A report interpretation method and related apparatus

Abstract

The application provides a report interpretation method and related device, and relates to the field of data interpretation. In the application, a plurality of target reports having a correlation relationship are obtained, the report types of the target reports are identified, key information is extracted from the target reports by using the report types of each target report to obtain a key information set, a target search result of at least one target key information in the key information set is searched in a knowledge graph, a professional interpretation result corresponding to the plurality of target reports is determined by using the key information set, a target reasoning path and a node interpretation of each node in the target reasoning path, a content adjustment operation is performed on the professional interpretation result according to a popularization interpretation of each node in the target reasoning path, and a popularization interpretation result corresponding to the professional interpretation result is obtained. In the application, the popularization interpretation result of the target report is obtained by interpreting the report, and then the user can understand the report content.

Description

Technical Field

[0001] The present application relates to the field of data interpretation, and more specifically, to a report interpretation method and related devices. Background Art

[0002] In actual scenarios, users will come into contact with various reports in real life or work scenarios, such as scientific research reports, experimental reports, project reports, pathology reports, etc.

[0003] After receiving the report, the user will read it. During the report reading process, due to reasons such as obscure terminology in the report, the user cannot understand the report content, which reduces the user experience. Summary of the Invention

[0004] In view of this, the present application provides a report interpretation method and related devices to achieve the purpose of obtaining a popular interpretation result of the report through the report interpretation operation, thereby enabling users to understand the content of the report.

[0005] In order to solve the above technical problems, this application adopts the following technical solutions:

[0006] A method for interpreting a report, comprising:

[0007] Get multiple target reports with related relationships;

[0008] Identifying a report type of the target report, and extracting key information from the target report using the report type of each target report to obtain a key information set;

[0009] Retrieving a target retrieval result of at least one target key information in the key information set in the knowledge graph; the target retrieval result includes a target reasoning path of the target key information and a node interpretation of each node in the target reasoning path; the node interpretation includes a professional interpretation and a popular interpretation; the target reasoning path includes a logical relationship between each node in the knowledge graph that has an association relationship with the target key information;

[0010] Determining professional interpretation results corresponding to the multiple target reports using the key information set, the target reasoning path, and the node interpretation of each node in the target reasoning path;

[0011] According to the popular interpretation of each node in the target reasoning path, a content adjustment operation is performed on the professional interpretation result to obtain a popular interpretation result corresponding to the professional interpretation result.

[0012] Optionally, identifying a report type of the target report, and extracting key information from the target report using the report type of each target report to obtain a key information set, including:

[0013] Performing text recognition and structure segmentation operations on the target report to obtain a segmentation result;

[0014] Analyzing the structure and layout of the target report using the segmentation results;

[0015] Determining the report type of the target report according to the structure layout and content of the target report;

[0016] Obtaining key information to be extracted corresponding to the report type of each target report, and extracting the key information from the target report;

[0017] The extracted key information is semantically aligned and conflict fused to obtain a key information set.

[0018] Optionally, retrieving a target retrieval result of at least one target key information in the key information set in the knowledge graph includes:

[0019] Determining at least one target key information from the key information set;

[0020] Obtain knowledge graph;

[0021] Generate an embedding vector for each node in the knowledge graph; the embedding vector includes features of the node and features of neighboring nodes of the node;

[0022] Using the target key information, a multi-hop path search is performed in the embedded vector to obtain a target reasoning path of the target key information;

[0023] Obtaining, from the knowledge graph, a node interpretation of each node in the target reasoning path;

[0024] The target retrieval result of the target key information is obtained by utilizing the target reasoning path of the target key information and the node interpretation of each node in the target reasoning path.

[0025] Optionally, using the target key information, performing a multi-hop path search in the embedded vector to obtain a target reasoning path of the target key information includes:

[0026] Using the target key information, perform a multi-hop path search in the embedded vector to obtain multiple candidate paths;

[0027] Obtaining the importance score and edge weight of each node in the candidate path;

[0028] determine a path score of the candidate path according to the importance scores of each node in the candidate path and the edge weights;

[0029] select a target reasoning path of the target key information from the candidate path according to the path score.

[0030] Optionally, a professional interpretation result corresponding to the multiple target reports is determined by using the set of key information, the target reasoning path, and a node interpretation of each node in the target reasoning path, including:

[0031] performing structural processing on the target reasoning path to obtain a structured path;

[0032] constructing a prompt word according to the set of key information, the structured path, and the node interpretation of each node in the target reasoning path;

[0033] inputting the prompt word into an interpretation model to obtain the professional interpretation result corresponding to the multiple target reports; the content output style of the interpretation model is a preset style, and the output content of the interpretation model adopts a preset structure.

[0034] Optionally, a popular interpretation result corresponding to the professional interpretation result is obtained by performing a content adjustment operation on the professional interpretation result according to a popularized interpretation of each node in the target reasoning path, including:

[0035] performing a term popularization processing operation on the professional interpretation result according to the popularized interpretation of each node in the target reasoning path to obtain a first interpretation result corresponding to the professional interpretation result;

[0036] performing a sentence style adjustment operation on the first interpretation result to obtain a second interpretation result;

[0037] performing a cognitive reconstruction and sentence rearrangement operation on the second interpretation result to obtain a popular interpretation result.

[0038] Optionally, after obtaining the popular interpretation result corresponding to the professional interpretation result, the method further includes:

[0039] outputting the popular interpretation result corresponding to the professional interpretation result in a hierarchical display manner and / or a preset style.

[0040] Optionally, after outputting the popular interpretation result corresponding to the professional interpretation result, the method further includes:

[0041] adjusting the weight of at least one edge in the target reasoning path according to feedback information of the popular interpretation result by a user.

[0042] A report interpretation device, comprising:

[0043] A report acquisition module is used to acquire multiple target reports with associated relationships;

[0044] an information extraction module, configured to identify a report type of the target report, and extract key information from the target report using the report type of each target report to obtain a key information set;

[0045] A retrieval module is configured to retrieve a target retrieval result of at least one target key information in the key information set from the knowledge graph; the target retrieval result includes a target reasoning path of the target key information and a node interpretation of each node in the target reasoning path; the node interpretation includes a professional interpretation and a popular interpretation; the target reasoning path includes a logical relationship between each node in the knowledge graph that has an associated relationship with the target key information;

[0046] an interpretation module, configured to determine professional interpretation results corresponding to the plurality of target reports by using the key information set, the target reasoning path, and the node interpretations of the nodes in the target reasoning path;

[0047] The content adjustment module is used to perform content adjustment operations on the professional interpretation results according to the popular interpretations of each node in the target reasoning path, so as to obtain popular interpretation results corresponding to the professional interpretation results.

[0048] An electronic device comprising at least one processor and a memory connected to the processor, wherein:

[0049] The memory is used to store computer programs;

[0050] The processor is configured to execute the computer program so as to enable the electronic device to implement the above-mentioned report interpretation method.

[0051] The present application provides a report interpretation method and related devices. In the present application, a plurality of target reports with an associated relationship are obtained, the report type of the target report is identified, and the report type of each target report is used to extract key information from the target report to obtain a key information set. The target retrieval result of at least one target key information in the key information set is retrieved in the knowledge graph. The professional interpretation results corresponding to the plurality of target reports are determined by using the key information set, the target reasoning path, and the node interpretation of each node in the target reasoning path. According to the popular interpretation of each node in the target reasoning path, the professional interpretation result is adjusted to obtain a popular interpretation result corresponding to the professional interpretation result. In the present application, the popular interpretation result of the target report is obtained by interpreting the report operation, so that the user can understand the content of the report. In addition, in the present application, when interpreting the professional interpretation result, the target retrieval result of at least one target key information in the key information set retrieved from the knowledge graph will be used, that is, the content related to the target report in the knowledge graph is used to assist in generating the professional interpretation result, thereby ensuring the correctness of the professional interpretation result. In addition, in this application, the professional interpretation results of the target report are first obtained, and then the popular interpretation results are further obtained based on the professional interpretation results. This can ensure that the basic data used for the popularization operation of the interpretation results is correct, thereby ensuring the correctness of the popular interpretation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are merely embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without any creative work.

[0053] Figure 1 A flowchart of a report interpretation method provided in an embodiment of the present application;

[0054] Figure 2 A flowchart of a method for determining a key information set provided in an embodiment of the present application;

[0055] Figure 3 A flowchart of a method for determining target search results provided in an embodiment of the present application;

[0056] Figure 4 A flowchart of a method for determining a professional interpretation result provided in an embodiment of the present application;

[0057] Figure 5 A flowchart of a method for determining a popular interpretation result provided in an embodiment of the present application;

[0058] Figure 6 A schematic diagram of the structure of a report interpretation device provided in an embodiment of the present application;

[0059] Figure 7 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0060] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0061] In actual scenarios, users will come into contact with various reports in real life or work scenarios, such as scientific research reports, experimental reports, project reports, pathology reports, etc.

[0062] Taking a pathology report as an example, after receiving the report, the user will read it. During the report reading process, due to reasons such as obscure terminology in the report, the user may not be able to understand the report content, which reduces the user experience.

[0063] To this end, the present application provides a report interpretation method and related devices. In the present application, a plurality of target reports with an associated relationship are obtained, the report type of the target report is identified, and the report type of each target report is used to extract key information from the target report to obtain a key information set. The target retrieval result of at least one target key information in the key information set is retrieved in the knowledge graph. The professional interpretation results corresponding to the plurality of target reports are determined using the key information set, the target reasoning path, and the node interpretation of each node in the target reasoning path. According to the popular interpretation of each node in the target reasoning path, the professional interpretation result is adjusted to obtain a popular interpretation result corresponding to the professional interpretation result. In the present application, the popular interpretation result of the target report is obtained by interpreting the report operation, so that the user can understand the content of the report. In addition, in the present application, when interpreting the professional interpretation result, the target retrieval result of at least one target key information in the key information set retrieved from the knowledge graph will be used, that is, the content related to the target report in the knowledge graph is used to assist in generating the professional interpretation result, thereby ensuring the correctness of the professional interpretation result. In addition, in this application, the professional interpretation results of the target report are first obtained, and then the popular interpretation results are further obtained based on the professional interpretation results. This can ensure that the basic data used for the popularization operation of the interpretation results is correct, thereby ensuring the correctness of the popular interpretation results.

[0064] In an implementation, with reference to Figure 1 A report interpretation method can include:

[0065] S11, obtaining a plurality of target reports having a correlation relationship.

[0066] In this application, a plurality of target reports having a correlation relationship can be input by a user, for example, a user selects multiple reports on a mini-program and clicks to confirm, and then receives multiple target reports. Taking a pathological report as an example, for the same user, the corresponding pathological report is at least divided into the following types:

[0067] 1. Histopathological report: describes morphology, grading, and diagnosis type;

[0068] 2. Molecular detection report: describes gene mutation, molecular marker detection, etc.

[0069] 3. Integrated diagnosis report: integrates histopathological and molecular detection information;

[0070] 4. Expert consultation report: contains multi-disciplinary consultation opinions.

[0071] In actual scenarios, the user will issue a target report after each operation, and the target report can be multiple reports in the above reports. These reports are reports for the same lesion site of the same user, that is, the multiple target reports have a correlation relationship.

[0072] This application defines different types of target reports, which can be.doc,.txt,.xml, etc. text type, or.pdf (scanned or native),.jpg,.png, etc. picture type, or system interface type, such as structured interface data in HIS (Hospital Information System, Hospital Information System) / EMR (Electronic Medical Record System, Electronic Medical Record System).

[0073] Compared with general physical examination reports, pathological reports are more complex in the following aspects:

[0074] 1. Multiple report types (histopathology, molecular detection, integrated diagnosis, expert consultation, etc.);

[0075] 2. Non-uniform expression (sometimes formal terms, sometimes abbreviations or notes);

[0076] 3. Strong professional, large number of term state combinations;

[0077] 4. A patient may have multiple reports, and the contents need to be used together.

[0078] Therefore, the report analysis method with semantic understanding ability, layout perception ability and strong generalization ability can be used to extract unified and structured semantic information, including diagnosis information, molecular markers and their states, tumor grade, etc. from pathological reports with different formats, complex content and variable structure, thereby providing standardized input for subsequent atlas reasoning and language generation modules.

[0079] S12, a report type of the target report is identified, and key information is extracted from the target report by using the report type of each target report to obtain a key information set.

[0080] In the present application, the report type to which each part of the content in the target report belongs is determined by identifying the visual anchor points in the target report. The report type can be, for example, a histopathological report, a molecular detection report, an integrated diagnosis report, and an expert consultation report.

[0081] For the target report, the extractable key information can be, for example:

[0082] 1) Immunohistochemical results: such as "GFAP (Glial Fibrillary Acidic Protein) (brain tissue +)" "Ki-67 (about 1-2%)" and the like;

[0083] 2) Pathological diagnosis information: such as "astrocytoma" "anaplastic oligodendroglioma" and the like;

[0084] 3) Tumor grade information: such as "CNS WHO 2" "CNS WHO 1-2" and the like;

[0085] 4) Histological features: such as "scattered multinucleated cells" "mitotic figures active" and the like;

[0086] 5) Molecular markers and their states: such as "IDH1 gene R132, no mutation" "ATRX gene, no mutation" "chromosome 1p / 19q, no co-deletion" and the like;

[0087] 6) Clinical recommendations or treatment plans: such as "recommended postoperative head MRI every 6 months" "consider stereotactic radiotherapy gamma knife" and the like;

[0088] 7) Report type identification: such as "histopathological report" "molecular detection report" "integrated diagnosis report" "expert consultation report" and the like.

[0089] The extracted key information can form a key information set.

[0090] S13, a target search result of at least one target key information in the key information set is searched in the knowledge graph.

[0091] The knowledge graph refers to a knowledge graph related to the user's lesion type. For example, if the user's lesion site is the brain, the knowledge graph is a knowledge graph related to head and neck tumors or a knowledge graph related to central nervous system tumors.

[0092] The data source for constructing the knowledge graph can be:

[0093] Authoritative knowledge base: UMLS (Unified Medical Language System), MeSH (National Center for Biotechnology Information Medical Subject Headings), PubTator 3.0 (Public Text Annotator, an artificial intelligence-driven tool for mining biomedical literature resources websites), OncoKB (Precision Oncology Knowledge Base);

[0094] Structured information of clinical guidelines: such as "WHO Classification of Tumours of the Central Nervous System, 5th edition (WHO CNS 2021)" and "Clinical Practice Guidelines: Central Nervous System Tumors 2024.V3";

[0095] Doctor annotation: supplement local diagnosis and treatment path.

[0096] The knowledge graph adopts a heterogeneous graph structure (Heterogeneous Graph). In the knowledge graph, nodes represent medical entities (such as diseases, molecular markers, diagnostic attributes, etc.), and edges represent the relationships between medical entities (such as cause, prediction, suggestion, etc.).

[0097] The type attribute of the node can refer to Table 1:

[0098] Table 1

[0099]

[0100] For the knowledge graph, when configuring the node attribute, each node is attached with:

[0101] Standard name (such as: IDH1 gene R132);

[0102] Popular alternative expression (such as: a gene mutation for judging the malignancy of brain glioma, the mutation has a good prognosis);

[0103] Source citation (such as guide chapter, literature identification ID);

[0104] Entity type label, etc.

[0105] The type attribute of the edge can refer to Table 2 specifically:

[0106] Table 2

[0107]

[0108] After the knowledge graph is constructed, a target retrieval result of at least one target key information in the key information set can be searched in the knowledge graph.

[0109] The target retrieval result includes a target reasoning path of the target key information and node interpretations of each node in the target reasoning path.

[0110] The target reasoning path can also be referred to as a path chain, and the target reasoning path includes a logical relationship of each node in the knowledge graph that has an association relationship with the target key information. For example, taking the target key information "IDH1 mutation" as an example, the target reasoning path can be "IDH1 mutation" → "slow progression" → "good prognosis" → "treatment recommendation". "Slow progression", "good prognosis", and "treatment recommendation" are nodes in the knowledge graph that have an association relationship with "IDH1 mutation", and the logical relationship of the three is "IDH1 mutation" → "slow progression" → "good prognosis" → "treatment recommendation".

[0111] Each node in the target reasoning path has a corresponding node interpretation, and the node interpretation includes a professional interpretation and a popularized interpretation. The professional interpretation refers to an interpretation of the node using professional terms, and the popularized interpretation refers to an interpretation obtained by popularizing the professional interpretation.

[0112] For example, the professional interpretation is "tumor growth is slow", and the popularized interpretation is "your tumor development is relatively mild, and the risk is low".

[0113] S14, using the key information set, the target reasoning path, and the node interpretation of each node in the target reasoning path, determining a plurality of target report corresponding professional interpretation results.

[0114] In this application, the professional interpretation results corresponding to the plurality of target reports refer to structured, editable, and semantically reasonable medical interpretation results. Among them, for the input plurality of target reports, since the plurality of target reports are all the content of the same disease, the final professional interpretation result is only one.

[0115] In the specific professional interpretation result, a model can be used to achieve the output result of the model, which is different from the general summary or question and answer output, and the output result has the following characteristics:

[0116] 1. With fixed structure: including "diagnosis induction + molecular mechanism explanation + suggestion prompt";

[0117] 2. Content closely related to pathological atlas reasoning results;

[0118] 3. Expressed in the style of professional doctors, for machine reconstruction or manual editing.

[0119] S15, according to the popular interpretation of each node in the target reasoning path, the content adjustment operation is performed on the professional interpretation result to obtain the popular interpretation result corresponding to the professional interpretation result.

[0120] In this application, the content adjustment operation on the professional interpretation result means converting the professional interpretation result into a popular interpretation result that can be understood by ordinary patients, emotionally acceptable, and logically easy to follow. The content adjustment operation can include:

[0121] 1. Term popular explanation (medical vocabulary despecialization): when explaining the term, the popular interpretation of each node in the target reasoning path can be used;

[0122] 2. Emotion color adjustment (encouraging tone replacement);

[0123] 3. Logic structure reconstruction (from "doctor to doctor" to "doctor to patient");

[0124] 4. Natural tone, mild semantics, and empathetic expression.

[0125] Through the above content adjustment operation, the popular interpretation result corresponding to the professional interpretation result is finally obtained. The popular interpretation result is easier for users to understand than the professional interpretation result.

[0126] In this embodiment, a plurality of target reports having a correlation relationship are obtained, the report types of the target reports are identified, key information is extracted from the target reports according to the report types of each target report to obtain a key information set, a target search result of at least one target key information in the key information set is searched in a knowledge graph, professional interpretation results corresponding to the plurality of target reports are determined according to the key information set, the target reasoning path and the node interpretation of each node in the target reasoning path, the professional interpretation results are subjected to a content adjustment operation according to the popularization interpretation of each node in the target reasoning path, and a popularization interpretation result corresponding to the professional interpretation result is obtained. In this application, the popularization interpretation result of the target report is obtained by interpreting the report, so that the user can understand the report content. In addition, in this application, when the professional interpretation result is obtained by interpretation, the target search result of at least one target key information in the key information set searched from the knowledge graph is used, that is, the content related to the target report in the knowledge graph is used to assist in generating the professional interpretation result, so as to ensure the correctness of the professional interpretation result. In addition, in this application, the professional interpretation result of the target report is obtained first, and then the popularization interpretation result is further obtained based on the professional interpretation result, which can ensure that the basic data used for the popularization interpretation operation is correct, and the correctness of the popularization interpretation result is ensured.

[0127] On the basis of any of the above embodiments, in another implementation manner, referring to Figure 2 , the report types of the target reports are identified, and the key information is extracted from the target reports according to the report types of each target report to obtain a key information set, which can include:

[0128] S21, text recognition and structure segmentation operation are performed on the target report to obtain a segmentation result.

[0129] In this application, if the target report adopts a PDF (Portable Document Format, Portable Document Format) / picture format, Donut (OCR-free Document Understanding Transformer, OCR-free Document Understanding Transformer model) or PaddleOCR (Paddle Optical Character Recognition, Paddle Optical Character Recognition tool based on Paddle) can be used to recognize paragraphs, tables, headers and footers and the like in the PDF / picture, and the original position mapping is retained for report paragraph division to obtain the segmentation result.

[0130] For text type data or system interface type data, paragraph division can be directly performed to obtain the segmentation result.

[0131] S22. Use the segmentation results to analyze the structural layout of the target report.

[0132] In this application, the segmentation results can be used through visual anchor technology to analyze the structural layout of the target report.

[0133] Visual anchor technology can be implemented using the document intelligent multimodal pre-training model (LayoutLMv3). After fine-tuning LayoutLMv3, it can be used to identify visual anchors in the target report based on the segmentation results of each paragraph and determine the structural layout of the target report.

[0134] S23. Determine the report type of the target report based on the structure layout and content of the target report.

[0135] In this application, the classifier uses the structure and content of the target report to identify the report type of each target report through location awareness + field semantic matching, such as histopathology report, molecular detection report, etc.). Location awareness + field semantic matching refers to:

[0136] The report type is identified by analyzing the structural layout (such as title position, table structure (such as table header) and text content (such as keywords such as "molecular detection report" and "integrated diagnostic report") in the document.

[0137] In practice, position awareness refers to determining the report type by the location of the report name within the target report. For example, "Molecular Detection Report" and "Integrated Diagnostic Report" typically appear in the report header or table header. Field semantic matching identifies keywords within the target report and matches them with keywords within each report type to determine the report type to which the target report belongs.

[0138] It should be noted that the integrated diagnostic report generally includes two parts: histopathology and molecular testing. The subsequent classifier returns the report interval (page number, paragraph position) and report type label at the same time.

[0139] S24. Obtain key information to be extracted corresponding to the report type of each target report, and extract the key information from the target report.

[0140] In this application, the key information to be extracted corresponding to the report type of each target report is pre-configured. For example, if the report type is a tissue pathology report, the corresponding key information to be extracted is pathological diagnosis information, immunohistochemistry results, histological characteristics, tumor grading information, etc.

[0141] Then, according to the key information to be extracted corresponding to the report type of each target report, a key information extraction operation is performed on the target report to obtain the content of each key information.

[0142] When extracting key information, models such as DeepSeek-V3 or GLM-4 (General Language Model) can be used to perform structured, directive extraction of target report paragraphs. These models support fine-tuning with small amounts of data to improve understanding of disease-specific terminology.

[0143] DeepSeek-V3 or GLM-4 extraction granularity supports:

[0144] Terms: tumor name, gene, chromosome;

[0145] Attributes: mutation, methylation, expression status;

[0146] Classification: CNS WHO classification.

[0147] Example of a prompt for DeepSeek-V3 or GLM-4: Extract the pathology diagnosis and molecular test results from the report. The output results are expressed in JSON (JavaScript Object Notation).

[0148] An example output from DeepSeek-V3 or GLM-4 is:

[0149] {

[0150] "diagnosis": "astrocytoma",

[0151] "grade": " CNS WHO Grade 2",

[0152] "marker": {

[0153] "IDH1 gene R132": "mutation",

[0154] "ATRX gene": "mutation",

[0155] "Chromosome 1p / 19q": "Not co-deleted"

[0156] }

[0157] }.

[0158] S25. Perform semantic alignment and conflict fusion operations on the extracted key information to obtain a key information set.

[0159] In practice, the same patient may have multiple different types of pathology reports, with duplicate or conflicting information. It is necessary to merge these reports, unify terminology, eliminate redundancy, and ensure data consistency and integrity.

[0160] When performing fusion, semantic alignment operations need to be performed on the same term. For example, if "astrocytoma" appears in a tissue pathology report and "IDH1 gene mutation" appears in a molecular test report, it is necessary to unify the diagnosis / marker fields through entity name standardization + semantic similarity fusion.

[0161] Entity name standardization: Synonyms or variants that appear in different reports are unified into a standard term, such as "astrocytic tumor" and "astrocytoma" are unified into "astrocytoma".

[0162] Semantic Similarity Calculation: We use the pre-trained BioBERT (Biomedical Language Representation Model) model to calculate the similarity between terms and identify possible synonyms or related terms. For example, "IDH1 gene R132" and "IDH1 gene" have high similarity and can be considered the same entity, unified as "IDH1 gene."

[0163] The above semantic alignment operation is aimed at processing terms that do not have conflicts. When there are conflicts between terms, a conflict resolution strategy is used to perform a conflict fusion operation.

[0164] In the conflict resolution strategy, priority rules for selecting and rejecting conflicting information are determined based on factors such as report type and chronological order. For example, the conclusions of the integrated diagnostic report are prioritized. If an integrated diagnostic report is unavailable, histological features are based on the histopathology report, and molecular information is based on the molecular testing report.

[0165] After semantic alignment and conflict fusion operations, unified structured data is generated to obtain a key information set, which contains all key medical entities and their status for subsequent analysis.

[0166] In one example, the key information set is output using a JSON structure, as shown below:

[0167] {

[0168] "patient_id": "P123456",

[0169] "reports": [

[0170] {

[0171] "type": "Histopathology report",

[0172] "diagnosis": "astrocytoma",

[0173] "histo": "scattered multinucleated cells",

[0174] "grade": "CNS WHO grade 2"

[0175] },

[0176] {

[0177] "type": "molecular testing report",

[0178] "markers": {

[0179] "IDH1 gene R132": "mutated",

[0180] "ATRX gene": "mutated",

[0181] "chromosome lp / 9q": "not co-deleted"

[0182] }

[0183] },

[0184] {

[0185] "type": "integrated diagnostic report",

[0186] "summary": "IDH-mutant diffuse glioma",

[0187] "diagnosis": "astrocytoma",

[0188] "grade": "CNS WHO grade 2",

[0189] "markers": {

[0190] "IDH1 gene R132": "mutated",

[0191] "ATRX gene": "mutated",

[0192] "chromosome lp / 9q": "not co-deleted"

[0193] }

[0194] } ]

[0196] }.

[0197] In the target report obtained in this embodiment, the key information of the target report is extracted, and subsequent analysis operations are only performed on the key information, thereby reducing the data processing amount of subsequent operations and improving efficiency.

[0198] On the basis of any of the above embodiments, in an implementation manner, the target retrieval result of the at least one target key information in the key information set is retrieved in the knowledge graph, which can include the following steps. Figure 3

[0199] S31, determining at least one target key information from the key information set.

[0200] In the key information set, when the key information of the integrated diagnostic report is included, the key information belonging to the integrated diagnostic report in the key information set is integrated and used as the target key information.

[0201] When the key information of the integrated diagnostic report is not included in the key information set, the key information belonging to the histopathological report and the molecular detection report in the key information set can be integrated and used as the target key information.

[0202] It should be noted that other strategies can also be selected to determine at least one target key information from the key information set.

[0203] S32, obtaining a knowledge graph.

[0204] The explanation of the knowledge graph is referred to the corresponding description above.

[0205] S33, generating an embedding vector of each node in the knowledge graph.

[0206] The embedding vector includes the features of the node and the features of the neighbor nodes of the node.

[0207] In order to improve the search efficiency, the embedding vector of each node in the knowledge graph can be generated. In an implementation manner, a graph neural network (GNN) can be used to generate the embedding vector. The GNN is a neural network capable of processing graph structure data, which is suitable for modeling complex relationships between nodes. Specifically, the GNN can be a graph attention network model, that is, the learning of node embedding can be realized through the graph attention network model.

[0208] In the knowledge graph, the node represents a medical entity, and the edge represents the relationship between the medical entities. Through the GNN network, an embedding vector can be generated for each node, which not only contains the features of the node itself, but also fuses the features of the neighbor nodes, so as to capture the context relationship between the medical entities. These embedding vectors can be used for subsequent path search, node classification and other tasks, thereby improving the understanding ability of the system for medical terms.​

[0209] S34, using the target key information, performing multi-hop path search in the embedding vector to obtain a target reasoning path of the target key information.

[0210] Specifically, when performing multi-hop path search, a multi-hop path search + edge weight control search mode can be used. When searching, the explainable path is recursively propagated from the input node corresponding to the target key information.

[0211] The multi-hop path search refers to a path search process in the knowledge graph, starting from a starting node, passing through multiple intermediate nodes, and finally reaching a target node. This search can simulate the diagnosis and reasoning process of a doctor.

[0212] The edge weight control refers to setting the weight of the edge in the graph during the path search process to affect the selection of the path. The edge weight can be set according to the recommended level of the medical guideline, the confidence of the node, and other factors.

[0213] In a specific implementation, the multi-hop path search can use a breadth-first search (BFS) algorithm. The breadth-first search combines an edge weight priority strategy to preferentially explore target reasoning paths with higher weights.

[0214] In one implementation, step S34 includes:

[0215] 1) Using the target key information, performing multi-hop path search in the embedding vector to obtain multiple candidate paths.

[0216] In a specific search, the construction of node embedding is the basis for path search, providing semantic information between nodes, and path search is the exploration of reasoning paths based on this. Therefore, in a specific implementation, the embedding vector of the node needs to be constructed through a graph neural network to obtain the context representation between medical entities. Then, the input node corresponding to the target key information can be searched using these embedding vectors to retrieve semantically similar nodes to find multiple candidate paths.

[0217] It should be noted that when performing multi-hop path search, semantic reasonableness should be ensured: for example, "IDH1 mutation" cannot directly jump to "immunotherapy recommendation", and there should be "better prognosis" and "slow growth" in between.

[0218] 2) Obtain the importance score of each node in the candidate path and the edge weight.

[0219] In the present application, after determining the candidate paths, the optimal target reasoning path needs to be selected from the multiple candidate paths in combination with the edge weight control strategy. When using the edge weight control strategy, the importance score of each node in the candidate path and the edge weight need to be known, wherein the edge weight is determined based on the guideline recommendation level + node confidence, and the importance score of the node can be configured according to actual conditions.

[0220] 3) Determine the path score of the candidate path according to the importance score of each node in the candidate path and the edge weight.

[0221] Wherein, after obtaining at least one candidate path through the above search step, the same term can lead to multiple nodes, and the authoritative path and the clinically meaningful path are preferentially retained. In specific implementation, the path scoring function can be used to preferentially select the path with more medical significance, avoid jumping reasoning, and ensure the rationality and authority of the reasoning path.

[0222] In an implementation manner, after obtaining at least one candidate path, the path scoring function is used to complete the sorting operation of the candidate path. The content of the path scoring function is:

[0223] ;

[0224] Wherein, is the score of the candidate path, is the edge weight in the candidate path, and the edge weight reflects the importance of the edge in medical reasoning, is the importance score of the node in the candidate path, and the importance score considers the medical meaning and confidence of the node.

[0225] The path score of each candidate path can be calculated through the path scoring function.

[0226] 4) Select the target reasoning path of the target key information from the candidate path according to the path score.

[0227] In the present application, the candidate path corresponding to the maximum path score is selected as the target reasoning path of the target key information.

[0228] Compared with the traditional medical information retrieval, the breadth-first search combined with the edge weight priority strategy in the present application has the following differences:

[0229] The traditional medical information retrieval usually stays at the definition level of a single term and lacks in-depth understanding of the causal relationship between terms. By constructing a knowledge graph containing causal relationships, the system can realize multi-hop reasoning along the medical logic path from a term, and gradually reveal the related diagnosis meaning, prognosis information and treatment suggestions. For example, the reasoning path from "IDH1 mutation" to the treatment suggestion.

[0230] Starting term: "IDH1 mutation"

[0231] First jump: "IDH1 mutation" → "slow tumor growth"

[0232] Second jump: "Tumor grows slowly" → "Prognosis is good"

[0233] The third jump: "Good prognosis" → "Chemoradiotherapy and chemotherapy can be delayed."

[0234] S35. Obtain the node interpretation of each node in the target reasoning path from the knowledge graph.

[0235] In this application, the node interpretation of each node is pre-configured in the knowledge graph and can be directly obtained.

[0236] S36. Obtain a target retrieval result of the target key information by using the target reasoning path of the target key information and the node interpretation of each node in the target reasoning path.

[0237] The target search results can be output in JSON format. In one example, the target search results are:

[0238] {

[0239] "term": "IDH1 mutation",

[0240] "diagnostic_path": [

[0241] {

[0242] "node": "Tumor grows slowly",

[0243] "type": "Clinical Inference",

[0244] "plain_text": "Your tumor is relatively mild and low risk."

[0245] },

[0246] {

[0247] "node": "Good prognosis",

[0248] "type": "Diagnostic attribute",

[0249] "plain_text": "Most of these patients respond well to treatment."

[0250] },

[0251] {

[0252] "node": "Can delay chemoradiotherapy",

[0253] "type": "Treatment Recommendation",

[0254] "plain_text": "Sometimes treatment can be postponed and only regular follow-up visits are required."

[0255] } ]

[0257] }.

[0258] In this application, the target retrieval result of the target key information can be obtained by summarizing the target reasoning path of the target key information and the node interpretation of each node in the target reasoning path.

[0259] In this embodiment, vector retrieval is performed in the knowledge graph to solve the problem of missing complex relationships between professional terms.

[0260] In this embodiment, multi-hop semantic path retrieval is used in the knowledge graph of medical causal logic to form a reasoning chain of reasoning path term → diagnostic meaning → prognostic information → recommended treatment. It not only provides factual retrieval content but also simulates the doctor's reasoning process. Compared with traditional retrieval technology, it can achieve explanatory knowledge guidance rather than simple information collage.

[0261] The target search results in this application can be used for subsequent model generation. This approach not only provides term definitions but also reveals their clinical diagnostic significance and impact on treatment decisions. This chain of reasoning can be used as a prompt input to subsequent models to help them generate more clinically valuable interpretations.

[0262] Based on any of the above embodiments, in one implementation, referring to Figure 4 , using the key information set, the target reasoning path, and the node interpretations of each node in the target reasoning path, to determine the professional interpretation results corresponding to multiple target reports, which may include:

[0263] S41. Structural processing is performed on the target reasoning path to obtain a structured path.

[0264] In this application, examples of target reasoning paths are:

[0265] For the medical entity "IDH1 mutation", the corresponding target reasoning path is: "IDH1 mutation" → "slow tumor growth" → "good prognosis" → "chemoradiotherapy can be delayed".

[0266] The extracted target reasoning path is structured to obtain a structured path.

[0267] S42, constructing a prompt word according to the key information set, the node interpretation of each node in the structured path and the target reasoning path.

[0268] In this application, the prompt is used to guide the large language model to generate content. Taking the target reasoning path as an example: IDH1 mutation → slow tumor growth → good prognosis → delayed radiotherapy and chemotherapy, when embedding the structured path into the prompt, text prompt embedding or vector embedding can be used.

[0269] Among them, the text prompt embedding refers to taking the structured path corresponding to the target reasoning path as a text prompt, directly as part of the prompt. In the prompt, the nodes and edges in the target reasoning path can form a structured prompt pair.

[0270] Vector embedding refers to using a graph attention network to encode the nodes and edges of the structured path to generate vector representations, and these vectors are used as part of the prompt to enhance the model's understanding of entity relationships. In addition, the vector representation obtained in this step can also be embedded into the intermediate layer of the model, which can be an attention layer, guiding the model to use relevant medical knowledge for interpretation.

[0271] In this application, the report interpretation operation is performed using an interpretation model, which can be a large language model such as DeepSeek-V3, GLM-4, etc. with task prompt understanding capability. The output of the interpretation model needs to have clear paragraph distribution and semantic labeling.

[0272] The input of the interpretation model includes two parts:

[0273] 1. Key information set, specific examples can be:

[0274] Report core terms (such as "astrocytoma", "IDH gene mutation", "ATRX gene mutation");

[0275] Report type (histopathological report / molecular detection report / integrated diagnosis report / expert consultation report, etc.);

[0276] Tumor grading information (such as "CNS WHO 2");

[0277] Detection results (mutant / no mutation / no deletion, etc.);

[0278] A JSON example of a key information set is:

[0279] {

[0280] "diagnosis": "astrocytoma",

[0281] "grade": "CNS WHO grade 2",

[0282] "markers": {

[0283] "IDH1 gene R132": "mutation",

[0284] "ATRX gene": "mutation",

[0285] "Chromosome 1p / 19q": "Not co-deleted"

[0286] }

[0287] }.

[0288] 2. Structural processing is performed on the target reasoning path to obtain a structured path.

[0289] Among them, the target reasoning path is "IDH1 mutation" → "slow growth" → "good prognosis" → "can delay chemoradiotherapy".

[0290] The structured path corresponding to the target reasoning path is: [

[0292] { "term": "IDH1 gene mutation", "implication": "low-grade glioma, good prognosis"},

[0293] { "term": "Chromosome 1p / 19q non-co-deletion", "implication": "Non-oligodendroglioma, better prognosis"}

[0294] ].

[0295] 3. Node interpretation of each node in the target reasoning path.

[0296] Among them, node interpretation includes professional interpretation and popular interpretation, among which popular interpretation is used to assist the large model in interpretation processing.

[0297] In one example, the prompts for interpreting the model include the following three types of structural tasks:

[0298] TASK_DIAGNOSIS: extract and summarize diagnostic semantic paragraphs;

[0299] TASK_MOLECULAR: Interpret molecular markers and their clinical significance;

[0300] TASK_ADVICE: Provides suggestions and next steps based on the graph reasoning path.

[0301] Task type labels are embedded in each prompt to control the model to focus on different semantic segments:

[0302] For example:

[0303]

Task Type

[0304]

Prompt Content

[0305]

Atlas Prompt

[0306] The prompt also includes:

[0307]

Note Content

[0308] The output of the interpretation model requires medical style semantic interpretation

[0309] In one example, a small number of structured examples (Few-shot Schema) can be used to guide (example prompt) or generate prompts for dynamic task labels. The specific implementation of Few-shot Schema guidance is as follows:

[0310] Introduce similar case examples as imitation templates to control the model's output style, format, and content density.

[0311] Example prompt:

[0312] Example input:

[0313]

Diagnosis

[0314]

Molecule

[0315]

Atlas

[0316] Example output:

[0317] The patient is an oligodendroglioma with a grade of CNS WHO 2, and the molecular detection supports the classic oligodendroglioma feature, with a better prognosis, and it is recommended to follow up and observe.

[0318] Please generate similar formats based on the following input:

[0319] [Diagnosis]: Astrocytoma...

[0320] S43, input the prompt words into the interpretation model to obtain professional interpretation results corresponding to the multiple target reports.

[0321] Among them, the content output style of the interpretation model is a preset style, and the output content of the interpretation model adopts a preset structure.

[0322] In an implementation manner, the content output style of the interpretation model is a medical professional style (preset style).

[0323] The output content adopts a preset structure: three sections (diagnosis → molecular explanation → treatment suggestion). In an example, the output format of the interpretation model is a structured paragraph:

[0324] {

[0325] "diagnosis_part": "The patient is diagnosed with astrocytoma, CNS WHO 2, which belongs to the category of low-grade glioma.",

[0326] "molecular_part": "The detection found that IDH1 gene R132 mutation, indicating that the tumor growth is relatively slow. ATRX gene mutation, chromosome 1p / 19q non-co-deletion indicates that the tumor is more inclined to astrocytoma.",

[0327] "suggestion_part": "It is currently recommended to perform regular review, and evaluate whether to start radiotherapy and chemotherapy regimen in combination with disease progression."

[0328] }。

[0329] In this application, the prompt words are input into the interpretation model, and the interpretation model can generate content containing professional explanations. For example: the patient's IDH1 gene has a mutation, which is usually associated with slow tumor growth and good prognosis. Therefore, it may be considered to delay radiotherapy and chemotherapy, and adopt a watchful waiting strategy.

[0330] The output of the interpretation model is the professional interpretation result corresponding to the multiple target reports, and an example of the professional interpretation result is shown above.

[0331] The professional interpretation result in this application supports the doctor's rapid review and manual supplement to ensure the accuracy of the professional interpretation result.

[0332] In this application, the professional interpretation result corresponding to the multiple target reports is obtained through the interpretation model, which provides basic data for subsequent popularization processing.

[0333] On the basis of any of the above embodiments, refer to Figure 5, according to the popular interpretation of each node in the target reasoning path, the content adjustment operation is performed on the professional interpretation result to obtain the popular interpretation result corresponding to the professional interpretation result, including:

[0334] S51, according to the popular interpretation of each node in the target reasoning path, the professional interpretation result is subjected to a term popularization processing operation to obtain a first interpretation result corresponding to the professional interpretation result.

[0335] Among them, the term popularization processing operation can adopt a de-jargonization way. In specific implementation, a DeepSeek, GLM, etc. instruction fine-tuning model can be used, and the instruction fine-tuning model rewrites the terms in the professional interpretation result by using the popular interpretation of each node in the target reasoning path, and adjusts them to simple language.

[0336] The prompt example of the instruction fine-tuning model is:

[0337] Please rewrite the following medical terms into language that ordinary people can understand, avoiding the use of professional terms while retaining the medical meaning:

[0338] Term: IDH1 gene R132 mutation

[0339] Hint: appears in patients with astrocytoma, its popular interpretation is…

[0340] Output example:

[0341] "There is a genetic change called IDH1, which usually means that the tumor grows slowly and the condition is relatively stable."

[0342] After the instruction fine-tuning model performs a term popularization processing operation on the professional interpretation result according to the popular interpretation of each node in the target reasoning path, a first interpretation result corresponding to the professional interpretation result is obtained.

[0343] S52, the sentence style adjustment operation is performed on the first interpretation result to obtain a second interpretation result.

[0344] In this application, an emotion-aware tone transformation module can be used to perform a sentence style adjustment operation on the first interpretation result. The technical core of the emotion-aware tone transformation module is:

[0345] Based on emotion dimension injection + large model sentence reconstruction + style memory template library to realize tone optimization.

[0346] The control parameter dimensions of the emotion-aware tone transformation module include:

[0347] Emotional polarity (positive / neutral / negative);

[0348] Tone style (doctor tone / empathetic tone / scientific explanation tone);

[0349] Focus information density (diagnosis / comfort / suggestion).

[0350] In this application, the emotional polarity can be selected as positive, the tone style as empathetic, and the focus information density as comfort.

[0351] The model in the emotional style regulation module can use Qwen2.5-instruct or InstructGLM (instructed version of GLM model) for sentence restructuring to improve empathy and positive guidance.

[0352] The prompt example of the model is:

[0353] Original sentence: "The tumor is low-grade and has a good prognosis."

[0354] Converted sentence: "Your tumor belongs to a type that grows slowly, and most patients can control the disease well. This is a positive signal."

[0355] By using the model in the emotional style regulation module described above to adjust the sentence style of the first interpretation result, the second interpretation result can be obtained.

[0356] S53, cognitive restructuring and word order rearrangement operation is performed on the second interpretation result to obtain a popularized interpretation result.

[0357] In this application, the cognitive restructuring and word order rearrangement module (Conceptual Reframing Engine) is used to perform cognitive restructuring and word order rearrangement operation on the second interpretation result.

[0358] The technical core of the cognitive restructuring and word order rearrangement module is:

[0359] The structure of the professional draft is "reconstructed" into an expression order that is more comfortable for patient cognition.

[0360] The cognitive restructuring and word order rearrangement module uses a "phenomenon -> explanation -> suggestion" structure rearrangement method to make the information closer to the patient's cognitive logic;

[0361] Structure prompt example:

[0362]

Input paragraph

[0363] The examination found a relatively mild lesion in the brain, an astrocytoma, accompanied by an active genetic change (IDH1 mutation), suggesting a potentially slower future course.

[0364] The cognitive restructuring and word order rearrangement module is used to perform cognitive restructuring and word order rearrangement on the second interpretation result to obtain a popularized interpretation result. An example of the popularized interpretation result is:

[0365] {

[0366] "patient_friendly_text": "Your test results show a relatively mild brain tumor...",

[0367] "key_terms": [

[0368] {

[0369] "term": "IDH1 mutation",

[0370] "simplified_explanation": "This usually means the tumor grows slowly and has a good prognosis"

[0371] }

[0372] ],

[0373] "suggestion_summary": "Suggest regular MRI checks to observe trends"

[0374] }。

[0375] In this embodiment, the professional interpretation result is converted into a popularized interpretation result that is professional, rigorous, understandable to ordinary patients, emotionally acceptable, and logically easy to follow. The patient is informed of the condition in a positive and encouraging tone, making it easier for users to understand and accept.

[0376] Based on any of the above embodiments, after obtaining the popularized interpretation result corresponding to the professional interpretation result, the popularized interpretation result corresponding to the professional interpretation result can also be output according to a hierarchical display method and / or a preset style.

[0377] In this embodiment, the popularized interpretation result uses a "phenomenon -> explanation -> suggestion" structural rearrangement method, including:

[0378] the main explanation text;

[0379] key terms and their popular explanations;

[0380] suggestion summary paragraph.

[0381] In the output of the popular interpretation result, supplementary explanation information from the knowledge graph can also be output. The supplementary explanation information can be viewed using the term click-to-view function.

[0382] When outputting information, the output information can be displayed in a hierarchical manner, with the logic configured from professional compression to popular expansion. In one example:

[0383] First layer: directly readable main text segment (popular explanation, text after emotional optimization);

[0384] Second layer: clickable term card (hover (display popular explanation when the mouse hovers over) + graph path);

[0385] Third layer: structured suggestion area (such as review frequency, possible treatment path);

[0386] Fourth layer (optional): doctor team notes, historical comparison.

[0387] In another implementation, the style / role of the output information can also be selected.

[0388] User-selectable output styles:

[0389] Doctor tone (formal, precise);

[0390] Reasonable tone (mild, encouraging);

[0391] Family hint style (auxiliary communication text);

[0392] Each style calls different preset prompts or output templates, and uses the corresponding model to adjust the style of the information to be output to obtain the final output information.

[0393] In one implementation, the output information also supports multilingual and cross-language output. When outputting in multiple languages, the output can support automatic switching between Chinese and English.

[0394] When converting languages, a large model can be used to generate the output information after language conversion, maintaining medical accuracy.

[0395] An example of language conversion by a large model is:

[0396] "IDH1 gene mutation" → "a gene change associated with a milder condition."

[0397] In this application, the popular interpretation results are presented to patients in a multimodal, hierarchical structure to improve comprehension efficiency, emotional acceptance, and engagement. In one implementation, after outputting the popular interpretation results corresponding to the professional interpretation results, the weight of at least one edge in the target reasoning path can be adjusted based on the user's feedback on the popular interpretation results.

[0398] In specific implementation, after the information is output, the user can use click & action to provide corresponding feedback. Examples of using click & action to provide corresponding feedback are:

[0399] Click the term / expand the description → Record points of interest;

[0400] Skip term / close module → Mark as “redundant / difficult to understand”;

[0401] Like / Don’t understand button → Feedback model effect;

[0402] Submit follow-up questions / new questions → Collect secondary Q&A requirements (for optimization in step 6).

[0403] In one implementation, the user feedback information obtained through the above-mentioned clicks and behaviors can be stored in a behavior data structure. In one example, the behavior data structure is:

[0404] {

[0405] "user_id": "U001",

[0406] "interaction": {

[0407] "clicked_terms": ["IDH1", "ATRX"],

[0408] "skipped_sections": ["Suggested Sections"],

[0409] "feedback": {

[0410] "understandable": true,

[0411] "emotion_tone": "soothing",

[0412] "followup_question": "Does chemotherapy no longer need to be done?"

[0413] }

[0414] }

[0415] }.

[0416] Subsequent construction of continuous learning mechanism, through the collection of user cognitive feedback, click behavior, language acceptance and follow-up behavior, reverse optimization of popular explanation version selection, emotional tone control weight, graph path priority order, polysemy term generation expression, personalized style adaptation, etc.

[0417] In an implementation manner, the data used for optimization is the feedback information of the user, including:

[0418] The collected user behavior data;

[0419] Interpretation of term usage records in the text;

[0420] Failed to generate examples (such as "user marked as not understood" sentence);

[0421] User follow-up question content.

[0422] In an implementation manner, the polysemy expression evolution learning (popular expression evolution) of the term can be performed based on the feedback information of the user

[0423] In a specific implementation, multiple popular expression versions are maintained for each term, and A / B testing is performed on the multiple version expressions of a certain term to collect click / understanding distribution.

[0424] Example: "IDH1 mutation" →

[0425] "one tumor change";

[0426] "lighter disease signal";

[0427] "slow development, good treatment response marker".

[0428] Subsequently, the user click rate, average reading time, and whether to trigger follow-up in the feedback information can be obtained, and the user click rate, average reading time, and whether to trigger follow-up are used as statistical indicators. Update the term popular explanation sorting strategy in a positive feedback manner.

[0429] In an implementation manner, the feedback information of the user can be used to implement a graph reasoning path optimization scheme. The implementation principle is:

[0430] Record the target reasoning path used when generating the popular interpretation result (such as: IDH1 mutation → slow progress → delayed treatment). If the popular interpretation result corresponding to the path is marked as "difficult to understand" by multiple users, the edge weight of the path is reduced, the path scoring function is updated, and the priority of high understanding degree path is improved.

[0431] In specific implementations, when the knowledge graph is called to generate popular interpretation results, the target inference path referenced by the large language model is recorded. The target inference path is not a static query definition, but a "multi-hop semantic search path" executed based on the knowledge graph structure.

[0432] This target reasoning path is either explicitly written into the prompt or implicitly passed into the model (e.g. as a context embedding or control vector).

[0433] After the task of generating the popularized explanation result is completed, this target reasoning path will be bound and recorded to the generated task ID and stored in the log system for subsequent feedback learning.

[0434] Users can perform actions within the user interface for popular explanations of results, such as marking them as "not understood" or "difficult to understand," skipping sections, and asking follow-up questions. The relationship between these actions and their corresponding target reasoning paths is analyzed. If a path repeatedly causes cognitive difficulties among multiple users, the weights of certain edges in that path (e.g., "slow progression" → "delayed treatment") are downgraded. The edge weights in the path scoring function are then updated, and this update influences the ranking of future path searches. Paths that are more intuitive, commonly used, or have a higher priority in guideline recommendations are prioritized. If a target reasoning path is repeatedly marked as "not understood," "difficult to understand," skipping sections, or asking follow-up questions, the path is marked as "for physician use only" or "avoid presentation to patients."

[0435] In one example, the target inference path is: "TP53 mutation" → "gene repair disorder" → "disease heterogeneity" → "poor prognosis." This path was marked as "difficult to understand" or "unclear terminology" by 78% of users. In this case, the edge "gene repair disorder" → "disease heterogeneity" will be optimized, with its edge weight reduced from 0.9 to 0.2.

[0436] In this application, a self-learning mechanism after user feedback is introduced to optimize the quality of knowledge graph paths and recommend better interpretation results to users.

[0437] In one implementation, user preferences (emotional style, paragraph length, and medical knowledge acceptance) can be determined based on user behavior clustering results to construct a user style preference profile. An example of a user style preference profile is: "tends to be positive, prefers long, popular sentences, and accepts more technical explanations."

[0438] The user style preference profile can be used to recommend personalized templates or configure parameters for the next generated task, providing users with popular interpretation results that better match the profile.

[0439] In one implementation method, this application also supports model re-tuning and reinforcement learning. Model re-tuning and reinforcement learning can include: updating the popular terminology library, fine-tuning the sentiment template vocabulary, retraining the knowledge graph edge weights, using RLHF (Reinforcement Learning from Human Feedback) or SFT (Supervised Fine-Tuning) to continuously enrich the fine-tuning data set of large models, regularly making lightweight updates to models such as DeepSeek and GLM, and strengthening high-scoring outputs to optimize model outputs.

[0440] In this application, a continuous learning mechanism is constructed to collect users' cognitive feedback, click behavior, language acceptance and follow-up behavior, and reversely optimize the parameters in the previous steps so that the final output of the popular interpretation result is more compatible with the user.

[0441] It should be noted that the model used in this application is fine-tuned using lightweight large model fine-tuning and prompt optimization technology, so that the large model can be accurately and targetedly optimized with only a small number of annotation reports. Ultimately, the entire process from extraction to generation takes into account the patient's emotions and cognitive level, truly realizing "making patients understand what they hear and see" and achieving the patient-centered purpose.

[0442] Based on the embodiment of the above-mentioned report interpretation method, another embodiment of the present application provides a report interpretation device, referring to Figure 6 , which may include:

[0443] A report acquisition module 11 is used to acquire multiple target reports with associated relationships;

[0444] An information extraction module 12 is configured to identify a report type of a target report, extract key information from the target report using the report type of each target report, and obtain a key information set;

[0445] The retrieval module 13 is configured to retrieve a target retrieval result of at least one target key information in the key information set from the knowledge graph; the target retrieval result includes a target reasoning path of the target key information and a node interpretation of each node in the target reasoning path; the node interpretation includes a professional interpretation and a popular interpretation; the target reasoning path includes a logical relationship between each node in the knowledge graph that has an association relationship with the target key information;

[0446] An interpretation module 14 is configured to determine professional interpretation results corresponding to multiple target reports using the key information set, the target reasoning path, and the node interpretations of each node in the target reasoning path;

[0447] The content adjustment module 15 is configured to perform a content adjustment operation on the professional interpretation result according to the popularized interpretation of each node in the target reasoning path, to obtain a popularized interpretation result corresponding to the professional interpretation result.

[0448] In an implementation manner, the information extraction module 12 includes:

[0449] The segmentation module is configured to perform a text recognition and structure segmentation operation on the target report, to obtain a segmentation result.

[0450] The layout determination sub-module is configured to analyze the structural layout of the target report by using the segmentation result.

[0451] The type determination sub-module is configured to determine the report type of the target report according to the structural layout and content of the target report.

[0452] The extraction sub-module is configured to obtain the key information to be extracted corresponding to the report type of each target report, and extract the key information from the target report.

[0453] The processing sub-module is configured to perform a semantic alignment and conflict fusion operation on the extracted key information, to obtain a key information set.

[0454] In an implementation manner, the retrieval module 13 includes:

[0455] The information determination sub-module is configured to determine at least one target key information from the key information set.

[0456] The graph obtaining sub-module is configured to obtain a knowledge graph.

[0457] The vector generation sub-module is configured to generate an embedding vector of each node in the knowledge graph; the embedding vector includes the features of the node and the features of the neighbor nodes of the node.

[0458] The search sub-module is configured to perform a multi-hop path search in the embedding vector by using the target key information, to obtain a target reasoning path of the target key information.

[0459] The interpretation obtaining sub-module is configured to obtain a node interpretation of each node in the target reasoning path from the knowledge graph.

[0460] The result determination sub-module is configured to obtain a target retrieval result of the target key information by using the target reasoning path of the target key information and the node interpretation of each node in the target reasoning path.

[0461] In an implementation manner, the search sub-module includes:

[0462] The search unit is configured to perform a multi-hop path search in the embedding vector by using the target key information, to obtain a plurality of candidate paths.

[0463] a data acquisition unit configured to acquire the importance scores of the nodes in the candidate paths and the edge weights;

[0464] a score determination unit configured to determine a path score of the candidate paths according to the importance scores of the nodes in the candidate paths and the edge weights;

[0465] a selection unit configured to select a target reasoning path of the target critical information from the candidate paths according to the path score.

[0466] In an implementation manner, the interpretation module 14 includes:

[0467] a structured processing submodule configured to perform structured processing on the target reasoning path to obtain a structured path;

[0468] a prompt word construction submodule configured to construct a prompt word according to the critical information set, the structured path, and the node interpretation of each node in the target reasoning path;

[0469] the interpretation submodule is configured to input the prompt word into an interpretation model to obtain professional interpretation results corresponding to the multiple target reports; the content output style of the interpretation model is a preset style, and the output content of the interpretation model adopts a preset structure.

[0470] In an implementation manner, the content adjustment module 15 includes:

[0471] a first adjustment submodule configured to perform a terminology popularization processing operation on the professional interpretation results according to the popularization interpretation of each node in the target reasoning path to obtain a first interpretation result corresponding to the professional interpretation results;

[0472] a second adjustment submodule configured to perform a sentence style adjustment operation on the first interpretation result to obtain a second interpretation result;

[0473] a third adjustment submodule configured to perform a cognitive reconstruction and a word order rearrangement operation on the second interpretation result to obtain a popularization interpretation result.

[0474] In an implementation manner, the method further includes:

[0475] a display module configured to output the popularization interpretation result corresponding to the professional interpretation results in a hierarchical display manner and / or a preset style.

[0476] In an implementation manner, the method further includes:

[0477] an optimization module configured to adjust the weight of at least one edge in the target reasoning path according to feedback information of the user on the popularization interpretation result.

[0478] In this embodiment, a plurality of target reports having a correlation relationship are acquired, the report types of the target reports are identified, key information is extracted from the target reports by using the report type of each target report to obtain a key information set, a target search result of at least one target key information in the key information set is searched in a knowledge graph, professional interpretation results corresponding to the plurality of target reports are determined by using the key information set, the target reasoning path and the node interpretation of each node in the target reasoning path, the professional interpretation results are subjected to a content adjustment operation according to the popularization interpretation of each node in the target reasoning path, and a popularization interpretation result corresponding to the professional interpretation result is obtained. In this application, the popularization interpretation result of the target report is obtained through the interpretation report operation, so that the user can understand the report content. In addition, in this application, when the professional interpretation result is obtained, the target search result of at least one target key information in the key information set searched from the knowledge graph is used, that is, the content related to the target report in the knowledge graph is used to assist in generating the professional interpretation result, so as to ensure the correctness of the professional interpretation result. In addition, in this application, the professional interpretation result of the target report is obtained first, and then the popularization interpretation result is further obtained based on the professional interpretation result, so as to ensure that the basic data used in the popularization interpretation operation is correct, and the correctness of the popularization interpretation result is ensured.

[0479] It should be noted that the working processes of various modules, sub-modules and units in the present application are described above with reference to the corresponding descriptions in the embodiments, and will not be described here.

[0480] In the embodiments of the present application, an electronic device is also provided, which includes at least one processor and a memory connected with the processor, wherein:

[0481] The memory is used to store a computer program;

[0482] The processor is used to execute the computer program, so that the electronic device can implement the report interpretation method described above.

[0483] Reference Figure 7 As shown in FIG. 1, a structural schematic diagram of an electronic device suitable for implementing the electronic device in the embodiments of the present application is shown. The electronic device in the embodiments of the present application can include but is not limited to fixed terminals such as mobile phones, notebook computers, PDAs (personal digital assistants), PADs (tablet computers), desktop computers and the like. Figure 7 The electronic device shown is only an example, and should not bring any limitation to the functions and use range of the embodiments of the present application.

[0484] As Figure 7As shown, the electronic device may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 601, which can perform various appropriate actions and processes based on programs stored in a read-only memory (ROM) 602 or programs loaded from a storage device 608 into a random access memory (RAM) 603. When the electronic device is powered on, the RAM 603 also stores various programs and data required for the operation of the electronic device. The processing device 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0485] Typically, the following devices may be connected to the I / O interface 605: an input device 606 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 607 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 608 including, for example, a memory card, a hard disk, etc.; and a communication device 609. The communication device 609 may allow the electronic device to communicate with other devices wirelessly or by wire to exchange data. Figure 7 The electronic device is shown with various devices, but it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed instead.

[0486] An embodiment of the present application also provides a computer program product including computer-readable instructions. When the computer-readable instructions are executed on an electronic device, the electronic device implements any report interpretation method provided in the embodiment of the present application.

[0487] A computer-readable storage medium is also provided in an embodiment of the present application. The storage medium carries one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can implement any report interpretation method provided in the embodiment of the present application.

[0488] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A report interpretation method, characterized in that: include: Get multiple target reports with related relationships; Identifying a report type of the target report, and extracting key information from the target report using the report type of each target report to obtain a key information set; Retrieving a target retrieval result of at least one target key information in the key information set in the knowledge graph; the target retrieval result includes a target reasoning path of the target key information and a node interpretation of each node in the target reasoning path; the node interpretation includes a professional interpretation and a popular interpretation; the target reasoning path includes a logical relationship between each node in the knowledge graph that has an association relationship with the target key information; Determine professional interpretation results corresponding to multiple target reports using the key information set, the target reasoning path, and node interpretations of each node in the target reasoning path; According to the popular interpretation of each node in the target reasoning path, the professional interpretation result is adjusted to obtain a popular interpretation result corresponding to the professional interpretation result; The key information set, the target reasoning path, and the node interpretation of each node in the target reasoning path are used to determine professional interpretation results corresponding to multiple target reports, including: Structuring the target reasoning path to obtain a structured path; Constructing prompt words according to the key information set, the structured path, and the node interpretations of each node in the target reasoning path; The prompt words are input into the interpretation model to obtain professional interpretation results corresponding to the multiple target reports; the content output style of the interpretation model is a preset style, and the output content of the interpretation model adopts a preset structure.

2. The report interpretation method according to claim 1, characterized in that: Identify the report type of the target report, and extract key information from the target report using the report type of each target report to obtain a key information set, including: Performing text recognition and structure segmentation operations on the target report to obtain a segmentation result; Analyzing the structure and layout of the target report using the segmentation results; Determining the report type of the target report according to the structure layout and content of the target report; Obtaining key information to be extracted corresponding to the report type of each target report, and extracting the key information from the target report; The extracted key information is semantically aligned and conflict fused to obtain a key information set.

3. The report interpretation method according to claim 1, characterized in that: Retrieving a target retrieval result of at least one target key information in the key information set in the knowledge graph includes: Determining at least one target key information from the key information set; Get the knowledge graph; Generate an embedding vector for each node in the knowledge graph; the embedding vector includes features of the node and features of neighboring nodes of the node; Using the target key information, a multi-hop path search is performed in the embedded vector to obtain a target reasoning path of the target key information; Obtaining, from the knowledge graph, a node interpretation of each node in the target reasoning path; The target retrieval result of the target key information is obtained by utilizing the target reasoning path of the target key information and the node interpretation of each node in the target reasoning path.

4. The report interpretation method according to claim 3, characterized in that: Using the target key information, a multi-hop path search is performed in the embedded vector to obtain a target reasoning path for the target key information, including: Using the target key information, perform a multi-hop path search in the embedded vector to obtain multiple candidate paths; Obtaining the importance score and edge weight of each node in the candidate path; Determining a path score of the candidate path based on the importance score and edge weight of each node in the candidate path; A target reasoning path of the target key information is selected from the candidate paths according to the path score.

5. The report interpretation method according to claim 1, characterized in that: According to the popular interpretation of each node in the target reasoning path, the professional interpretation result is adjusted to obtain a popular interpretation result corresponding to the professional interpretation result, including: Performing a terminology popularization operation on the professional interpretation result according to the popular interpretation of each node in the target reasoning path to obtain a first interpretation result corresponding to the professional interpretation result; Performing a sentence style adjustment operation on the first interpretation result to obtain a second interpretation result; The second interpretation result is subjected to cognitive reconstruction and word order rearrangement operations to obtain a popular interpretation result.

6. The report interpretation method according to claim 1, characterized in that: After obtaining the popular interpretation results corresponding to the professional interpretation results, it also includes: The popular interpretation results corresponding to the professional interpretation results are output in a layered display manner and / or a preset style.

7. The report interpretation method according to claim 6, characterized in that: After outputting the popular interpretation result corresponding to the professional interpretation result, the method further includes: The weight of at least one edge in the target reasoning path is adjusted according to user feedback on the popular interpretation result.

8. A report interpretation device, characterized in that: include: A report acquisition module is used to acquire multiple target reports with associated relationships; an information extraction module, configured to identify a report type of the target report, and extract key information from the target report using the report type of each target report to obtain a key information set; A retrieval module is configured to retrieve a target retrieval result of at least one target key information in the key information set from the knowledge graph; the target retrieval result includes a target reasoning path of the target key information and a node interpretation of each node in the target reasoning path; the node interpretation includes a professional interpretation and a popular interpretation; the target reasoning path includes a logical relationship between each node in the knowledge graph that has an associated relationship with the target key information; An interpretation module, configured to determine professional interpretation results corresponding to a plurality of target reports using the key information set, the target reasoning path, and node interpretations of each node in the target reasoning path; A content adjustment module, configured to perform content adjustment operations on the professional interpretation results according to the popular interpretations of each node in the target reasoning path, to obtain popular interpretation results corresponding to the professional interpretation results; The interpretation module is used to determine professional interpretation results corresponding to multiple target reports using the key information set, the target reasoning path, and the node interpretations of each node in the target reasoning path, specifically including: Structuring the target reasoning path to obtain a structured path; Constructing prompt words according to the key information set, the structured path, and the node interpretations of each node in the target reasoning path; The prompt words are input into the interpretation model to obtain professional interpretation results corresponding to the multiple target reports; the content output style of the interpretation model is a preset style, and the output content of the interpretation model adopts a preset structure.

9. An electronic device, characterized in that: comprising at least one processor and a memory connected to the processor, wherein: The memory is used to store computer programs; The processor is configured to execute the computer program so as to enable the electronic device to implement the report interpretation method according to any one of claims 1 to 7.

Images