An interactive system for interpreting tumor gene test reports

Abstract

The present invention provides an interactive system for interpreting tumor gene test reports. The interactive system includes a user instruction recognition module, a multimodal file parsing module, a knowledge base module, a language processing module, and a result feedback module. After the user enters the instruction and uploads personal information such as medical information and genetic test reports, the system performs semantic analysis through the instruction recognition module, extracts key information through the file parsing module, and matches relevant tumor gene data in combination with the knowledge base module to extract individual tumor gene information. The language processing module converts the individual tumor gene information into a natural language answer, which is returned to the user by the result feedback module. The various modules of the system work together to support multiple rounds of interaction, obtain clearer instructions or more comprehensive information through guided inquiries and guided information supplements, and dynamically update the answer content to ensure that the output logic is clear, easy to understand, and meets the personalized feedback results of user needs.

Description

Technical Field

[0001] The present invention belongs to the field of medical care informatics, and specifically relates to an interactive system for medical test report interpretation scenarios, and more specifically to an interactive system for tumor gene test report interpretation scenarios. Background Art

[0002] With the rapid development of precision medicine, tumor genetic testing has become a core component of cancer diagnosis and treatment. Currently available testing products primarily use high-throughput sequencing (NGS) technology to analyze markers such as genetic variants (such as EGFR and KRAS mutations), tumor mutation burden (TMB), and microsatellite instability (MSI), providing a basis for targeted therapy, immunotherapy, and prognostic assessment. However, existing interpretation systems for tumor genetic testing reports generally suffer from the following issues.

[0003] 1. Traditional systems rely on manual or semi-automated processes to parse genetic test reports, making it difficult to efficiently process multimodal data (such as reports in PDF and image formats). This results in incomplete or biased extraction of key information. For example, information such as tumor mutation burden (TMB) and gene variant sites in unstructured text requires manual re-entry, which is inefficient and prone to errors.

[0004] 2. Existing systems are mostly one-way question-and-answer systems. Traditional text-based question-and-answer systems allow users to ask questions through text. Based on the user's input, the system generates answers using a general language model or rule engine. However, this process lacks a dynamic guidance mechanism. When user input is ambiguous or incomplete (for example, simply uploading a report without specifying the requirements), the system cannot proactively guide the user to provide necessary information (such as tumor staging and treatment history), resulting in inefficient interaction.

[0005] 3. General-purpose large language models lack specialized expertise in the oncology field, making it difficult to accurately integrate multi-source medical knowledge (such as drug sensitivity evidence and clinical trial data). Consequently, the accuracy and depth of responses are limited, failing to meet specialized needs. For example, while comparative document CN119493847A uses a large language model to parse gene reports, it lacks a customized knowledge base for oncology scenarios, resulting in one-sided responses and a lack of practical application.

[0006] 4. Traditional genetic testing reports rely heavily on professionals to adjust and verify the report interpretation results, which is inefficient, costly, and poorly accessible, and the results are easily influenced by subjectivity. Summary of the Invention

[0007] In order to solve the problems of insufficient information processing capabilities, poor user interaction experience, insufficient professionalism in the field of tumors, and deep dependence on professionals in the existing technology when interpreting tumor gene test reports, the present invention provides an interactive system for tumor gene test report interpretation scenarios.

[0008] The technical solution adopted by the present invention is: an interactive system for interpreting tumor gene test reports, including:

[0009] A user instruction recognition module is used to perform semantic analysis on user instructions, determine user identity and obtain user intent, and generate knowledge base matching logic based on the user intent;

[0010] A multimodal file parsing module is used to parse personal information uploaded by users and obtain parsed content. Based on the parsed content, it uses a pre-configured large language model to extract knowledge base matching information. The personal information includes individual medical information and tumor gene test report files.

[0011] A knowledge base module, configured to extract individual tumor gene information from a knowledge base based on the knowledge base matching logic and knowledge base matching information; the knowledge base includes a clinical case database, a drug database, a gene variation database, a clinical trial database, and a medical literature database;

[0012] A language processing module is used to generate answer data based on the individual tumor gene information, and use a large language model to perform logical reasoning and language polishing on the answer data to generate answer content;

[0013] The result feedback module includes an inquiry unit, an information supplement unit, and a content return unit; the content return unit is used to stream the answer content to the user; when the user instruction recognition module determines that the user instruction is ambiguous, the inquiry unit performs a guided inquiry on the user and updates the user instruction; when the multimodal file parsing module determines that the obtained parsed content is an information-deficient file, the information supplement unit performs a guided information supplement on the user and updates the personal information; after one or more guided inquiries and / or guided information supplements, the language processing module updates the answer content according to the updated user instruction and / or updated personal information, and streams the updated answer content to the user through the content return unit.

[0014] Preferably, the user instruction recognition module generates knowledge base matching logic through the following steps:

[0015] Perform semantic analysis on the acquired user instructions to obtain user intent;

[0016] Classify user intent based on historical data of user commands and medical domain corpus training classification models;

[0017] Based on the classification results, the type of knowledge base to be called and the data priority are determined, and the knowledge base matching logic corresponding to the user's intention is generated.

[0018] Preferably, the multimodal file parsing module includes:

[0019] A format conversion unit, used to convert personal information uploaded by users in image format or PDF format into structured text;

[0020] A key indicator extraction unit is used to extract key indicators from structured text through a large language model and store them in a database as knowledge base matching information; the key indicators include at least one of tumor type, tumor stage, tumor grade, treatment history, genetic information, gene mutation type, tumor mutation load, programmed death ligand expression status, microsatellite instability status, and micromolecule residual status.

[0021] Preferably, the drug database includes at least one of the following: the relationship between genes and drug sensitivity / resistance, the level of evidence for the relationship between genes and drug sensitivity, drug instructions, medical guideline information, drug clinical trial results, and pharmacoeconomic data.

[0022] Preferably, the pharmacoeconomic data includes at least one of FDA approval information, NMPA approval information, and medical insurance coverage status; the medical insurance coverage status includes at least one of approved drug manufacturers, approved reference prices, approved cancer types, and approved applicable targets.

[0023] Preferably, the knowledge base module further comprises: a similar case matching unit, configured to match similar cases from the clinical case library based on the extracted individual tumor gene information, and embed the matching results into the answer data.

[0024] Preferably, the language processing module also includes: a language optimization unit, which is used to adjust the expression of the answer content according to the user identity, where the user identity includes medical professionals or non-medical professionals; when the user is a non-medical professional, the language optimization unit converts the professional language in the answer content into popular language.

[0025] Preferably, the answer includes: gene information, targeted drug usage tips, and similar case information;

[0026] The gene information includes at least one of information on the association between genes and tumors, evidence of the association between genes and tumors, and gene variation analysis information;

[0027] The targeted drug usage prompt includes at least one of the relationship between genes and targeted drug sensitivity / resistance, the evidence level of the relationship between genes and targeted drug sensitivity, targeted drug instructions, targeted drug medical guide information, targeted drug clinical trial results, and targeted drug economic data.

[0028] Preferably, the method for the result feedback module to provide guided information supplement to the user includes:

[0029] The multimodal document parsing module determines whether the knowledge base matching information extracted from the current parsed content meets the preset minimum information set; if not, the result feedback module sends a supplementary information request to the user, and the request includes uploading missing key indicators and / or confirming ambiguous key indicators.

[0030] Preferably, the minimum information set includes at least one key indicator, and the key indicators include tumor type, tumor stage, gene information, gene mutation type, tumor mutation load, programmed death ligand expression status, microsatellite instability status, and micromolecule residual status.

[0031] Preferably, the result feedback module further includes a prediction unit, which proactively predicts and raises suggestive inquiries based on user intention and parsed content, and embeds the suggestive inquiries into the guided inquiry.

[0032] Preferably, the result feedback module also includes an adaptive unit; used to adjust the expression of guided inquiries and guided information supplements according to the user identity, where the user identity includes medical professionals or non-medical professionals; when the user is a non-medical professional, the adaptive unit converts the professional language in the guided inquiries and guided information supplements into popular language.

[0033] Beneficial effects of the present invention:

[0034] 1. The multimodal file parsing module combines OCR technology with a large language model to automatically parse unstructured data in personal information in different formats, such as images and PDFs, accurately extracting key indicators such as tumor type and genetic variation, reducing manual intervention and improving the system's multimodal data processing capabilities.

[0035] 2. The result feedback module achieves dynamic interactive guidance by providing guided inquiries and guided information supplementation to users. When user input is vague or information is missing, the guided answer mechanism is triggered, and the user is actively asked to supplement necessary information such as tumor stage and treatment history, and the answer content is optimized through multiple interactions.

[0036] 3. The knowledge base module integrates multiple oncology knowledge bases (clinical cases, pharmacoeconomics, medical guidelines, etc.) and combines them with the logical reasoning and language optimization capabilities of the large language model deployed by the language processing module to automatically generate interpretation results that are both professional and popular.

[0037] 4. The language optimization unit in the language processing module adaptively adjusts the presentation of answers based on user identity (doctor / patient), achieving differentiated and personalized output to meet the needs of different user groups. For example, the professional language description "EGFR p.L858R mutation is sensitive to osimertinib (Level A evidence)" can be converted into the popular language description "Your genetic test results indicate that you are suitable for osimertinib, which is recommended by authoritative guidelines."

[0038] In summary, the interactive system provided by the present invention automatically helps users efficiently interpret tumor gene detection reports by improving multimodal data parsing capabilities, adding a dynamic interactive guidance mechanism, and integrating multi-source tumor field knowledge bases. It also generates logically clear and easy-to-understand answer content through an answer mechanism that can be interacted with multiple times. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a structural diagram of an interactive system for interpreting tumor gene test reports provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0040] The following describes the embodiments of the present invention through specific embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through different specific embodiments. The details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other unless they conflict.

[0041] This embodiment provides an interactive system for interpreting tumor gene test reports. The interactive system includes a user instruction recognition module, a multimodal file parsing module, a knowledge base module, a language processing module, and a result feedback module. The user enters an instruction and uploads personal information, including individual medical information and a tumor gene test report file. The user instruction recognition module and the multimodal file parsing module perform semantic analysis on the user instruction and parse the personal information, respectively, to obtain corresponding knowledge base matching logic and knowledge base matching information. The knowledge base module extracts the individual tumor gene information from the knowledge base and generates corresponding answer data and answer content in natural language state through the language processing module. Finally, the answer content is streamed back to the user through the result feedback module. In particular, with the combined effect of the user instruction recognition module, the multimodal file parsing module, and the result feedback module, and the support of the knowledge base module and the language processing module, this embodiment also supports multiple interactions between the system and the user during the answer process, providing guided inquiries and guided information supplementation to the user. With the support of the knowledge base module and the language processing module, the answer content is updated based on the supplemented personal information and more specific user instructions to generate logically clear, easy-to-understand, and user-specific answer content, which is then returned to the user.

[0042] In a specific embodiment, the user command recognition module is configured to perform semantic analysis on user commands, determine user identity, and obtain user intent, and generate knowledge base matching logic based on the user intent. After performing semantic analysis on the user command, the user command recognition module determines that the user's oncology knowledge is relatively limited, thereby determining the user's identity as a non-medical professional (e.g., a patient). The user command recognition module can generate the knowledge base matching logic by performing semantic analysis on the obtained user command to obtain user intent; training a classification model based on historical user command data and a medical domain corpus to classify the user intent; and determining the type of knowledge base to be invoked and the data priority based on the classification results, thereby generating knowledge base matching logic corresponding to the user intent. However, if the user command recognition module determines, after performing semantic analysis on the user command, that the user command is semantically ambiguous and it is difficult to clearly determine the user intent, the result feedback module will conduct guided inquiries with the user and update the user command. For example, the user command can be a simple "Help me interpret this NGS test report" or a more specific description, such as "Help me interpret this NGS report and recommend appropriate drugs and corresponding clinical trials." When the user instruction is a vague or incomplete question, the result feedback module is triggered to conduct guided inquiries to the user, thereby obtaining a clearer user intention, which helps to generate the knowledge base matching logic corresponding to the user intention.

[0043] In a specific embodiment, the multimodal file parsing module is used to parse user-uploaded personal information and obtain parsed content. Based on the parsed content, the module uses a pre-configured large language model to extract knowledge base matching information. This personal information includes individual medical information and tumor gene test report files. Individual medical information generally includes basic sample information (e.g., name, gender, age, family history, medical records, sample type, etc.) as well as test information (e.g., product, hospital, institution, project), allowing the extraction of key indicators such as gender, age, genetic factors, tumor type, tumor stage, and tumor grade as knowledge base matching information. Tumor gene test report files generally include corresponding gene information and its variant test results (e.g., nucleotide sequence information, amino acid sequence information, exons / introns, transcripts, variant type, variant frequency, TMB, MSI), allowing the extraction of key indicators such as gene information, gene mutation type, tumor mutation burden, programmed death ligand expression status, microsatellite instability status, and micromolecule residual status as knowledge base matching information. However, if the multimodal document parsing module parses the uploaded personal information and determines that the parsed content is insufficient and does not meet the minimum standards for personal information interpretation, the result feedback module provides the user with guided information supplementation and updates the personal information. More specifically, the multimodal document parsing module determines whether the knowledge base matching information extracted from the current parsed content meets a preset minimum information set. If not, the result feedback module sends a request for supplementary information to the user, which includes uploading missing key indicators and / or confirming ambiguous key indicators. The minimum information set includes at least one key indicator, including tumor type, tumor stage, genetic information, gene mutation type, tumor mutation load, programmed death ligand expression status, microsatellite instability status, and micromolecule residue status. For example, if a user only uploads a tumor genetic test report file without medical records, key indicators such as tumor type, tumor stage, or tumor grade cannot be extracted from the parsed content. This triggers the result feedback module to provide the user with guided information supplementation, thereby obtaining a more comprehensive parsed content and helping the large language model extract more comprehensive knowledge base matching information from the parsed content.

[0044] In a preferred embodiment, the multimodal file parsing module includes a format conversion unit and a key indicator extraction unit. The format conversion unit is used to convert user-uploaded personal information in various formats (e.g., images, PDFs, tables, text, etc.) into structured text, such as by using ORC technology to identify images or PDF files. The key indicator extraction unit is used to extract key indicators from the structured text using a large language model and store them in a database as knowledge base matching information. These key indicators include at least one of tumor type, tumor stage, tumor grade, genetic information, gene mutation type, tumor mutation load, programmed death ligand expression status, microsatellite instability status, and micromolecule residual status.

[0045] In a specific embodiment, the knowledge base module is used to extract individual tumor gene information from a knowledge base based on the knowledge base matching logic and knowledge base matching information. The knowledge base includes a clinical case database, a drug database, a gene mutation database, a clinical trial database, and a medical literature database.

[0046] In a preferred embodiment, the drug database includes at least one of the following: the relationship between genes and drug sensitivity / resistance, the level of evidence for this relationship, drug instructions, medical guideline information, drug clinical trial results, and pharmacoeconomic data. The pharmacoeconomic data includes at least one of FDA approval information, NMPA approval information, and medical insurance coverage status; medical insurance coverage status includes at least one of approved drug manufacturers, approved reference prices, approved cancer types, and approved indicated targets. The inclusion of pharmacoeconomic data facilitates the interactive system to recommend treatment options of varying cost to patients based on drug economics, allowing patients to compare efficacy, safety, and affordability.

[0047] In a preferred embodiment, the knowledge base module also includes a similar case matching unit. This unit is used to match similar cases from the clinical case library based on the extracted individual tumor genetic information and embed the matching results into the response data. For example, this unit can match similar cases from the clinical case library based on factors such as the patient's tumor type, tumor stage, tumor grade, genetic variation information, and treatment history, and present them for user reference.

[0048] In a specific embodiment, the language processing module is used to generate answer data based on the individual tumor gene information, and use a large language model to perform logical reasoning and language polishing on the answer data to generate answer content. In a specific embodiment, the answer content generally includes gene information, targeted drug use prompts, and similar case information. The gene information should include at least one of the association information between genes and tumors, evidence of the association between genes and tumors, and gene mutation analysis information. The targeted drug use prompts include at least one of the relationship between genes and targeted drug sensitivity / resistance, the level of evidence of the relationship between genes and targeted drug sensitivity, targeted drug instructions, targeted drug medical guide information, targeted drug clinical trial results, and targeted drug economic data.

[0049] In a preferred embodiment, the language processing module also includes a language optimization unit. The language optimization unit is configured to adjust the presentation of answers based on the user's identity, including medical professionals and non-medical professionals. If the user is a non-medical professional, the language optimization unit converts the specialized language in the answers into colloquial language. For example, if the user command recognition module performs semantic analysis on the user's command and determines that the user's oncology knowledge is limited, thus determining the user's identity as a non-medical professional (e.g., a patient), the language optimization unit uses a large language model to perform logical reasoning and language adjustments on the answer data, generating answers in colloquial language. For example, the specialized language statement "EGFRp.L858R mutation is sensitive to osimertinib (Level A evidence)" is converted into colloquial language "Your genetic test results indicate that you are suitable for osimertinib, which has been recommended by authoritative guidelines." The unit then provides recommendations tailored to the patient's specific situation, such as recommending treatment options of varying costs based on drug cost-effectiveness, and comparing efficacy, safety, and affordability.

[0050] In a specific embodiment, the result feedback module includes an inquiry unit, an information supplement unit, and a content return unit. The content return unit is used to stream the answer content to the user; when the user instruction recognition module determines that the user instruction is ambiguous, the inquiry unit conducts a guided inquiry to the user and updates the user instruction; when the multimodal file parsing module determines that the parsed content obtained is an information-deficient file, the information supplement unit conducts a guided information supplement to the user and updates the personal information; after one or more guided inquiries and / or guided information supplements, the language processing module updates the answer content according to the updated user instruction and / or updated personal information, and streams the updated answer content to the user through the content return unit.

[0051] In a preferred embodiment, the method for the result feedback module to provide guided information supplementation to the user includes: the multimodal document parsing module determines whether the knowledge base matching information extracted from the current parsed content meets a preset minimum information set; if not, the result feedback module sends a supplementary information request to the user, the request including uploading missing key indicators and / or confirming ambiguous key indicators. The minimum information set includes at least one key indicator, including tumor type, tumor stage, genetic information, gene mutation type, tumor mutation load, programmed death ligand expression status, microsatellite instability status, and micromolecule residual status.

[0052] In a preferred embodiment, the result feedback module also includes a prediction unit and an adaptive unit. The prediction unit is used to proactively predict and propose suggested queries based on user intent and parsed content. Based on statistical data and pre-trained models in the knowledge base, the prediction unit predicts high-frequency user requests and generates suggested queries, proposing potential questions or suggestions, thereby obtaining more specific user instructions and enriching the response content. For example, if a report indicates that a variant is suitable for a certain type of targeted drug, the system automatically asks, "Would you like to learn about clinical trials of that targeted drug?" The adaptive unit is used to determine the user's identity based on the user's instructions and adjust the presentation of guided queries and guided information supplements. User identities can include medical professionals or non-medical professionals. If the user is a non-medical professional, the adaptive unit converts the professional language in the guided queries and guided information supplements into colloquial language to adjust the query's comprehension level. For example, if the user's answer is less professional, the system uses the large model to polish a simpler query, such as "Do you know the stage of the tumor?"

[0053] For example, in one specific embodiment, a user uploaded an incomplete genetic test report and entered the user instruction "How to treat cancer?" The user instruction recognition module performs semantic analysis on the user instruction and determines that the user instruction is ambiguous. The multimodal file parsing module parses the genetic test report and determines that the key indicators of tumor type and gene mutation type are missing. Therefore, the knowledge base matching information extracted from the current parsed content does not meet the preset minimum information set. The parsed content is an information-deficient file, which triggers the query unit and information supplement unit of the result feedback module to conduct guided inquiries and guided information supplements to the user. At the same time, the adaptive unit of the result feedback module determines that the user is a non-medical professional based on the user instruction, so it converts the professional language in the guided inquiries and guided information supplements into popular language.

[0054] Therefore, the system first provides user guidance in plain language: Hello! We've recognized that you've uploaded a genetic test report. To better interpret your tumor genetic test report, I need some information. I haven't found any tumor type information in your report. Please tell me what type of tumor your report is for, for example, lung cancer, breast cancer, or another type.

[0055] User returned: It is a report of lung cancer.

[0056] The system provides a guided supplement: Thank you for your response! You mentioned lung cancer. Does the report include specific genetic mutation information, such as EGFR, KRAS, or other gene mutations? If so, please provide the specific mutation type.

[0057] User response: I took a look and it’s the EGFR L858R mutation.

[0058] The system adds "tumor type - lung cancer" and "gene variation - EGFR L858R mutation" to personal information, which helps to better extract knowledge base matching information.

[0059] In addition, the query unit of the result feedback module uses plain language to conduct guided inquiries to obtain a clearer understanding of user intent. Simultaneously, the prediction unit of the result feedback module proactively predicts user intent and the parsed content, embedding suggested queries into the guided inquiry: "Thank you for your answer! You mentioned how to treat cancer. Do you want me to recommend specific medications?"

[0060] User returns: Yes.

[0061] The system thus captures clear user intent, helping to better generate knowledge base matching logic and answer: Thank you for your answer! I will provide you with a detailed report interpretation and focus on drug-related recommendations. Please wait.

[0062] Based on the updated user instructions and personal information, the corresponding knowledge base matching logic (routine report interpretation, focusing on drug information) and knowledge base matching information (tumor information, gene mutations, drug instructions, pharmacoeconomics, medical guidelines, and clinical cases) are obtained. The knowledge base module extracts individual tumor gene information from the knowledge base based on the knowledge base matching logic and knowledge base matching information (including the relationship between the specific variant gene corresponding to the tumor and cancer, the relationship between genes and sensitivity / resistance, medication instructions, pharmacoeconomics, medication instructions, medical guidelines, and drugs confirmed in clinical case reports) and focuses on making relevant drug recommendations. The specific information extracted is as follows:

[0063] 1. The relationship between genes and cancer: EGFR mutations occur in 49.5% of non-small cell lung cancers. EGFR is an oncogene whose mutations lead to signaling activation, promoting the development of various tumors. Signaling activation caused by EGFR gene mutations occurs in the development of many tumors, including non-small cell lung cancer, head and neck cancer, esophageal cancer, gastric cancer, intestinal cancer, liver cancer, and breast cancer. The EGFRL858R mutation is common and can lead to sustained kinase activation, enhancing the invasive ability of cancer cells. EGFRL858R is a sensitive mutation, sensitizing the cancer to multiple EGFR-TKI drugs.

[0064] 2. Relationship between genes and sensitivity / resistance: The NMPA / FDA or guidelines for this cancer type have a Level A evidence level. In non-small cell lung cancer, EGFR gene-sensitive mutations are sensitive to gefitinib, vometinib, osimertinib, erlotinib, gefitinib, icotinib, befortinib, afatinib, bevacizumab + erlotinib, and erlotinib + ramucirumab. Specific clinical trial results, drug instructions, and corresponding links to the NCCN and CSCO guidelines are also provided. Evidence from small clinical studies for this cancer type is Level C. EAI045 is an investigational, fourth-generation EGFR allosteric inhibitor that targets EGFR resistance mutations. Specific clinical trial results and relevant clinical trial recruitment information are also provided.

[0065] 3. Partial pharmacoeconomics involved: Based on the drugs mentioned above, the corresponding pharmacoeconomic information was extracted, as shown in Table 1.

[0066] Table 1. Some of the pharmacoeconomics involved

[0067] Drug name Product Name Pharmaceutical production companies Dosage Specification price Average monthly usage Gefitinib tablets Ireko Qilu Pharmaceutical (Hainan) Co., Ltd. oral tablets 0.25g*10 tablets 420 yuan The average monthly consumption is 3 boxes, with a total cost of 1,260 yuan Gefitinib tablets Good luck Chia Tai Tianqing Pharmaceutical Group Co., Ltd. oral tablets 0.25g*10 tablets 450 yuan The average monthly consumption is 3 boxes, with a total cost of 1,350 yuan Gefitinib tablets Iressa AstraZeneca Pharmaceuticals Ltd. oral tablets 0.25g*10 tablets 1596 yuan The average monthly usage is 3 boxes, and the total cost is 4,788 yuan. Gefitinib tablets Ai Xingkang Jiangsu Hengrui Medicine Co., Ltd. tablet 0.25g*10 tablets 498 yuan The monthly consumption is 3 boxes, and the total price is 1494 yuan Vometinib Mesylate Tablets Aversa Shanghai Elis Pharmaceuticals Co., Ltd. oral tablets 40mg*28 tablets 2682.4 yuan The average monthly consumption is 2-3 boxes, and the cost is about 5364.8-8047.2 yuan. Osimertinib Mesylate Tablets Teresa AstraZeneca Pharmaceuticals Ltd. oral tablets 80mg\*30t 4966.2 yuan 1 box / month, total cost 4966.2 yuan Erlotinib Hydrochloride Tablets Loret Shanghai Acnuo Pharmaceutical Co., Ltd. oral tablets 0.15g*7 tablets 497 yuan Unknown Erlotinib Hydrochloride Tablets Tarceva Shanghai Roche Pharmaceuticals Co., Ltd. oral tablets 150mg*7 tablets 567 yuan One month's consumption is 4 boxes, and the total purchase price is 2268 yuan

[0068] 4. Similar Cases: In the first case, the patient was a male smoker with advanced non-small cell lung cancer harboring EGFR L858R and E746_A750del mutations. He received targeted therapy with gefitinib and achieved symptomatic remission. The second case presented a more complex case, harboring both HER2 amplification and the EGFR L858R mutation. Initial treatment with afatinib resulted in a partial response that lasted 21 months. The third case, however, had a different mutation—EGFR G719A and T725M. Although rare, treatment with erlotinib resulted in stable disease for 169 days.

[0069] After integrating the individual tumor gene information extracted above, the language processing module uses a large language model to perform logical reasoning on the response data to generate the response content. The language optimization unit, based on the user's instructions, determines that the user is a non-medical professional and converts the professional language in the generated response content into popular language. The following is the response content returned to the user:

[0070] Based on the level of evidence and CSCO guideline recommendations, osimertinib, ametinib, vumetinib, afatinib, dacomitinib, gefitinib, erlotinib, and icotinib are options. Considering different financial circumstances and drug efficacy, you can choose the appropriate EGFR-TKI for treatment based on your individual situation. For patients with better financial resources, third-generation epidermal growth factor receptor tyrosine kinase inhibitors (EGFR-TKIs) such as osimertinib, vumetinib, and ametinib are recommended. These drugs have demonstrated strong efficacy in inhibiting EGFR-sensitizing mutations and the T790M resistance mutation, while also offering a high safety profile. For example, osimertinib, after medical insurance, costs approximately 510 RMB per 80mg tablet. For patients with moderate financial resources, first-generation EGFR-TKIs such as gefitinib, erlotinib, and icotinib can be considered. These drugs are relatively affordable and effectively inhibit EGFR-sensitizing mutations. For example, the price of gefitinib and erlotinib after medical insurance is about RMB 195 per 150mg tablet, while erlotinib has been included in the medical insurance coverage, which helps to reduce the financial burden on patients.

[0071] The embodiments described above are merely descriptions of preferred implementations of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should fall within the scope of protection of the present invention.

Claims

1. An interactive system for interpreting tumor gene detection reports, characterized in that: include: A user instruction recognition module is used to perform semantic analysis on user instructions, determine user identity and obtain user intent, and generate knowledge base matching logic based on the user intent; A multimodal file parsing module is configured to parse personal information uploaded by users and obtain parsed content, and extract knowledge base matching information based on the parsed content using a pre-configured large language model. The personal information includes individual medical information and tumor gene test report files. The multimodal file parsing module includes a format conversion unit and a key indicator extraction unit. The format conversion unit is configured to convert the personal information uploaded by users into structured text. The key indicator extraction unit is configured to extract key indicators from the structured text using a large language model and store them in a database as knowledge base matching information. The key indicators include at least one of tumor type, tumor stage, tumor grade, gene information, gene mutation type, tumor mutation load, programmed death ligand expression status, microsatellite instability status, and micromolecule residual status. A knowledge base module, configured to extract individual tumor gene information from a knowledge base based on the knowledge base matching logic and knowledge base matching information; the knowledge base includes: a clinical case database, a drug database, a gene mutation database, a clinical trial database, and a medical literature database; A language processing module is used to generate answer data based on the individual tumor gene information, and use a large language model to perform logical reasoning and language polishing on the answer data to generate answer content; The result feedback module includes an inquiry unit, an information supplement unit, and a content return unit; the content return unit is configured to return an answer to the user; when the user instruction recognition module determines that the user instruction is ambiguous, the inquiry unit performs a guided inquiry on the user and updates the user instruction; when the multimodal file parsing module determines that the parsed content obtained is an information-deficient file, the information supplement unit performs a guided information supplement on the user and updates the personal information; after one or more guided inquiries and / or guided information supplements, the language processing module updates the answer based on the updated user instruction and / or updated personal information, and streams the updated answer to the user via the content return unit; wherein the method for the information supplement unit to provide guided information supplement to the user includes: the multimodal file parsing module determines whether the knowledge base matching information extracted from the current parsed content meets a preset minimum information set; if not, the information supplement unit sends a supplementary information request to the user, the request including uploading missing key indicators and / or confirming ambiguous key indicators; the minimum information set includes at least one key indicator.

2. The interactive system according to claim 1, characterized in that The user instruction recognition module generates knowledge base matching logic through the following steps: Perform semantic analysis on the acquired user instructions to obtain user intent; A classification model is trained based on historical data of user instructions and a medical corpus to classify user intent. The type of knowledge base to be called and the data priority are determined based on the classification results, and the knowledge base matching logic corresponding to the user intent is generated.

3. The interactive system according to claim 1, characterized in that The drug database includes at least one of the following: the relationship between genes and drug sensitivity / resistance, the level of evidence for the relationship between genes and drug sensitivity, drug instructions, medical guideline information, drug clinical trial results, and pharmacoeconomic data.

4. The interactive system according to claim 3, characterized in that The pharmacoeconomic data includes at least one of FDA approval information, NMPA approval information, and medical insurance coverage status; the medical insurance coverage status includes at least one of approved drug manufacturers, approved reference prices, approved cancer types, and approved applicable targets.

5. The interactive system according to claim 1, characterized in that The knowledge base module also includes: A similar case matching unit is used to match similar cases from the clinical case library based on the extracted individual tumor gene information, and embed the matching results into the answer data.

6. The interactive system according to claim 1, characterized in that The language processing module also includes: The language optimization unit is used to adjust the expression of the answer content according to the user's identity, where the user identity includes medical professionals or non-medical professionals; when the user is a non-medical professional, the language optimization unit converts the professional language in the answer content into popular language.

7. The interactive system according to claim 1 or 5, characterized in that: The answers include: genetic information, targeted drug usage tips, and similar case information; The gene information includes at least one of information on the association between genes and tumors, evidence of the association between genes and tumors, and gene variation analysis information; The targeted drug usage prompt includes at least one of the relationship between genes and targeted drug sensitivity / resistance, the evidence level of the relationship between genes and targeted drug sensitivity, targeted drug instructions, targeted drug medical guide information, targeted drug clinical trial results, and targeted drug economic data.

8. The interactive system according to claim 1, characterized in that The result feedback module further includes a prediction unit and / or an adaptive unit; The prediction unit is used to proactively predict and propose suggested queries based on user intention and parsed content, and embed the suggested queries into the guided query; The adaptive unit is used to adjust the expression of guided inquiries and guided information supplements according to the user identity, where the user identity includes medical professionals or non-medical professionals; when the user is a non-medical professional, the adaptive unit converts the professional language in the guided inquiries and guided information supplements into popular language.

Images