, and using NLP deep analysis to overcome the bottlenecks in both structured and unstructured data processing, a foundation for all-dimensional analysis is laid.

[0015] (2) Collaboration between rule base and explainable AI: The rule engine quickly captures explicit violations (such as drug overdose), and the AI ​​engine analyzes implicit logic to achieve a dynamic balance between the "Diagnosis and Treatment Guidelines" and "Clinical Practice".

[0016] (3) Three-level dynamic rating drives resource optimization and allocation + full-link coverage before, during and after the event: blocking violations before the event, intervening in risks in real time during the event, and accurately tracing and assigning responsibility after the event.

[0017] (4) Evidence location: The large model accurately extracts key medical arguments in the medical record. Sentence generation: Based on clinical logic, professional review documents are automatically generated, eliminating the loopholes of manual writing. Attached Figure Description

[0018] Figure 1 This is a basic flowchart of a deep semantic analysis method for medical behavior compliance in some embodiments of the present invention.

[0019] Figure 2 This is a schematic diagram illustrating the specific process of a deep semantic analysis method for medical behavior compliance in some embodiments of the present invention.

[0020] Figure 3 This is a schematic diagram of the structure of a deep semantic analysis system for medical behavior compliance in some embodiments of the present invention;

[0021] Figure 4 This is a schematic diagram of the structure of an electronic device in some embodiments of the present invention. Detailed Implementation

[0022] 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.

[0023] refer to Figure 1 and Figure 2 In a first aspect of the present invention, a deep semantic analysis method for medical behavior compliance is provided, comprising: S100. acquiring multi-source heterogeneous medical data and medical compliance data; standardizing the medical data and medical compliance data; extracting clinical semantic elements from the standardized medical data based on natural language processing methods; training a first AI model for identifying illegal medical behaviors based on the medical compliance data; training a second AI model for parsing non-illegal medical behaviors and judging the compliance of non-illegal medical behaviors based on the standardized medical data and clinical semantic elements; acquiring patient medical data and multimodal appeal materials in real time; dynamically prioritizing the appeal materials using a multi-factor weighting method based on the trained first AI model and second AI model; identifying medical behavior evidence from the multimodal appeal materials using a third AI model based on the priority classification results; and structuring the medical behavior evidence.

[0024] It should be noted that the above-mentioned first AI model, second AI model or third AI model only represent functional division. Those skilled in the art can integrate or separate the first AI model, second AI model or third AI model based on the training or fine-tuning of large language models (such as the Qianwen series models, Deepseek series models, Kimi series models, etc.).

[0025] In step S100 of some embodiments of the present invention, training a first AI model for identifying illegal medical practices based on medical compliance data includes: training a first AI model for identifying illegal medical practices based on medical policies and regulations, a clinical knowledge base, and a historical case database.

[0026] In some embodiments of the present invention, step S100 includes:

[0027] S101. Acquire multi-source heterogeneous medical data and medical compliance data;

[0028] Specifically, through standardized interfaces, data interaction and integration are achieved among heterogeneous systems such as core business data (HIS hospital information system), medical technology data (LIS laboratory system / PACS imaging system), management data (EMR electronic medical records / medical insurance platform), and cost data (billing system / consumables management system). This unifies the management of multi-source heterogeneous data and constructs a medical data lake. When medical data is incorporated into the unified access layer from the data source, it undergoes data cleaning and preprocessing, including format conversion (such as date standardization, gender coding, etc.), terminology alignment (such as disease code standardization), entity linking, and other operations, ultimately constructing a complete patient-centered view of diagnosis and treatment data.

[0029] S102. Standardize the aforementioned medical data and medical compliance data;

[0030] After medical data is incorporated into a unified access layer from various source systems (such as HIS, LIS, EMR, etc.) through standardized interfaces, a rigorous data governance process is immediately implemented. Data mapping rules and ETL (Extract-Transform-Load) scripts are applied to forcibly transform heterogeneous data into a unified format for the target data lake. Date / time strings from different sources, such as "20230708", "08 / 07 / 2023", and "2023-07-08 14:30:00", are uniformly converted to the standard format YYYY-MM-DD (HH:mm:ss). Gender descriptions, such as "Male", "Male", and "1", are uniformly mapped to standard codes, such as "M" for male, "F" for female, and "U" for unknown.

[0031] Using medical terminology mapping and coding recognition algorithms, key structured fields are parsed and extracted from patient diagnosis and treatment data, including basic patient information, diagnostic codes (ICD), surgical procedure codes (ICD-9-CM-3 / ICD-10-PCS), drug codes, chargeable items, and examination and test requests and results (numerical type).

[0032] S103. Extract clinical semantic elements from standardized medical data based on natural language processing methods;

[0033] Specifically, NLP technology (medical entity recognition, relation extraction and event analysis) is used to accurately extract core clinical semantic elements (such as symptoms, signs, diagnostic basis, key examination results, treatment measures and their temporal relationships) from unstructured texts such as medical records and examination reports, providing structured input for subsequent analysis.

[0034] In step S200 of some embodiments of the present invention, the following is included:

[0035] S201. Based on medical compliance data, train the first AI model to identify unethical medical practices;

[0036] Specifically, the rule base engine judges the compliance of medical practices based on clear policies, regulations and clinical knowledge bases, and forms a historical case library by accumulating AI cases, thereby realizing the dynamic expansion of the rule base content.

[0037] S202. Based on standardized medical data and clinical semantic elements, train a second AI model to analyze non-violation medical behaviors and determine the compliance of non-violation medical behaviors.

[0038] Specifically, a large AI model is used, and professional corpora such as medical knowledge graphs and clinical pathway guidelines are used for pre-training to construct a clinical semantic understanding layer and a causal reasoning module for "symptoms → examination → diagnosis → treatment".

[0039] Step S202 includes: constructing a medical knowledge graph based on standardized medical data and clinical semantic elements; and constructing and training a clinical semantic understanding layer and a causal reasoning module through the medical knowledge graph. The input data of the AI ​​big data model consists of patient-related information extracted from the medical data lake, including basic information, diagnostic descriptions, medical records, examination reports, etc.

[0040] It is understandable that the first AI model and the second AI model constitute a dual-engine architecture of "explainable AI" + "rule base"; the rule base engine is responsible for handling violations with clear rules (such as drug overdose and contraindication conflicts); for cases not covered by the rule base, the explainable AI engine performs medical behavior reasoning and compliance judgment by analyzing ambiguous scenarios (such as the rationality of indications and the necessity of treatment).

[0041] In some embodiments of the present invention, the dynamic priority classification of the appeal materials based on a multi-factor weighted method using a first AI model and a second AI model that have been trained respectively includes: S301. Analyzing the violation markers, violation rates, and cost deviations of the appeal materials based on the trained first AI model; S302. Determining the risk confidence level of the appeal materials based on the trained second AI model; S303. Calculating the dynamic priority score of the appeal materials using the entropy weight method based on the violation markers, violation rates, cost deviations, and risk confidence levels of the appeal materials.

[0042] Furthermore, in step S303, the calculation of the dynamic priority score of the appeal material based on the violation mark, violation rate, cost deviation, and risk confidence level of the appeal material using the entropy weight method includes: constructing a multi-factor dynamic scoring matrix based on the violation mark, violation rate, cost deviation, and risk confidence level of the appeal material; determining the weight of each factor in the scoring matrix using the entropy weight method; and calculating the dynamic priority score of the appeal material based on the score and weight of each factor.

[0043] Specifically, the dynamic rating of appeal priorities adopts a multi-factor weighted scoring model, which constructs a dynamic scoring matrix based on four dimensions: rule engine violation markers (such as the number of times drugs were overdose), AI engine risk confidence (a continuous value of 0-1), patient / doctor historical violation rate (sliding window statistics), and cost deviation (|actual cost - medical insurance standard| / medical insurance standard). The factor weights are automatically calculated using the entropy weight method (EWM) (e.g., confidence weight 0.6, cost deviation weight 0.3).

[0044] A three-tiered classification mechanism is used to dynamically grade the model's judgment results: high priority (suspected violation, requiring priority processing), medium-low priority (confirmed violation), and automatic approval (compliant with usage guidelines). Simultaneously, judgment criteria are generated based on case data, and relevant key medical behavioral information is listed.

[0045] It is understandable that the above training or case analysis results may be used to conduct risk scoring, and dynamic rating may be carried out based on the scores.

[0046] In step S400 of the above embodiments, based on the priority classification results, medical behavior evidence is identified from the multimodal appeal materials using a third AI model; and the medical behavior evidence is structured.

[0047] Specifically, multimodal key point recognition employs an intelligent document pipeline to achieve evidence localization and highlighting. Evidence localization utilizes a large-scale model to intelligently analyze relevant document content (textual evidence, tabular evidence, and image evidence). Through AI-powered medical text recognition, clinical entity boundary detection is achieved. A Table Transformer architecture is used to parse laboratory reports, and the AI-powered model's built-in visual model is used to locate key features in the reports. Highlighting is achieved through a Python highlighting engine that supports DOM tree modification in PDF / Word documents (using pdfminer + python-docx) to implement cross-modal association tagging (e.g., linking the pathology report text "PD-L1<1%" with the immunohistochemistry image region for highlighting). The output includes evidence fragment packages with spatiotemporal stamps (<document page number, coordinates, confidence level> triples) to support subsequent appeal generation.

[0048] Finally, the appeal statement generation process includes: input data, including patient data, medical project information, and intelligent review results. The rule template engine dynamically populates structured elements (patient condition, medical necessity, clinical decision basis) to ensure 100% coverage of existing medical insurance rule base items; it fine-tunes the large language model (training data: historical appeal documents + clinical guidelines) to handle new / complex items, generating personalized text that conforms to the diagnosis and treatment logic chain (symptoms → examination → treatment → compliance justification).

[0049] Engineering extensions: Supports manually adding new project rules and synchronizing the template library through the configuration interface; multiple projects are automatically clustered according to the diagnosis and treatment timeline to generate coherent text; when the text is too long, it enables modular segmentation generation of "diagnosis-treatment-evidence" and outputs a standardized JSON structure (including text and evidence location index).

[0050] In particular, manual review is added based on the above steps to improve the accuracy of the model and reduce illusions.

[0051] Based on the data lake (multi-source heterogeneous medical data and medical compliance data), medical behavior is assessed using inpatient admission diagnosis data and historical diagnosis data, and the assessment results are quickly fed back to standardize medical practices. In-process review: During a patient's hospitalization, based on real-time data from the medical data lake and dynamic data generated during hospitalization, compliance assessment of medical behavior is conducted to dynamically evaluate whether any violations have occurred. Post-process review: After the patient's treatment is completed and before medical insurance settlement, a final compliance assessment of medical behavior is conducted based on complete medical record data.

[0052] Preferably, manual calibration is employed. Cases where model assessments show bias or errors are corrected through manual review. The calibration results serve as valuable feedback data to improve the model's judgment mechanism. Rule updates are implemented regularly based on policy changes, updates to clinical practice, and feedback from manual calibration. The review rule base is then updated accordingly, and the model's analysis rules are adjusted to achieve iterative optimization and upgrades.

[0053] In a specific embodiment of the present invention, the following steps are included: S1, integration of multi-source medical data; S2, semantic parsing of medical text and extraction of structured / unstructured data; S3, dual-engine architecture of "explainable AI" + "rule base"; S4, dynamic rating of appeal priority; S5, multimodal key point recognition; S6, generation of appeal statements; S7, manual review; S8, model feedback and optimization.

[0054] Example 2

[0055] refer to Figure 3In a second aspect, the present invention provides a deep semantic analysis system 1 for medical behavior compliance, comprising: an acquisition module 11 for acquiring multi-source heterogeneous medical data and medical compliance data; standardizing the medical data and medical compliance data; extracting clinical semantic elements from the standardized medical data based on natural language processing methods; a training module 12 for training a first AI model for identifying illegal medical behavior based on medical compliance data; and training a second AI model for analyzing non-illegal medical behavior and judging the compliance of non-illegal medical behavior based on the standardized medical data and clinical semantic elements; a classification module 13 for acquiring patient medical data and multimodal appeal materials in real time; dynamically prioritizing the appeal materials using a multi-factor weighting method based on the trained first AI model and second AI model; and an identification module 14 for identifying medical behavior evidence from the multimodal appeal materials based on the priority classification results using a third AI model; and structuring the medical behavior evidence.

[0056] Furthermore, the classification module 13 includes: an analysis unit, used to analyze the violation markers, violation rates, and cost deviations of the appeal materials based on a first AI model that has been trained; a determination unit, used to determine the risk confidence level of the appeal materials based on a second AI model that has been trained; and a calculation unit, used to calculate the dynamic priority score of the appeal materials using the entropy weight method based on the violation markers, violation rates, cost deviations, and risk confidence levels of the appeal materials.

[0057] Example 3

[0058] 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.

[0059] 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.

[0060] 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.

[0061] 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 thereof. 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.

[0062] 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:

[0063] 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).

[0064] 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.

[0065] 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 action compliance-oriented deep semantic analysis method, characterized in that, The method comprises the following steps: acquiring multi-source heterogeneous medical data and medical compliance data; standardizing the medical data and the medical compliance data; extracting clinical semantic elements from the standardized medical data based on a natural language processing method; training a first AI model for identifying illegal medical behaviors based on the medical compliance data; training a second AI model for analyzing non-illegal medical behaviors and judging the compliance of the non-illegal medical behaviors based on the standardized medical data and the clinical semantic elements; acquiring medical data and multi-modal complaint materials of a patient in real time; performing dynamic priority classification on the complaint materials by a multi-factor weighting method based on the trained first AI model and the second AI model; analyzing illegal labels, illegal rates and cost deviation degrees of the complaint materials based on the trained first AI model; determining risk confidence of the complaint materials based on the trained second AI model; calculating a dynamic priority score of the complaint materials by an entropy weight method based on the illegal labels, the illegal rates, the cost deviation degrees and the risk confidence of the complaint materials; identifying medical behavior evidence from the multi-modal complaint materials by a third AI model based on the priority classification result; and structuring the medical behavior evidence.

2. The medical action compliance oriented deep semantic analysis method of claim 1, wherein, The step of training the first AI model for identifying illegal medical behaviors based on the medical compliance data comprises the following steps: training the first AI model for identifying illegal medical behaviors based on medical policies and regulations, a clinical knowledge base and a historical case base.

3. The medical action compliance oriented deep semantic analysis method of claim 1, wherein, The step of training the second AI model for analyzing non-illegal medical behaviors and judging the compliance of the non-illegal medical behaviors based on the standardized medical data and the clinical semantic elements comprises the following steps: constructing a medical knowledge graph based on the standardized medical data and the clinical semantic elements; constructing and training a clinical semantic understanding layer and a causal reasoning module through the medical knowledge graph.

4. The medical action compliance oriented deep semantic analysis method of claim 1, wherein, The step of calculating the dynamic priority score of the complaint materials by the entropy weight method based on the illegal labels, the illegal rates, the cost deviation degrees and the risk confidence of the complaint materials comprises the following steps: constructing a multi-factor dynamic scoring matrix based on the illegal labels, the illegal rates, the cost deviation degrees and the risk confidence of the complaint materials; determining the weight of each factor in the scoring matrix by the entropy weight method; calculating the dynamic priority score of the complaint materials based on the score and the weight of each factor.

5. The medical action compliance oriented deep semantic analysis method of claim 1, wherein, The step of identifying medical behavior evidence from the multi-modal complaint materials by the third AI model comprises the following steps: detecting clinical entity boundaries, analyzing test reports and positioning report key objects from the multi-modal complaint materials by the third AI model.

6. A medical action compliance oriented deep semantic analysis system, characterized in that, The method comprises the following steps: an acquisition module, configured to acquire multi-source heterogeneous medical data and medical compliance data; standardizing the medical data and the medical compliance data; extracting clinical semantic elements from the standardized medical data based on a natural language processing method; a training module, configured to train a first AI model for identifying illegal medical behaviors based on the medical compliance data; and train a second AI model for analyzing non-illegal medical behaviors and judging the compliance of the non-illegal medical behaviors based on the standardized medical data and the clinical semantic elements; The classification module is configured to acquire medical data and multi-modal complaint materials of a patient in real time, perform dynamic priority classification on the complaint materials by a multi-factor weighting method based on the trained first AI model and the second AI model, analyze rule violation marks, rule violation rates and cost deviation degrees of the complaint materials based on the trained first AI model, determine risk confidence of the complaint materials based on the trained second AI model, and calculate a dynamic priority score of the complaint materials by an entropy weight method based on the rule violation marks, the rule violation rates, the cost deviation degrees and the risk confidence of the complaint materials. The identification module is configured to identify medical behavior evidence from the multi-modal complaint materials by a third AI model based on the priority classification result, and structure the medical behavior evidence. one or more processors; a storage device for storing one or more programs, which, when executed by the one or more processors, cause the one or more processors to implement the medical behavior compliance deep semantic analysis method according to any one of claims 1 to 5.

7. An electronic device, comprising: The computer program, when executed by a processor, implements the medical behavior compliance deep semantic analysis method according to any one of claims 1 to 5. ​ 8. A computer readable medium having stored thereon a computer program, wherein, ​

Images