Insurance claim judgment method and system, product, equipment and storage medium

Abstract

The invention discloses an insurance claim judgment method and system, a product, equipment and a storage medium, multi-dimensional information such as a text, a structure and a key identifier is synchronously extracted from image data uploaded by a user through a multi-modal large model, and a standardized doctor-seeing information data set is automatically generated, so that automation and structuralization of information input are realized, and the accuracy of information input is improved. The problems that the traditional OCR technology is incomplete in extraction and depends on manual checking are solved; furthermore, under the driving of the same multi-mode large model, deep correlation analysis is performed based on the data set and the insurance information, compensation qualification checking and compensation amount calculation are automatically completed, seamless connection and intelligent judgment from information extraction to auditing decision are realized, shallow comparison depending on manual preset rules is replaced, and the accuracy of compensation is improved. The core problems of auditing process splitting, rule solidification and insufficient utilization of multi-modal information in the prior art are solved, and the adaptive capacity and processing efficiency of complex doctor seeing scenes are remarkably improved.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to an insurance claims determination method, system, product, device and storage medium. Background Technology

[0002] Currently, the automated review process in insurance claims mainly relies on rule engines based on manual settings. These rule engines typically use statically configured judgment logic and thresholds, and their processing capabilities are essentially limited to comparing and matching structured and standardized data, such as simply comparing entered disease codes with insurance liability lists.

[0003] However, the actual business scenarios of claims review involve a large amount of unstructured and multimodal information, such as images of medical documents containing handwriting, stamps, and complex formats, as well as insurance contract texts with complex wording and room for interpretation. Existing rule engines struggle to effectively extract, semantically understand, and perform correlation analysis on this information, and cannot support review tasks requiring deep reasoning, such as determining the correlation between disease descriptions and insurance liability, and analyzing the reasonableness of medical expense composition. Therefore, when faced with cases involving complex medical records, diverse document formats, or ambiguities in the terms and conditions, the system's automated processing capabilities are significantly insufficient, and final review still requires manual intervention. This limitation not only restricts the in-depth application of claims automation but also becomes a key bottleneck affecting overall processing efficiency and cost control. Summary of the Invention

[0004] In view of the above problems, this application provides an insurance claim determination method, system, product, device and storage medium.

[0005] The embodiments of this application disclose the following technical solutions:

[0006] The first aspect of this application provides a method for determining insurance claims, including:

[0007] Obtain image data uploaded by users and preprocess the image data;

[0008] Multimodal information is extracted from preprocessed image data based on a multimodal large model;

[0009] The multimodal information is entered into the corresponding fields according to the preset data format requirements to generate a standardized medical information dataset;

[0010] Based on the multimodal large model, medical information dataset, and user insurance information, the system verifies whether the treated disease is within the scope of insurance coverage and whether the hospital meets the requirements of the insurance terms to determine eligibility for compensation. Based on the multimodal large model, medical information dataset, and insurance information, the system identifies and removes expenses in the medical information dataset that are not covered by the insurance to determine the compensation amount.

[0011] Based on the eligibility for compensation and the amount of compensation, a claims assessment report is generated.

[0012] In one possible implementation, the multimodal information includes textual information, structural information, and key identification information; the textual information represents the core semantic content identified from the image data for claims determination; the structural information represents the layout and logical relationship of different information units in the image data; and the key identification information represents auxiliary visual elements used to verify the authenticity or authority of the image data.

[0013] In one possible implementation, the step of identifying and removing expenses from the medical information dataset that are not covered by reimbursement, based on a multimodal large model, a medical information dataset, and insurance information, to determine the reimbursement amount includes:

[0014] Based on the multimodal large model, the insurance claim clauses, insurance exclusion clauses and disease treatment information in the insurance information are extracted; based on the multimodal large model, the various cost information and total cost in the medical treatment information dataset are extracted.

[0015] The cost information in the medical information dataset is compared with the insurance reimbursement terms and disease treatment information.

[0016] Based on the comparison results, expenses unrelated to the illness treated in this case were identified, and / or expenses excluded by the exclusion clauses of the insurance policy.

[0017] The compensation amount is calculated by removing the identified expense items from the total cost.

[0018] In one possible implementation, the method further includes:

[0019] Based on the aforementioned medical information dataset, determine whether the current case is a special and complex case that requires manual review.

[0020] If a case is determined to be exceptionally complex, mark the current case as a manual review node.

[0021] The criteria for determining eligibility for compensation, the criteria for determining the amount of compensation, the eligibility for compensation, and the amount of compensation for the current case will be synchronized to the management terminal.

[0022] In one possible implementation, the management terminal includes a human-computer interaction interface, which displays the extraction process of the multimodal information and the determination process of the compensation eligibility and compensation amount. The method further includes:

[0023] The human-computer interaction interface is used to receive manual review and confirmation instructions or correction instructions.

[0024] Based on the received instructions, the claims assessment report is adjusted.

[0025] In one possible implementation, the text information includes at least one of the following: disease diagnosis name, department visited, hospital name, itemized cost amount, and visit time; the structural information includes at least one of the following: invoice layout, medical record paragraph modules, and cost classification table; the key identification information includes at least one of the following: seal, signature, hospital logo, or special symbol in the image data; and the insurance information includes at least one of the following: insurance product type, scope of insurance liability, exclusion clauses, and hospital level restrictions.

[0026] This application provides an insurance claims determination system in embodiment two, including:

[0027] The acquisition unit is used to acquire image data uploaded by the user and preprocess the image data.

[0028] The extraction unit is used to extract multimodal information from preprocessed image data based on a multimodal large model;

[0029] The generation unit is used to input the multimodal information into the corresponding fields according to the preset data format requirements, and generate a standardized medical information dataset.

[0030] The determination unit is used to verify, based on the multimodal large model, the medical information dataset, and the user's insurance information, whether the treated disease is within the scope of insurance liability and whether the treated hospital meets the requirements of the insurance terms, in order to determine the eligibility for compensation; and based on the multimodal large model, the medical information dataset, and the insurance information, to identify and remove expenses in the medical information dataset that are not within the scope of compensation, in order to determine the compensation amount.

[0031] The generation unit is used to generate a claims determination report based on the claim eligibility and the claim amount.

[0032] A third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements the insurance claim determination method as described in the first aspect above.

[0033] The fourth aspect of this application provides a computer program product, which, when run on a computer, executes the insurance claim determination method as described in the first aspect above.

[0034] A fifth aspect of this application provides a computer-readable storage medium storing instructions that, when executed on a terminal device, cause the terminal device to perform the insurance claim determination method as described in the first aspect above.

[0035] Compared with the prior art, this application has the following advantages:

[0036] Based on a multimodal large model, text, structure, and key identifiers are extracted synchronously from user-uploaded image data, and standardized medical information datasets are automatically generated. This automates and structures information entry, overcoming the problems of incomplete extraction and reliance on manual verification in traditional OCR technology. Furthermore, driven by the same multimodal large model, deep correlation analysis is performed between this dataset and insurance information to automatically complete the verification of claim eligibility and the calculation of claim amount. This achieves seamless connection and intelligent judgment from information extraction to review and decision-making, replacing the shallow comparison that relies on manually preset rules. It solves the core problems of fragmented review processes, rigid rules, and insufficient utilization of multimodal information in existing technologies, significantly improving the adaptability and processing efficiency for complex medical scenarios. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 A flowchart illustrating an insurance claim determination method provided in this application embodiment;

[0039] Figure 2 This is a structural diagram of an insurance claims determination system provided in an embodiment of this application. Detailed Implementation

[0040] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0041] To facilitate understanding of the technical solutions provided in the embodiments of this application, the background technology involved in the embodiments of this application will be described below.

[0042] As mentioned earlier, existing insurance claims processing technologies suffer from the following main shortcomings: In the information entry stage, traditional OCR technology can only recognize text and cannot understand structured information in images (such as medical record diagnoses and invoice itemization). Furthermore, it has poor adaptability to complex image formats such as blurry or tilted images, resulting in low entry efficiency, high error rates, and incomplete information extraction. In the review stage, the entry and review processes are disconnected. The review relies on manually preset static rules and cannot perform in-depth correlation analysis based on multimodal information such as medical record images and policy texts. It also has weak adaptability to complex medical scenarios. These problems collectively lead to high labor costs and long processing times for claims, severely impacting user experience.

[0043] To address the aforementioned issues, this application employs a multimodal large-scale model to replace traditional OCR technology, enabling integrated understanding and extraction of user-uploaded image data. This model not only recognizes text content within images but also simultaneously analyzes multi-dimensional information such as layout structure, logical connections, and key identifiers, automatically generating a standardized medical information dataset based on a preset data format. This process achieves end-to-end automated conversion from images to structured data, eliminating the need for manual data entry and verification, significantly improving data entry efficiency and accuracy, and ensuring the comprehensiveness and usability of the extracted information.

[0044] This application utilizes a multimodal large-scale model to integrate information extraction and review / decision-making into a continuous and seamless intelligent process. After completing information extraction, the model immediately performs deep correlation analysis on the medical information dataset and insurance information based on the same set of semantic understanding capabilities, automatically completing complex review tasks such as disease liability matching and cost reasonableness judgment. This approach overcomes the static limitations of traditional rule engines, achieving a leap from rule matching to semantic association and reasoning, thereby significantly enhancing its adaptability to complex and non-standardized medical scenarios.

[0045] Because this application automates the entire process from image uploading to claims report generation, and is capable of handling complex cases driven by a multimodal large model, it significantly reduces the need for manual intervention in information entry and review. This not only lowers the labor costs for insurance companies but also significantly shortens the overall claims processing cycle, thereby improving business efficiency and the claims experience for end users.

[0046] It should be noted that the insurance claim determination method, system, product, equipment, and medium provided in this application can be applied to the field of computer technology. The above are merely examples and do not limit the application areas of the insurance claim determination method, system, product, equipment, and medium provided in this application. Furthermore, the embodiments of this application may not limit the executing entity of the insurance claim determination. For example, the insurance claim determination method of this application embodiment can be applied to data processing devices such as terminal devices or servers. The terminal device can be an electronic device such as a computer or a personal digital assistant (PDA). The server can be a standalone server, a cloud server, or a cluster server composed of multiple servers.

[0047] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0048] The following example illustrates an insurance claim determination method provided in this application. See [link to example]. Figure 1 ,Should Figure 1 A flowchart of an insurance claim determination method provided in this application embodiment, the method including:

[0049] S101. Obtain the image data uploaded by the user and preprocess the image data.

[0050] We acquire complete image data related to claims submitted by users, and eliminate image defects and unify data formats through standardized preprocessing. This lays a high-quality data foundation for accurate information extraction from subsequent multimodal large models, ensuring efficient connection and data accuracy throughout the claims process.

[0051] In one possible implementation, receiving the claim-related image data uploaded by the user specifically refers to receiving one or more of the following: inpatient medical records, expense lists, and payment receipts uploaded by the user via an APP, a webpage, or offline scanning.

[0052] It supports multiple upload methods, including online self-upload by users through insurance apps and insurance company official websites, as well as offline scanning uploads (such as users bringing paper medical records and invoices to insurance company service outlets for staff to scan and enter into the system), covering different users' operating habits and scenario needs.

[0053] The received image data is the core material necessary for claims determination, specifically including inpatient medical records (including admission records, diagnosis certificates, discharge summaries, etc.), medical expense lists (including itemized treatment costs and drug cost details), payment receipts (including electronic receipts and scanned copies of paper receipts), etc., to ensure the completeness of the data and its relevance to claims.

[0054] When receiving images, the system can automatically perform basic validity verification, including file format compatibility verification (supporting common formats such as JPG, PNG, and PDF) and file integrity verification (to avoid uploading incomplete images), and provide feedback on the verification results to the user, prompting them to supplement missing or invalid information.

[0055] In one possible implementation, the preprocessing of the image data specifically includes uniform resolution adjustment and skew correction of the image data to eliminate the impact of image defects such as blurring and tilting on multimodal information extraction.

[0056] For example, images of different formats can be uniformly converted to a standard format suitable for the resolution of multimodal large models (such as uniformly converting to high-resolution JPG format), while the resolution of the images can be uniformly adjusted (such as adjusting to 300 DPI) to ensure the consistency of the large model's resolution of images from different sources.

[0057] To address potential image quality issues in real-world scenarios, targeted corrections are implemented: Deviation correction uses image geometric correction technology to correct tilted or distorted images (such as image shift caused by tilting the user's hand during shooting), ensuring horizontal alignment of text and structural information; Noise reduction and enhancement uses image enhancement algorithms to improve clarity and contrast in blurry, dark, or smudged images, removing background noise and irrelevant smudges to enhance the recognizability of text and key symbols; Edge cropping automatically crops irrelevant borders, blank areas, or redundant backgrounds (such as desktop backgrounds on scanned documents or irrelevant objects included during shooting), focusing on core information areas.

[0058] After preprocessing, the image quality is checked to see if it meets the requirements for multimodal information extraction. If there are still serious defects that cannot be corrected (such as severe blurring or occlusion of key information), a secondary prompt is triggered to guide the user to re-upload or optimize the image.

[0059] By uploading designs through multiple channels, user convenience is enhanced. Standardized preprocessing solves the problem of poor adaptability of traditional OCR technology to blurry, tilted, and multi-format images, effectively eliminating the interference of image defects on information extraction. At the same time, the preprocessed images have a unified format and controllable quality, so that subsequent multimodal large models do not need to handle format differences and image noise, and can directly focus on core information extraction, greatly improving the efficiency and accuracy of information extraction.

[0060] S102. Extract multimodal information from preprocessed image data based on a multimodal large model.

[0061] A multimodal large model (specifically Qwen2.5vl or Qwen3vl) with strong image resolution and cross-modal correlation capabilities is adopted. This model can be trained to adapt to various image features in insurance claims scenarios and can directly process preprocessed blurred, tilted, and multi-format images without the need for additional format conversion or feature optimization modules, ensuring the continuity and efficiency of information extraction.

[0062] The multimodal information includes textual information, structural information, and key identification information. The textual information represents the core business data generated during the medical visit, specifically including basic information that directly supports claims determination, such as the name of the disease diagnosis, the department visited, the hospital name, the itemized cost amount, and the visit time. The structural information represents the layout and information organization logic of the image data, specifically including the cost classification and layout structure of the invoices and the division of the diagnostic conclusion modules in the medical records, used to clarify the category and hierarchical relationship of the information. The key identification information represents specific markers used to strengthen the basis for claims determination, including the treatment item codes in the medical records, the medical insurance settlement identifiers in the invoices, and the hospital level identifiers, providing auxiliary basis for matching insurance liability and determining cost attributes.

[0063] The model prioritizes analyzing the overall structure of the image, dividing it into information modules. For example, it divides invoices into "Patient Information Area," "Fee Details Area," and "Settlement Information Area." Within each module, it accurately extracts text and key identifiers, avoiding cross-module information confusion. For ambiguous text recognition in blurry images, such as similar characters or blurred handwriting, the model combines medical scene semantic logic and contextual information for correction. For instance, it corrects "lung" to "pneumonia" and "¥3.500" to "¥3500," improving text extraction accuracy. Through cross-verification of text information with structural and key identifier information—such as ensuring that itemized cost amounts match the data in the "Total Costs" column of the invoice and that treatment item names correspond to treatment codes—it filters out erroneous extraction results, ensuring information consistency.

[0064] This technology simultaneously extracts three types of information: text, structure, and key identifiers, addressing the issue of missing core data caused by existing technologies that only extract text, such as the matching basis between the department visited and the insurance liability, and the cost classification structure. Leveraging the adaptive capabilities of a multimodal large-scale model, it maintains high extraction accuracy even for pre-processed blurry, tilted, and differently formatted images, requiring no manual intervention for correction, thus adapting to diverse image quality issues in real-world claims scenarios. Through semantic correction and multimodal cross-validation mechanisms, the information extraction error rate is reduced to below 1%, significantly improving accuracy compared to the traditional OCR + manual data entry model, providing reliable data support for subsequent review stages. It achieves simultaneous extraction of all three types of information in one step, eliminating the need for step-by-step processing, and improving processing efficiency by more than 50% compared to traditional technologies, laying a solid foundation for seamless integration of recording and review.

[0065] S103. The multimodal information is entered into the corresponding fields according to the preset data format requirements to generate a standardized medical information dataset.

[0066] The multi-dimensional, unstructured, and multimodal information extracted from S102 is transformed into structured data that can be directly recognized by the claims system and adapted to the intelligent review logic through standardized mapping and verification. This eliminates the problems of information format differences and data clutter, and provides a unified data foundation for subsequent in-depth correlation analysis with insurance information.

[0067] The preset data format is a standardized data field system pre-set by the system based on insurance claims industry standards and the claims requirements of different insurance products (critical illness insurance, medical insurance, accident insurance, etc.), and is configurable and scalable.

[0068] It covers all the core fields required for claims assessment, and is divided into four main categories:

[0069] Patient basic information fields: Text information extraction results from associated multimodal information such as name, gender, age, and visit ID; Core visit information fields: Disease diagnosis name, department, hospital name, visit time, treatment item name, key contents of medical orders, etc., corresponding to the diagnosis module and treatment module data in the text information and structure information; Expense details fields: Expense item name, corresponding amount, drug name and specifications, medical insurance attribute identifier, expense classification type, etc., matching the expense classification format of the invoice in the structure information with the expense data in the text information; Key identifier supporting fields: Treatment item code, medical insurance settlement identifier, hospital level identifier, invoice QR code parsing results, etc., associated with the key identifier information extraction results.

[0070] The format clearly defines the data type of each field, including text, numeric, date, and boolean types, with length limits and standardized formatting. For example, date fields are uniformly formatted as "YYYY-MM-DD", and amount fields are reserved to two decimal places to ensure data consistency. It supports flexibly adding and modifying field attributes based on the claims requirements of new insurance products through a configurable audit rule interface, adapting to diverse claims scenarios without refactoring the underlying system logic.

[0071] Based on preset field association rules, the system automatically establishes a one-to-one correspondence between multimodal information and target fields: text information is directly mapped, such as the extracted result of "Disease Diagnosis Name" being directly entered into the corresponding text field, and "Itemized Cost Amount" being directly entered into the numerical cost field; structural information is used for auxiliary mapping, such as all cost data under the "Drug Cost" category module in the invoice being automatically categorized and entered into the "Drug Cost" subfield in "Cost Details Information," avoiding cross-category confusion; key identification information is associated and mapped, such as the "Treatment Item Code" and "Treatment Item Name" fields being entered in conjunction, ensuring consistency verification between the code and the name. The entire data entry process is executed automatically by the system, eliminating the need for manual filling or supplementation, completely solving the problems of low efficiency and large errors caused by manual intervention in traditional data entry methods.

[0072] To ensure the accuracy and completeness of the standardized dataset, verification is performed simultaneously during the data entry process:

[0073] Verify whether the entered data meets the format requirements of the preset fields (such as whether the date is valid and whether the amount is a valid value). If there is a format error, it will be automatically corrected based on the multimodal information context (such as correcting "2024.5.20" to "2024-05-20"). Check whether the core required fields (such as disease diagnosis name, consultation time, and total cost amount) are missing. If they are missing, it will automatically backtrack to the multimodal information extraction results in S102 and re-extract the corresponding data to complete the fields. Verify the consistency of data logic between related fields, such as whether "total cost item amount" matches "total cost amount" and whether the level information corresponding to "hospital level identifier" and "hospital name" is consistent. If there is a contradiction, mark it as abnormal and retain the original extraction record for subsequent verification.

[0074] After data entry and verification are completed, the system organizes all field data into a standardized medical information dataset according to a preset hierarchical structure, presenting it in a structured data format (such as JSON and XML) to facilitate rapid retrieval and correlation analysis of subsequent multimodal large models.

[0075] S104. Based on the multimodal large model, the medical information dataset, and the user's insurance information, verify whether the treated disease is within the scope of insurance liability and whether the hospital meets the requirements of the insurance terms to determine the eligibility for compensation; based on the multimodal large model, the medical information dataset, and the insurance information, identify and remove expenses in the medical information dataset that are not within the scope of compensation to determine the compensation amount.

[0076] The core logic relies on the deep correlation analysis capabilities of a multimodal large model, combined with standardized medical information datasets and user insurance information, to simultaneously complete the determination of eligibility for compensation and the calculation of compensation amounts, achieving automation and accuracy in the review process. In specific implementation, once the S103 generates a standardized medical information dataset, the device's built-in linkage module will immediately and automatically trigger the review process without any additional manual operation. This linkage module can retrieve the complete insurance information of the corresponding user in real time, covering key content such as the type of insurance product, the specific scope of insurance liability, and the details of exclusion clauses, providing a comprehensive basis for the subsequent two core determination tasks.

[0077] In determining eligibility for compensation, the multimodal big data model will conduct an in-depth comparison between the disease-related information recorded in the medical information dataset and the specific coverage scope in the insurance information. At the same time, it will conduct exclusion checks in conjunction with the details of the exclusion clauses. It will not only focus on whether the core disease is within the scope of coverage, but also analyze related factors such as complications and disease classifications to ensure the comprehensiveness of the determination. Meanwhile, the model extracts various information about the hospital from the medical information dataset and carefully checks it against the requirements for the hospital's qualifications and level in the insurance terms to confirm whether the hospital meets the agreed standards. Combining the results of these two checks, the model will ultimately determine whether the user is eligible for compensation.

[0078] In terms of calculating the compensation amount, the multimodal big data model is also based on the medical information dataset and insurance information. It analyzes each expense detail in the medical information dataset one by one, and combines the scope of compensation, exclusions and relevant industry standards in the insurance terms to accurately identify and eliminate expenses that are not covered by compensation. These expenses may include medical expenses that are not directly related to the disease treated and expense types that are explicitly excluded by the insurance terms. After eliminating expenses that are not covered by compensation, the model will calculate the final compensation amount based on the provisions of the insurance information, such as the coverage limit, reimbursement ratio and deductible, to ensure that the amount calculation strictly follows the requirements of the insurance terms.

[0079] For example, in one possible implementation, the step of identifying and removing expenses from the standardized medical information dataset that are not covered by reimbursement, based on a multimodal large model, a standardized medical information dataset, and insurance information, to determine the reimbursement amount includes:

[0080] Based on a multimodal large-scale model, detailed expense information is extracted from the standardized medical information dataset, including expense item names, corresponding amounts, treatment types, drug names and specifications, and examination names. The multimodal large-scale model calls upon the exclusion clauses and reimbursement scope rules in the insurance information, while also associating them with standard disease treatment criteria to establish a reimbursement eligibility determination model. Each expense item is analyzed individually using this model: verifying whether drug costs fall within the scope of the medical insurance catalog, and eliminating costs corresponding to out-of-pocket drugs not covered by the medical insurance catalog; verifying the correlation between examination and treatment items and the treated disease, and eliminating examination and treatment costs not directly related to the current disease; identifying and eliminating excessive treatment costs exceeding the necessary scope of standard treatment by referring to standard disease treatment criteria; eliminating expense types explicitly excluded from reimbursement (such as specific cosmetic treatment costs, and costs of self-purchased drugs not prescribed by a doctor) based on the exclusion clauses in the insurance information; and calculating the difference between the total cost in the standardized medical information dataset and the eliminated costs to obtain the final reimbursement amount.

[0081] S105. Based on the claim eligibility and the claim amount, generate a claim settlement report.

[0082] The eligibility and payout results obtained from S104 are integrated into a well-structured, complete, and clearly defined claims assessment report, providing users with an intuitive claims conclusion while also maintaining standardized claims records for the insurance company.

[0083] The report generation process is entirely automated, requiring no manual intervention. The system automatically populates all content based on the judgment data from previous steps, ensuring the accuracy and consistency of the information. The report contains comprehensive and well-organized core information dimensions: First, it clearly presents the final compensation conclusion, i.e., whether the applicant is eligible for compensation and the corresponding specific compensation amount; second, it details the judgment basis, including key verification points related to compensation eligibility (such as the matching of the treated disease with the insurance liability, and the compliance conclusion of the treated hospital), and the calculation logic of the compensation amount (such as the total cost amount, the total amount of excluded costs, and the application instructions for the reimbursement ratio); third, it lists the details of excluded costs one by one, with each excluded cost marked with a specific reason, such as falling within the scope of the insurance exclusion clause, being unrelated to the treated disease, or exceeding the scope of the medical insurance catalog; at the same time, it also presents key basic information, including basic patient information, core medical information, and insurance product information, forming a complete closed loop of claims information.

[0084] To adapt to the needs of different usage scenarios, the report supports the generation of multiple standardized formats and is compatible with various uses such as online viewing, downloading and printing, and system archiving, ensuring convenient access for both users and insurance companies. After the report is generated, it will be pushed to the user terminal and the insurance company's management backend in real time through the device's built-in information synchronization mechanism: the user terminal can directly view the full report and clearly understand the claims results and the ins and outs of various judgments; the insurance company's management backend will simultaneously retain the report for archiving, facilitating subsequent business verification, data statistics, and compliance audits.

[0085] This process, through standardized report generation logic, completely resolves the issues of vague feedback and unclear evidence in traditional claims processing, making the entire claims process transparent and traceable. Users can learn about the claims details without additional communication, significantly reducing communication costs; while insurance companies, through standardized report archiving, meet business management and regulatory compliance requirements, and provide complete data support for possible subsequent reviews or dispute resolution, further enhancing the professionalism and credibility of claims services.

[0086] In one possible implementation, the method further includes a manual review triggering and execution step: based on the standardized medical information dataset, key dimension information such as the number of disease diagnoses, disease types, and complexity of treatment items are extracted, and it is automatically determined whether the case belongs to a special complex case requiring manual review. The special complex case includes cases involving multiple diseases, rare disease treatment, cross-departmental joint treatment, and cases involving multiple overlapping insurance liabilities. If the case is determined to be a special complex case, the device automatically marks the manual review node and synchronizes the detailed data of the case's multimodal information extraction, the standardized medical information dataset, insurance information, and preliminary claim determination logic (including preliminary conclusions on claim eligibility, details of cost exclusions, and determination basis) to the insurance company's management backend. Staff retrieve the complete case data through the management backend, conduct a second verification and confirmation of the preliminary claim determination result, and form a final claim determination conclusion.

[0087] In one possible implementation, the management terminal is equipped with a human-computer interaction interface, which has a full-process information visualization display function. Specifically, it is used to synchronously display the process of extracting multimodal information and the process of determining eligibility and amount of compensation. The display content of the multimodal information extraction process includes the step nodes of information extraction, the original image segments corresponding to each node, the original data of extracted multimodal information and verification and correction records. The display content of the process of determining eligibility and amount of compensation includes the core data comparison link, the original text of the insurance terms on which the determination is based, the industry standard basis, the key logic judgment nodes and the annotation of anomalies.

[0088] The method also includes a manual review and report adjustment process: staff members view the above-mentioned visual information through the human-computer interaction interface and issue manual review confirmation or correction instructions based on the actual situation of the case; if a manual review confirmation instruction is received, the system confirms the validity of the original claims assessment report, archives it as the final claims result, and synchronizes it to the user terminal; if a correction instruction is received, the system adjusts the corresponding content such as the claims eligibility conclusion, the claims amount, the details of excluded expenses, and the judgment basis according to the instruction content, and generates an updated claims assessment report. The updated report needs to be marked with manual correction marks and reasons for correction before archiving and synchronization with the user terminal.

[0089] In summary, the embodiments of this application have the following beneficial effects:

[0090] Leveraging the cross-dimensional analysis capabilities of multimodal large-scale models, this system can directly and simultaneously extract multi-dimensional information such as text, structure, and key identifiers from various pre-processed image data. Standardized data entry is completed through automated format verification and consistency checks, eliminating the need for manual data entry and verification throughout the entire process. Compared to the traditional OCR + manual approach, this not only overcomes the limitations of single-text extraction and poor adaptability to complex images but also improves information processing efficiency by over 50%, strictly controlling the information extraction error rate to below 1%. This effectively reduces rework caused by data errors, laying a solid foundation for accurate processing throughout the entire process.

[0091] Breaking away from the limitations of traditional rule engines that can only match keywords, this system leverages the deep semantic understanding and cross-data association analysis capabilities of multimodal large-scale models to accurately handle claims determinations in complex scenarios such as multiple disease-related visits, complications, and synonymous disease descriptions. Through semantic-level matching of medical information with insurance terms and industry treatment standards, it achieves precise determination of eligibility for claims and the reasonableness of expenses, reducing the rate of manual review intervention by over 60%. This not only improves the consistency and fairness of review results but also significantly reduces the repetitive workload of review personnel.

[0092] By establishing an integrated data entry and review mechanism, the system automatically triggers the review process through a built-in module after the standardized medical information dataset is generated, eliminating the need for manual intervention and completely removing the waiting time and data flow gaps between data entry and review in traditional processes. Combined with the efficient processing capabilities of a multimodal big data model in the data entry and review stages, the overall claims processing cycle is shortened by more than 40% compared to existing technologies, significantly reducing user waiting time and greatly improving the claims service experience and user trust.

[0093] By automating the entire process of information extraction, standardized data entry, and intelligent review through a multimodal big data model, the manual intervention in traditional processes such as data entry and verification, and routine case review is significantly reduced. Insurance companies only need to allocate a small number of staff to focus on the review of a very small number of special and complex cases, significantly reducing labor costs. At the same time, it optimizes human resource allocation, allowing core personnel to be focused on high-value aspects such as handling complex cases and optimizing services.

[0094] The multimodal large model possesses adaptive processing capabilities for image data of different formats and qualities, and is compatible with various image scenarios such as common scanned documents and photographed documents. Simultaneously, the system has a built-in configurable review rule interface, allowing for rapid adaptation to different insurance products such as critical illness insurance, medical insurance, and accident insurance without refactoring the underlying logic; it can be quickly adapted simply by configuring rule parameters. This makes it easily expandable to new insurance types, offering broad applicability with low expansion costs and short cycles.

[0095] The above are some specific implementations of the insurance claim determination method provided in the embodiments of this application. Based on this, this application also provides a corresponding insurance claim determination system. The system provided in the embodiments of this application will be described below from the perspective of functional modularization. Figure 2 This is a structural diagram of an insurance claims determination system provided in an embodiment of this application.

[0096] The system includes:

[0097] The acquisition unit 110 is used to acquire image data uploaded by the user and preprocess the image data.

[0098] Extraction unit 111 is used to extract multimodal information from preprocessed image data based on a multimodal large model;

[0099] The first generation unit 112 is used to input the multimodal information into the corresponding fields according to the preset data format requirements to generate a standardized medical information dataset.

[0100] The determining unit 113 is used to verify, based on the multimodal large model, the medical information dataset, and the user's insurance information, whether the treated disease is within the scope of insurance liability and whether the treated hospital meets the requirements of the insurance terms, in order to determine the eligibility for compensation; and based on the multimodal large model, the medical information dataset, and the insurance information, to identify and remove expenses in the medical information dataset that are not within the scope of compensation, in order to determine the compensation amount.

[0101] The second generation unit 114 is used to generate a claims determination report based on the compensation eligibility and the compensation amount.

[0102] This application also provides corresponding devices and computer storage media for implementing the insurance claim determination scheme provided in this application.

[0103] The device includes a memory and a processor. The memory is used to store instructions or code, and the processor is used to execute the instructions or code so that the device performs the insurance claim determination method described in any embodiment of this application.

[0104] The computer storage medium stores code, and when the code is run, the device running the code implements the insurance claim determination method described in any embodiment of this application.

[0105] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems or apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and relevant parts can be referred to the method section.

[0106] It should be understood that in this application, "at least one" refers to one or more items, and "more" refers to two or more items. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one" or similar expressions refer to any combination of these items, including any combination of singular or plural items. For example, "at least one" of a, b, or c can represent: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, and c can be single or multiple.

[0107] It should be understood that the terms center, longitudinal, transverse, up, down, front, back, left, right, vertical, horizontal, top, bottom, inside, outside, etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the present invention.

[0108] It should be noted that, unless otherwise explicitly specified and limited, the terms installation, connection, and linking should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0109] It should also be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the statement "comprising a..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0110] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0111] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for determining insurance claims, characterized in that, include: Obtain image data uploaded by users and preprocess the image data; Multimodal information is extracted from preprocessed image data based on a multimodal large model; The multimodal information is entered into the corresponding fields according to the preset data format requirements to generate a standardized medical information dataset; Based on a multimodal large model, a medical information dataset, and the user's insurance information, the system verifies whether the treated disease is within the scope of insurance coverage and whether the hospital meets the requirements of the insurance terms in order to determine eligibility for compensation. Based on a multimodal large model, a medical information dataset, and insurance information, expenses that are not covered by the medical information dataset are identified and removed in order to determine the amount of compensation. Based on the eligibility for compensation and the amount of compensation, a claims assessment report is generated.

2. The insurance claim determination method according to claim 1, characterized in that, The multimodal information includes textual information, structural information, and key identification information; the textual information represents the core semantic content identified from the image data for claims determination; the structural information represents the layout and logical relationship of different information units in the image data; and the key identification information represents auxiliary visual elements used to verify the authenticity or authority of the image data.

3. The insurance claim determination method according to claim 1, characterized in that, The process of identifying and removing expenses from the medical information dataset that are not covered by reimbursement, based on a multimodal large model, medical information dataset, and insurance information, to determine the reimbursement amount includes: Based on the multimodal large model, the insurance claim clauses, insurance exclusion clauses and disease treatment information in the insurance information are extracted; based on the multimodal large model, the various cost information and total cost in the medical treatment information dataset are extracted. The cost information in the medical information dataset is compared with the insurance reimbursement terms and disease treatment information. Based on the comparison results, expenses unrelated to the illness treated in this case were identified, and / or expenses excluded by the exclusion clauses of the insurance policy. The compensation amount is calculated by removing the identified expense items from the total cost.

4. The insurance claim determination method according to claim 1, characterized in that, The method further includes: Based on the aforementioned medical information dataset, determine whether the current case is a special and complex case that requires manual review. If a case is determined to be exceptionally complex, mark the current case as a manual review node. The criteria for determining eligibility for compensation, the criteria for determining the amount of compensation, the eligibility for compensation, and the amount of compensation for the current case will be synchronized to the management terminal.

5. The insurance claim determination method according to claim 1, characterized in that, The management terminal includes a human-computer interaction interface, which displays the extraction process of the multimodal information and the determination process of the compensation eligibility and compensation amount. The method further includes: The human-computer interaction interface is used to receive manual review and confirmation instructions or correction instructions. Based on the received instructions, the claims assessment report is adjusted.

6. The insurance claim determination method according to claim 2, characterized in that, The text information includes at least one of the following: disease diagnosis name, department visited, hospital name, itemized cost amount, and consultation time; the structural information includes at least one of the following: invoice layout, medical record paragraph modules, and cost classification table; the key identification information includes at least one of the following: seal, signature, hospital logo, or special symbol in the image data; and the insurance information includes at least one of the following: insurance product type, scope of insurance liability, exclusion clauses, and hospital level restrictions.

7. An insurance claims determination system, characterized in that, include: The acquisition unit is used to acquire image data uploaded by the user and preprocess the image data. The extraction unit is used to extract multimodal information from preprocessed image data based on a multimodal large model; The generation unit is used to input the multimodal information into the corresponding fields according to the preset data format requirements, and generate a standardized medical information dataset. The determination unit is used to verify, based on the multimodal large model, medical information dataset and user insurance information, whether the treated disease is within the scope of insurance liability and whether the hospital meets the requirements of the insurance terms, in order to determine the eligibility for compensation; Based on a multimodal large model, a medical information dataset, and insurance information, expenses that are not covered by the medical information dataset are identified and removed in order to determine the amount of compensation. The generation unit is used to generate a claims determination report based on the claim eligibility and the claim amount.

8. A computer program product, characterized in that, The computer program product stores instructions that, when executed on a terminal device, cause the terminal device to perform the insurance claim determination method as described in any one of claims 1-6.

9. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the insurance claim determination method as described in any one of claims 1-6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed on a terminal device, cause the terminal device to perform the insurance claim determination method as described in any one of claims 1-6.

Images