Large model-based medical instrument registration intelligent evaluation system and method

Abstract

The invention relates to the technical field of medical instrument intelligent review, and discloses a medical instrument registration intelligent review system and method based on a large model, and the system comprises a user side and a cloud server side. The user side comprises an enterprise user side and an evaluation mechanism side; a file generation module, a file submission module and a form filling module are arranged in the enterprise user side; an intelligent review module and an intelligent distribution module are arranged in the cloud server side; the intelligent review module comprises a form review unit and a technical review unit; the form review unit is used for intelligently obtaining a form review result; the technical review unit is used for intelligently acquiring a technical review result; and the review institution end is used for rechecking the form review result and the technical review result by reviews and forming a final review result. According to the method, the declaration material can be intelligently generated, the review efficiency, accuracy and scientificity are improved by utilizing the large model, review decision can be assisted, the labor cost is reduced, the regulation adaptability is high, and the intelligent development of medical instrument registration review is promoted.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent review of medical devices, and particularly relates to an intelligent review system and method for medical device registration based on a large model. Background Art

[0002] Medical device registration review, as a necessary procedure before the market launch of medical device products, is the core link to ensure the safety, effectiveness, and quality controllability of products. It is highly professional and technical, with a strict regulatory system, covering multiple aspects such as design, production, clinical trials, and post-market supervision, and requires extremely high knowledge and analysis capabilities of review personnel. Moreover, the review standards vary significantly depending on product types, risk levels, and application scenarios, and the regulations are updated frequently and the technical specifications are complex, which also poses high requirements for the knowledge reserve and adaptability of review personnel.

[0003] In recent years, with the rapid development of artificial intelligence (AI), profound changes have been brought about in various fields of society. Since OpenAI released the conversational general artificial intelligence tool (Chat Generative Pretrained Transformer, ChatGPT), large language models (LLMs) with tens of billions of parameters have quickly become advanced natural language processing tools. In the field of medical device registration review, intelligent systems have also begun to be applied. The pre-training method of LLMs based on massive data enables them to have a wide range of knowledge reserves and in-depth understanding capabilities, especially showing significant advantages when dealing with professional knowledge in fields such as medical regulations and technical standards, and can assist review personnel in quickly retrieving and analyzing a large number of regulatory documents and technical materials, improving the review efficiency. At the same time, intelligent systems can also provide references and suggestions for new review work by learning from past review cases.

[0004] However, there are still some drawbacks and difficulties in the intelligent systems in the field of medical device registration and review. In terms of drawbacks, the accuracy of intelligent systems needs to be improved. Although large language models show powerful capabilities in processing large amounts of text, in a highly specialized field such as medical device registration and review, misunderstandings of regulations and technical standards can lead to serious consequences. Due to the wide variety and complex technologies of medical devices, it is difficult for intelligent systems to fully and accurately understand and apply the special requirements of various products. Moreover, the interpretability of intelligent systems is poor. For the conclusions and suggestions drawn, it is difficult to clearly explain the reasoning process to reviewers, which will affect the reviewers' trust in the results in the review work that requires rigorous demonstration. In terms of difficulties, the frequent updates of regulations result in a lag in the knowledge update of intelligent systems. Medical device regulations and technical specifications are constantly changing, and intelligent systems need to update data and models in a timely manner to meet new requirements, but this is difficult in actual operation. In addition, the risk levels and application scenarios of different medical device products vary greatly. How to develop appropriate intelligent review strategies for different types of products is a major challenge for intelligent systems. Summary of the Invention

[0005] The present invention aims to provide an intelligent review system and method for medical device registration based on a large model, which can intelligently generate application materials, improve the review efficiency, accuracy and scientificity by using the large model, can assist in review decision-making, reduce labor costs, and has strong regulatory adaptability, contributing to the intelligent development of medical device registration and review.

[0006] To achieve the above object, the present invention provides the following basic solutions.

[0007] Solution 1

[0008] An intelligent review system for medical device registration based on a large model, comprising a user side and a cloud server side;

[0009] The user side includes an enterprise user side and a review agency side; the enterprise user side is provided with a document generation module, a document submission module and a form filling module; the document generation module includes an information input unit, an intelligent interaction unit and a material processing unit; the information input unit is used for enterprise users to input basic information; the intelligent interaction unit is provided with a question-and-answer model, which is used to conduct question-and-answer interactions with enterprise users based on the target application documents and generate the first application materials according to the interaction results; the material processing unit is used to review and supplement the first application materials and form the second application materials; the document submission module is used for enterprise users to submit the second application materials to the cloud server side; the form filling module is used to dynamically generate forms based on the sorting of approval importance for users to fill in.

[0010] The cloud server is equipped with an intelligent review module and an intelligent allocation module; the intelligent review module includes a formal review unit and a technical review unit; the formal review unit is used to judge the integrity and correctness of the second application materials and obtain a formal review result; the technical review unit is used to extract key technical information from the second application materials, conduct risk assessment based on the key technical information, and obtain a technical review result; the intelligent allocation module is used to match suitable reviewers for different application materials according to the classification code of medical devices, and optimize the allocation in combination with the current workload of the reviewers;

[0011] The review agency terminal is used for reviewers to review and verify the formal review result and the technical review result, and form a final review result.

[0012] Solution Two

[0013] A method for intelligent review of medical device registration based on a large model, applying a system for intelligent review of medical device registration as described in Solution One for medical device registration review; includes the following steps:

[0014] Step 1, the enterprise user enters basic information through the user terminal to form the second application materials, and uploads them to the cloud server for storage and processing;

[0015] Step 2, the cloud server conducts intelligent review on the second application materials and matches suitable reviewers for different application materials;

[0016] Step 3, the reviewers review and verify the formal review result and the technical review result, and form a final review result;

[0017] Step 4, the final review result is submitted to the cloud server, and the enterprise user can view the final review result through the user terminal.

[0018] The working principle and advantages of the present invention are as follows:

[0019] A system and method for intelligent review of medical device registration based on a large model according to the present invention can intelligently generate application materials, improve the review efficiency, accuracy and scientificity by using the large model, can assist in review decision-making, reduce labor costs, and has strong regulatory adaptability, which helps to promote the intelligent development of medical device registration review. The key points are:

[0020] First, this solution can assist enterprise users in intelligently generating application materials, which helps reduce labor costs. Through the file generation module on the enterprise user side, especially the Q&A model in the intelligent interaction unit, the system conducts Q&A interactions with enterprise users based on the target application documents, guiding enterprise users to provide key information, thereby intelligently generating the first application materials, and then forming the second application materials after review and supplementation by the material processing unit. This process assists enterprise users in efficiently generating relevant materials according to the format and requirements of the application materials, reducing the difficulty and time cost of enterprise preparation of application materials.

[0021] Second, this solution can intelligently assist in the review decision-making, improving the accuracy, scientificity, and efficiency of the review. With the powerful computing and natural language processing capabilities of the large model, the review agency side can quickly analyze medical device registration data and regulatory information. Moreover, the intelligent review module on the cloud server side can simultaneously conduct formal reviews and technical reviews on a large number of application materials. Compared with the traditional manual review method, it significantly shortens the review cycle and improves the overall review efficiency. Additionally, the review based on the intelligent model can more accurately judge whether the application materials meet the regulatory requirements and identify potential risk points. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a schematic diagram of the system layout architecture of Embodiment 1 of an intelligent medical device registration review system and method based on a large model of the present invention;

[0023] Figure 2 It is a schematic diagram of the overall system structure of Embodiment 1 of an intelligent medical device registration review system and method based on a large model of the present invention;

[0024] Figure 3 It is a schematic diagram of the user side structure of Embodiment 1 of an intelligent medical device registration review system and method based on a large model of the present invention;

[0025] Figure 4 It is a schematic diagram of the cloud server side structure of Embodiment 1 of an intelligent medical device registration review system and method based on a large model of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0026] The following is a more detailed description through specific embodiments:

[0027] Embodiment 1

[0028] The embodiment is basically as shown in Figure 1 and Figure 2 shown: An intelligent medical device registration review system based on a large model includes a user side and a cloud server side.

[0029] The user end includes an enterprise user end, an administrator end and an evaluation agency end. The user end supports multi-platform compatibility, including a Web end, a mobile end and a desktop client, and provides a consistent user experience.

[0030] The enterprise user terminal is provided with a language selection module, a login registration module, a file generation module, a file submission module, a form filling module and a review and reading module. Figure 3 shown.

[0031] The language selection module is used for enterprise users to select the system language; the system language includes but is not limited to Chinese and English. The login and registration module is used for enterprise personnel to register and log in to their accounts, and set basic account information. The basic account information includes but is not limited to company name, business license, and product information.

[0032] The file generation module includes an information input unit, an intelligent interaction unit and a material processing unit.

[0033] The information input unit is used for enterprise users to input basic information; the basic information includes but is not limited to user identity information, enterprise qualifications, applied medical device category, etc., and is used as the basic data source for subsequent application materials generation.

[0034] A question-and-answer model is provided in the intelligent interaction unit, and the question-and-answer model is used to conduct question-and-answer interaction with enterprise users based on the target declaration document, and generate the first declaration materials according to the interaction results; during the interaction process, the question-and-answer model can guide enterprise users to provide key information, clarify data inconsistencies or supplement missing content in the form of questions and answers, so as to optimize the generation process of declaration materials.

[0035] When the question-answering model interacts with the enterprise user on the question-answering basis based on the target application document and generates the first application materials according to the interaction result, the following sub-steps are included:

[0036] S1. Obtain target declaration documents, which include declaration information and declaration material requirements.

[0037] For example, if a company is applying for registration of an "ECG monitor", the target application documents include the following:

[0038] Declaration information: company name, product name, model, registration type (such as first registration or change of registration);

[0039] Application materials requirements: including technical documents, product instructions, clinical validation reports, risk assessment reports, declaration of conformity, etc.

[0040] S2, converts the target declaration document into a word vector set, and parses the declaration issues in the target declaration document based on the word vector set.

[0041] For example, does the product meet specific standards (such as electrical safety standards like IEC 60601)?

[0042] Is a clinical validation report provided to prove the effectiveness of the product in a medical environment?

[0043] Has the product undergone a risk assessment, and does the assessment result comply with relevant regulatory requirements?

[0044] Is the technical specification of the product provided, including power, measurement range, accuracy, etc.?

[0045] S3. Match and retrieve in the knowledge base of the document generation module according to the declared questions, and generate multiple answers for each declared question based on the retrieval results. Here, the knowledge base is preset in the document generation module, which pre-stores a large amount of data related to medical device regulations, standards, technical requirements, and actual application cases to help enterprises quickly match compliance requirements and generate reasonable answers when submitting registration applications.

[0046] For example, for "Does it meet the IEC 60601 electrical safety standard?"

[0047] Answer 1: Yes, this electrocardiogram monitor meets the IEC 60601 electrical safety standard and has passed third-party certification;

[0048] Answer 2: No, it has not passed the IEC 60601 certification and is under rectification.

[0049] S4. Intelligently generate the first declaration materials based on the question answers filled in by enterprise users and the requirements of declaration materials.

[0050] In this step, enterprise personnel can further review and improve the first declaration materials generated by the system, including: checking the generated technical documents to ensure the accuracy and integrity of all information. Checking the details of the clinical validation report to ensure it meets the latest clinical research standards. Further supplementing some technical parameters in the risk assessment report to meet the review requirements.

[0051] The material processing unit is used to review and supplement the first declaration materials to ensure that the integrity and standardization of the materials meet the specific requirements of medical device product declaration, and form the second declaration materials.

[0052] Specifically, when supplementing the first declaration materials, the material processing unit includes the following sub-steps:

[0053] (1) Identify missing items from the first application materials - Discover missing content in the application materials through regulatory comparison and NLP technology. For example, if the regulations require that the application materials must include a detailed description of the intended use and scope of application of the medical device, and the relevant content in the application materials is vaguely expressed or missing, the material processing unit will identify it in a timely manner.

[0054] (2) Priority selection - Select suitable corpora for filling in the missing items from multi-source data such as regulations, historical cases, and guidelines.

[0055] Specifically, extract the regulatory provisions directly related to the missing items from the regulations, and at the same time search for historical cases to find the handling methods for the same missing items when applying for similar medical device products, and refer to the authoritative medical device application guidelines. By comprehensively analyzing multi-source information, determine the most suitable corpus for supplementing the missing items. For example, if the missing item is about the safety indicators of a medical device, the corpus will be preferentially selected from the part of the regulations on medical device safety standards, historical cases of supplementing safety indicators of the same type of products, and relevant safety guidelines.

[0056] (3) Further screen and process the selected corpora - Corpus cleaning, similarity calculation, and consistency verification.

[0057] Specifically, first clean the selected corpora to remove the noise data and irrelevant information. Then perform similarity calculation, match the cleaned corpora with the theme and requirements of the missing items, and select the most fitting corpus. Next, perform consistency verification to ensure that the selected corpora are consistent with the overall application materials in terms of terms, formats, logic, etc. For example, for the professional terms of medical devices, ensure that they are consistent with the terms already used in the application materials to avoid the situation of different expressions for the same concept.

[0058] (4) Automatically generate suggestions: Generate supplementary text based on the extracted content for the user to review.

[0059] Based on the extracted and processed corpus content, use natural language generation technology to generate supplementary text. For example, for the missing part of the clinical trial data, generate a supplementary text containing key contents such as the trial purpose, methods, and results according to the selected corpus.

[0060] Then integrate the supplementary text approved by the user into the first application materials to form a complete and standardized second application materials, providing a reliable data basis for the subsequent intelligent review of medical devices.

[0061] The document submission module is used for enterprise users to submit the second application materials (including their relevant documents) to the cloud server; the relevant documents include but are not limited to clinical evaluation materials, quality management systems, product specifications and label drafts, and non-clinical materials.

[0062] The form filling module is used to dynamically generate a form for users to fill in based on the approval importance ranking.

[0063] When the form filling module dynamically generates a form, it includes the following sub-steps:

[0064] S1. Clean and parse the input regulatory texts, standard documents, and materials submitted by users, including but not limited to removing noise data, standardizing formats, and performing structured conversion.

[0065] S2. The form filling module is equipped with a TF-IDF model; the TF-IDF model extracts keywords from the input text.

[0066] The specific formula is as follows:

[0067] TF-IDF(t, d) = TF(t, d) × IDF(t);

[0068]

[0069] where f t,d is the number of times the word t appears in the document d; ∑ k∈d f k,d is the total number of times all words appear in the document d.

[0070] In this step, entity recognition and dependency parsing are also performed. Specifically, a deep learning model is called to identify regulatory clauses, technical terms, and other key entities from the input text, and the syntactic structure of the input text is parsed to extract the core requirements related to regulations, standards, and risk points.

[0071] Based on the above data, subsequent importance ranking can be assisted to facilitate the legal person to fill in the more important information first.

[0072] S3. Assign weights to each keyword based on the word frequency, semantic relevance, and context position of the keyword, and sort them according to the importance score; the importance score is measured based on the weight.

[0073] S4. Use the keywords with importance scores higher than the preset threshold as the basis for generating form questions in the form. Among them, for regulatory compliance questions, select-type form questions are generated; for technical descriptive content, fill-in-the-blank form questions or descriptive form questions are generated.

[0074] An example of a select-type form question is as follows - in the regulatory standard document, the TF-IDF analysis finds that the important score of "IEC 60601" in the electrical medical device document is high, exceeding the set threshold, indicating that this regulation is extremely crucial in the process of applying for this product. Question: Does your device comply with the IEC 0601-1 electrical safety standard?

[0075] An example of a fill-in-the-blank form question is as follows - in the product description document, the TF-IDF model identifies that the score of "Electromagnetic Compatibility (EMC) test" is relatively high, indicating that this test is crucial for the description of the device performance. Question: Please provide the electromagnetic compatibility (EMC) test parameters of your device: (fill-in-the-blank question) - Test frequency range: (fill-in-the-blank) _ MHz - Radiation emission limit: (fill-in-the-blank) dBμV / m.

[0076] The review access module is used for enterprise users to view the review status and final review results of medical device registration in real time. Specifically, the review access module includes a review status browsing unit and a review report access unit.

[0077] The review status browsing unit is used for enterprise users to view the review status of medical device registration in real time.

[0078] The review report access unit is used for enterprise users to view the final review report and results returned by the cloud server.

[0079] The administrator side includes an information management module, a system monitoring module, and a database update module.

[0080] The information management module is used for the administrator to manage the user information of enterprise users, enterprise information, as well as the background information and qualification information of reviewers or review agencies. The user information includes, but is not limited to, registration and login information, basic information, and information related to medical device review. The enterprise information includes, but is not limited to, company name, unified social credit code, and company product information. The background information includes, but is not limited to, agency name, review scope, and reviewer information. The qualification information includes qualification certificate information. In addition, the information management module is also used for the administrator to view the account status, and the account status includes, but is not limited to, activation, suspension, and cancellation.

[0081] The system monitoring module is used to monitor the running status of the system in real time and record the operation logs of the user side. The operation logs include login records, submission records, and system exception situations.

[0082] The database update module is used to collect and dynamically update medical device associated documents and historical review cases (including review results), and store them in the review knowledge base and the review report case library respectively. The medical device associated documents include but are not limited to the latest medical device regulations, policy documents, industry technical standards and guidelines related to medical devices.

[0083] The review agency terminal is used for reviewers to review the formal review results and technical review results and form a final review result.

[0084] The review agency terminal includes a registration and login unit, a manual review unit, a final review unit of the review department, and a review result submission unit. The registration and login unit is used for reviewers (including review experts and review department personnel) to register information and log in to the system. The manual review unit is used for review experts to manually review and supplement the review results generated by the intelligent review module and form a review result to ensure the accuracy and reasonableness of the review report and the supplementary opinions given. The final review unit of the review department is used for review department personnel to conduct a final review of the review results of review experts. The review result submission unit is used to upload the results after the final review of the review to the cloud server terminal.

[0085] The cloud server terminal is equipped with a data storage module, a data processing module, a data transmission module, a database construction module, a model training module, an intelligent review module, a knowledge update module, and an intelligent allocation module, as Figure 4 shown.

[0086] The data storage module is used for the data of users and the cloud server terminal. The data transmission module is used for data interaction with the user terminal.

[0087] The data processing module includes a data parsing unit, a data cleaning unit, a data structured conversion unit, and a data consistency check unit. The data parsing unit is used to automatically parse the application materials (including their related documents) submitted by enterprise users and extract key data content, including text content, table structure, picture information, etc. The data cleaning unit is used to perform deduplication, outlier detection, missing value filling, and format standardization processing on the parsed data. The data structured conversion unit is used to convert the unstructured data in the key data into a structured format recognizable by the system for subsequent module analysis and processing. The data consistency check unit is used to check the integrity and consistency of the data in the application materials (including their related documents) to ensure that the information in different parts matches each other without conflicts.

[0088] In the database construction module, there are a review report case library, a review material library, and a review knowledge library. Specifically, in the review report case library, there are all reviewed medical device review reports uploaded by the review agency side, and the review reports include those that have passed and those that have not passed the review. In the review material library, there are review materials related to various medical device products that need to be reviewed and uploaded by the enterprise user side. In the review knowledge library, there are medical device regulations, policy documents, industry technical standards and guidelines related to medical devices, review guiding principles for various medical devices at home and abroad, national standards, and guideline materials. Specifically, in this embodiment, the guideline materials include review guiding principles, national standards, and guideline materials for various domestic medical devices, such as the GB 9706 series: "General Safety Requirements and Performance of Medical Electrical Equipment", etc.; industry standards (YY), such as the YY / T 0287 series: "Requirements for Quality Management Systems for Medical Devices - Requirements for Regulatory Purposes"; guiding principles, such as "Technical Guidance for the Biological Evaluation and Clinical Evaluation of Medical Devices", etc.

[0089] The model training module includes a model storage unit, a model calling unit, a model adaptation unit, and a domain adaptation unit.

[0090] The model storage unit is used to store multiple pre-trained deep learning models. In this embodiment, the multiple deep learning models specifically include multi-modal models such as a text analysis model, a table data processing model, and an image recognition model. The pre-training includes training using a medical device domain dataset.

[0091] The model calling unit is used to provide API interfaces for different deep learning models to support other modules in the system to call relevant deep learning models for task processing.

[0092] The model adaptation unit is used to match deep learning models for different types of information data (including but not limited to structured text, tables, images) to ensure that the input data is consistent with the model structure. In this embodiment, meta-learning is also used to make the deep learning model adapt to multiple data formats.

[0093] The domain adaptation unit is used for secondary training of deep learning models. The secondary training includes: adapting the deep learning model from the general medical device domain to the medical device registration and review domain - constructing a dedicated medical device review domain corpus (including but not limited to medical device review regulations text, standard documents, review cases), continuously adding the latest regulations, standards, and cases to the training data using incremental learning, using Prompt Engineering technology, enhancing the model's understanding of domain problems through explicit prompts, strengthening the model's understanding of the specific domain, and improving the applicability of the deep learning model to medical device review standards.

[0094] The intelligent review module includes a formal review unit, a technical review unit, and a review report generation unit.

[0095] The formal review unit is used to judge the integrity and correctness of the second application materials and obtain the formal review results; the technical review unit is used to extract key technical information from the second application materials, conduct risk assessment based on the key technical information, and obtain the technical review results.

[0096] Specifically, in this embodiment, when the formal review unit extracts key technical information, it extracts high-weight content based on regulatory standards, product categories, industry guidelines, and historical cases, combined with NLP technology, as the key technical information.

[0097] The risk assessment includes semantic comparison with the second application materials based on the regulatory standard documents to confirm whether the risk items mentioned in the regulatory standard documents are mentioned in the second application materials, and sorting out the mentioned risk items as one of the technical review results.

[0098] The review report generation unit is used to combine the formal review results and the technical review results to convert and generate a review report in a standard format, that is, the initial review report. The content therein includes but is not limited to regulatory compliance, technical analysis, and risk assessment. After the initial review report is generated, it also provides for the review experts to manually review and supplement the initial review report, and submit the final review report to the cloud server to complete the closed-loop review process (that is, the closed-loop review process is completed by the review agency side).

[0099] Specifically, the operation steps of the formal review unit and the technical review unit are as follows:

[0100] S1, conduct intelligent review on the integrity of the second application materials and the accuracy of the filled information. If the review is passed, proceed to S2.

[0101] Automatically detect whether the materials submitted by the user are complete through intelligent algorithms, and focus on checking whether the content such as registration forms, application documents, and test reports is complete and whether the information filling complies with the specifications to ensure that the materials meet the basic review standards. If there are any omissions or errors, the user will be automatically prompted to supplement or modify.

[0102] S2, use the language large model as the basic model (specifically, here a deep learning model can be called from the cloud server as the basic model), disassemble the relevant legal regulations, standards, guidelines, and review guiding principle documents of medical devices submitted into rule entries, and build a rule library. The rule library includes data format requirements, field integrity of registration forms, necessity of key documents, etc., as the prompt of the large model.

[0103] S3, convert the second application materials and the filled form information into structured data.

[0104] Specifically, the technical review unit automatically identifies various format files (such as PDF, Word, Excel) submitted by users, extracts key information and converts it into structured data, facilitating subsequent rule comparison and semantic analysis, and improving data processing efficiency and accuracy.

[0105] S4. Based on the large model, compare the semantic consistency between the content in the application materials and the standard technical requirements.

[0106] S5. Make detailed annotations on the items that do not conform to the rules, output relevant statements and comparison results, and store them.

[0107] S6. Output the review results in JSON format, including the items that do not pass the rules and the corresponding reasons.

[0108] S7. Generate a review report (corresponding to the initial review report) based on the large model and the review results. If the review fails, the regulations with risks should be output in the review report.

[0109] The knowledge update module adopts an incremental learning mechanism. By periodically loading newly added review cases and regulatory update data, it optimizes the semantic understanding ability of the deep learning model and iterates the latest laws and regulations.

[0110] The intelligent allocation module is used to match suitable review personnel for different application materials according to the classification code of medical devices, and optimize the allocation in combination with the current workload of the review personnel. The intelligent allocation module includes an information processing unit, an expert information management unit, and a task allocation decision-making unit.

[0111] The information processing unit is used to pre-receive and parse the registration information of medical devices, including but not limited to classification code, technical characteristics, risk level; the expert information management unit is used to pre-collect and store the basic information of review personnel, including but not limited to professional field, work experience, current load.

[0112] When the intelligent allocation module matches suitable review personnel for different application materials, the task allocation decision-making unit makes a decision for allocation, including the following sub-steps:

[0113] S1. Match the professional fields, and calculate the similarity between the classification code and the keywords in the expert's professional field. The calculation formula is as follows:

[0114]

[0115] Among them, keyword weight (Weight) - assign weights to each keyword according to factors such as regulatory requirements, technical difficulty, and clinical importance (such as the weight of "implantation" is 1.2, and the weight of "risk" is 1.0);

[0116] Similarity Score - Using deep learning technology to capture the deep semantic relationships between different fields, ensure the context compliance of keywords, and calculate the semantic similarity between materials and expert fields based on a similarity measurement algorithm, with a value range of 0 - 1;

[0117] Number of keywords (n) - The total number of keywords extracted from the declared materials.

[0118] After calculating S 领域 further refine the operation process of professional field matching:

[0119] First, preprocess the data, parse the declared materials, and extract core information (technical standards, detection requirements, etc.); process the expert database and establish professional labels (good at regulations, equipment types, clinical applications, etc.).

[0120] Then, perform multi - dimensional matching: (1) Match the regulation adaptability: Check the matching degree between the regulations familiar to the experts (such as FDA, CE, NMPA) and the material requirements; (2) Match the equipment type: Conduct precise screening according to the equipment classification (such as imaging, in vitro diagnosis, implantable, etc.); (3) Adapt to the clinical background: For products that require clinical evaluation, give priority to matching experts with a medical background.

[0121] Finally, set the expert matching threshold. Set an appropriate matching threshold to achieve priority allocation, alternate allocation, and manual intervention review. For example, if S 领域 <0.5, manual intervention review is required; when 0.5 ≤ S 领域 <0.75, alternate allocation is required; when S 领域 ≥0.75, it is priority allocation.

[0122] S2, Workload Optimization, evaluate the workload of the matched experts and optimize the allocation results. The calculation formula is as follows:

[0123]

[0124] Task Complexity Factor (C) - According to the classification of medical devices (such as Class I, Class II, Class III), different weights are assigned to ensure that high - complexity devices occupy more resources;

[0125] Current Task Volume - Obtained by counting all tasks being processed;

[0126] Expert's Maximum Task Capacity - Set the maximum task limit for each expert according to historical processing capabilities.

[0127] After calculating W 专家After that, expert allocation is carried out. First, expert workload classification is performed, and low workload, medium workload, and high workload are allocated by setting thresholds. For example, in the case of low workload (W 专家 <0.5), the expert is in an idle state and new tasks can be preferentially arranged; in the case of medium workload (0.5 ≤ W 专家 <0.8), moderate arrangements need to be made, and the current task completion time of the expert needs to be considered; in the case of high workload (W 专家 ≥0.8), new tasks are not allocated temporarily and are transferred to other experts.

[0128] Then, according to the task priority allocation, the task priority is adjusted according to the urgency of the materials (such as recalling products first). The "minimum task first" allocation strategy is adopted to ensure that experts work evenly. The load balancing weighted strategy has the following calculation formula:

[0129] Final score = αS 领域 +β(1 - W 专家 )

[0130] Set α > β to give priority to the matching of professional fields.

[0131] In this step, through intelligent allocation optimization, it can be ensured that each medical device declaration material is reviewed by the most suitable expert, while reasonably balancing the workload of experts, improving the review efficiency, shortening the approval cycle, and enhancing the quality and compliance of medical device registration.

[0132] A medical device registration intelligent review system based on a large model provided by this embodiment can intelligently generate declaration materials, and uses the large model to improve the review efficiency, accuracy, and scientificity. It can assist in review decision-making, reduce labor costs, and has strong regulatory adaptability, which helps to promote the intelligent development of medical device registration review.

[0133] Specifically, this solution can assist enterprise users in intelligently generating relevant medical device declaration materials according to the format and material requirements of the declaration materials. The review agency can quickly analyze medical device registration data and regulatory information by introducing the powerful computing and natural language processing capabilities of the large model, greatly improving the review efficiency; by using the in-depth learning and precise matching of the large model for documents and regulations, the accuracy and scientificity of the review are significantly improved; the system also has an intelligent auxiliary decision-making function, and provides high-value references for reviewers by identifying key risk points and compliance assessments, reducing subjective errors. At the same time, this system greatly reduces the need for manual intervention, saves labor costs, and its adaptive learning ability based on the large model ensures that the system can maintain high adaptability and scalability as the regulations are updated. In addition, the friendly user interaction design improves the operation experience, comprehensively promotes the development of medical device registration review towards high efficiency, precision, and intelligence, and provides strong support for the innovation and high-quality development of the medical device industry.

[0134] Example 2

[0135] An intelligent review method for medical device registration based on a large model, which applies an intelligent review system for medical device registration as described in Example 1 to conduct medical device registration review; includes the following steps:

[0136] Step 1, the enterprise user enters basic information through the user terminal to form the second application material, and uploads it to the cloud server for storage and processing;

[0137] Step 2, the cloud server conducts intelligent review on the second application material and matches suitable review personnel for different application materials;

[0138] Step 3, the review personnel review the formal review results and technical review results and form the final review result;

[0139] Step 4, the final review result is submitted to the cloud server, and the enterprise user can view the final review result through the user terminal.

[0140] Step 5, convert the final review result into the final review report and file it, and update the review knowledge base at the same time.

[0141] In this embodiment, the cloud server also supports integration with the third-party platform of the medical device regulatory department. The specific steps include:

[0142] S1, determine the integration target and the technical requirements of the third-party platform.

[0143] S2, design API interfaces and data formats compatible with the third-party platform.

[0144] S3, develop a data conversion module and interface adapters that support different protocols.

[0145] S4, deploy a data exchange mechanism and configure security authentication and log monitoring.

[0146] S5, verify functions, performance, and exception handling to ensure the stable operation of the system.

[0147] S6, complete the deployment and continuously monitor and optimize the system functions.

[0148] The intelligent review method for medical device registration based on a large model provided in this embodiment can intelligently generate application materials, improve the efficiency and quality of the review work by means of the large model, has good regulatory adaptability, can be adapted to multiple platforms, and has strong versatility.

[0149] The above are only embodiments of the present invention. Common general knowledge such as specific structures and characteristics known in the art is not described in detail herein. Those of ordinary skill in the art know all the common general knowledge in the technical field to which the invention pertains before the filing date or the priority date, are able to obtain all the prior art in this field, and have the ability to apply the conventional experimental means before this date. Those of ordinary skill in the art can, under the inspiration given in this application, complete and implement this solution in combination with their own abilities. Some typical well-known structures or well-known methods should not become an obstacle for those of ordinary skill in the art to implement this application. It should be noted that for those skilled in the art, without departing from the structure of the present invention, several deformations and improvements can still be made, and these should also be regarded as the protection scope of the present invention, and these will not affect the implementation effect of the present invention and the practicality of the patent.

Claims

1. A medical device registration intelligent review system based on a large model, characterized in that: Including user side and cloud server side; The user end includes an enterprise user end and an evaluation agency end; the enterprise user end is provided with a file generation module, a file submission module and a form filling module; the file generation module includes an information entry unit, an intelligent interaction unit and a material processing unit; the information entry unit is used for enterprise users to enter basic information; The intelligent interaction unit is provided with a question-and-answer model, which is used to conduct question-and-answer interaction with enterprise users based on the target application documents, and generate the first application materials according to the interaction results; the material processing unit is used to review and supplement the first application materials, and form the second application materials; The file submission module is used for enterprise users to submit second application materials to the cloud server; the form filling module is used to dynamically generate a form for users to fill in based on the approval importance ranking; The cloud server is provided with an intelligent review module and an intelligent allocation module; the intelligent review module includes a formal review unit and a technical review unit; the formal review unit is used to judge the completeness and correctness of the second application materials and obtain the formal review results; the technical review unit is used to extract key technical information from the second application materials, and perform risk assessment based on the key technical information, and obtain the technical review results; the intelligent allocation module is used to match suitable reviewers for different application materials according to the classification code of the medical device, and optimize the allocation in combination with the current workload of the reviewers; The review agency end is used for reviewers to review the formal review results and technical review results and form a final review result.

2. According to claim 1, a medical device registration intelligent review system based on a large model is characterized in that: The form filling module includes the following sub-steps when dynamically generating a form: S1, cleaning and parsing the input regulatory texts, standard documents and user-submitted materials, including but not limited to removing noise data, format standardization and structural conversion; S2, the form filling module is provided with a TF-IDF model; the TF-IDF model extracts keywords from the input text; S3, assigns weights to each keyword based on its frequency, semantic relevance, and contextual position, and sorts them by importance score; The importance score is based on a weighted measure; S4, using keywords with keyword importance scores higher than a preset threshold as the basis for generating form questions in the form, wherein for regulatory compliance questions, multiple-choice form questions are generated; for technical descriptive content, fill-in-the-blank form questions or description form questions are generated.

3. According to the large model-based medical device registration intelligent review system of claim 1, it is characterized in that: When the question-answering model interacts with the enterprise user on the question-answering basis based on the target application document and generates the first application materials according to the interaction result, the following sub-steps are included: S1, obtaining a target declaration document, wherein the target declaration document includes declaration information and declaration material requirements; S2, converting the target declaration document into a word vector set, and parsing the declaration problems in the target declaration document according to the word vector set; S3, performing matching searches in the knowledge base of the document generation module according to the declared questions, and generating multiple answers to each declared question according to the search results; S4, intelligently generates the first declaration materials based on the answers to the questions filled in by the enterprise users and the declaration materials requirements.

4. According to claim 1, a medical device registration intelligent review system based on a large model is characterized in that: The intelligent allocation module includes the following sub-steps when matching appropriate reviewers for different application materials: S1, matching professional fields, using classification codes and keywords in expert professional fields to calculate similarity, the calculation formula is as follows: S2, workload optimization, evaluates the workload of the matched experts and optimizes the allocation results. The calculation formula is as follows: Where C is the task complexity factor.

5. According to claim 1, a medical device registration intelligent review system based on a large model is characterized in that: The enterprise user terminal is also provided with a review and inquiry module; the review and inquiry module is used for enterprise users to check the review status and final review results of medical device registration in real time.

6. According to claim 1, a medical device registration intelligent review system based on a large model is characterized in that: The cloud server also has a database construction module; the database construction module includes a review report case library, a review material library and a review knowledge base.

7. According to claim 3, a medical device registration intelligent review system based on a large model is characterized in that: The user end also includes an administrator end; the administrator end includes an information management module, a system monitoring module and a database update module; the information management module is used for the administrator to manage the user information, enterprise information of the enterprise users, as well as the background information and qualification information of the reviewers or review agencies; the system monitoring module is used to monitor the operating status of the system in real time and record the operation log of the user end; the database update module is used to collect and dynamically update medical device related files and historical review cases, and store them in the review knowledge base and the review report case library respectively.

8. According to claim 1, a medical device registration intelligent review system based on a large model is characterized in that: The cloud server is also provided with a model training module; the model training module includes a model storage unit, a model calling unit, a model adaptation unit and a domain adaptation unit; the model storage unit is used to store multiple pre-trained deep learning models; the pre-training includes training using a medical device field data set; the model calling unit is used to provide API interfaces for different deep learning models; the model adaptation unit is used to match deep learning models for different types of information data; and the domain adaptation unit is used to retrain the deep learning model.

9. A medical device registration intelligent review method based on a large model, characterized in that: Applying a medical device registration intelligent review system based on a large model as described in any one of claims 1 to 8 to conduct a medical device registration review; comprising the following steps: Step 1: Enterprise users enter basic information through the user terminal and form the second application materials, which are uploaded to the cloud server for storage and processing; Step 2: The cloud server conducts intelligent review of the second application materials and matches appropriate reviewers for different application materials; Step 3: The reviewer reviews the formal review results and technical review results and forms the final review results; Step 4: The final review result is submitted to the cloud server, and the enterprise user checks the final review result through the user end.

10. The method for intelligent review of medical device registration based on a large model according to claim 9, characterized in that: It also includes step 5, which converts the final review results into a final review report and archives it, while updating the review knowledge base.

Images