Large model and interaction-based medical diagnosis report generation method and related equipment

Abstract

The invention discloses a medical diagnosis report generation method based on a large model and interaction and related equipment, and belongs to the technical field of natural language processing and medical informatization. According to the invention, through multi-modal data acquisition and dynamic inquiry path adjustment, collection of comprehensive clinical information is ensured. And a symptom-examination result consistency verification and diagnosis basis sufficiency evaluation mechanism is introduced, so that the logic preciseness of medical judgment is ensured. By constructing the medical knowledge graph, related diagnosis and treatment guidelines are automatically provided, and omissions and wrong references caused by manual consultation are avoided. A first draft of differential diagnosis and treatment suggestions is automatically generated by means of a large language model, and the professionalism and safety of report content are improved in combination with examination and verification of a clinical rule engine. A doctor participates in modification through multiple rounds of interaction in the whole process, the main right of professional judgment is reserved, intelligent support provided by AI is fully utilized, and therefore the finally generated diagnosis report conforms to the actual clinical situation and meets medical specifications and specialized requirements.

Description

Technical Field

[0001] This invention relates to the fields of natural language processing and medical information technology, and in particular to a method and related equipment for generating medical diagnostic reports based on large models and interaction. Background Technology

[0002] Currently, the writing of medical diagnostic reports (such as imaging reports, laboratory test reports, and pathology reports) mainly relies on doctors manually completing them or using simple electronic medical record templates. This traditional method has the following significant problems: First, manual writing is time-consuming and laborious, and in busy clinical work, it is easy to miss key clinical information (such as important signs, abnormal test results, etc.) or inconsistencies may occur. Second, existing electronic medical record templates are often fixed and rigid, unable to be dynamically adjusted according to the characteristics of different specialties and individual patient differences. Doctors lack intelligent guidance when filling them out, which may lead to incomplete report content or medical logic errors. Third, in terms of citing diagnostic evidence, doctors need to manually consult clinical guidelines and medical literature, which is inefficient and easily misses the latest evidence-based medicine. With the surge in medical data and the increasing requirements for standardized diagnosis and treatment, these problems are becoming increasingly prominent.

[0003] While some report generation tools have emerged in the field of medical informatics, most only offer basic data entry or simple text replacement, with limited intelligence. For example, some systems can only collect information according to a fixed process and cannot dynamically adjust the consultation path based on the patient's specific condition. Furthermore, current technologies lack intelligent verification mechanisms for medical logical consistency, making it difficult to promptly detect contradictions between clinical manifestations and examination results. In addition, although large language models perform well in general text generation, existing solutions are still significantly inadequate in scenarios requiring the integration of specialized medical knowledge and multiple interactions with clinicians to complete diagnostic reports; current technologies cannot provide sufficient intelligent assistance. Therefore, an innovative technological solution is urgently needed to address these issues and achieve intelligent and personalized medical diagnostic report generation. Summary of the Invention

[0004] To at least partially address one of the technical problems existing in the prior art, the present invention aims to provide a method and related equipment for generating medical diagnostic reports based on large models and interaction, applicable to scenarios where structured diagnostic reports are generated through interaction with users using a large language model (LLM).

[0005] The first technical solution adopted in this invention is:

[0006] A method for generating medical diagnostic reports based on large models and interactions includes the following steps:

[0007] Information collection and preprocessing: Acquire patients' medical information through multimodal data collection methods;

[0008] Intelligent guided completion: Based on medical information, the large model generates a consultation questionnaire to allow patients to supplement information, and adds medical information based on the information supplemented by the patients; it detects medical inconsistencies in the medical information and prompts doctors to review in real time to ensure the accuracy and consistency of clinical descriptions;

[0009] Feature extraction and interaction: The large model identifies symptom and sign entities based on medical information and constructs a medical knowledge graph; it detects contradictions between signs and signs and prompts doctors to review and confirm.

[0010] Diagnostic criteria and interaction: Disease diagnoses are matched based on the constructed medical knowledge graph to form diagnostic criteria for doctors to confirm and adjust.

[0011] Treatment suggestions and interaction: Generate a draft of a personalized treatment plan based on the diagnosis results, and support doctors to make interactive modifications, and finally obtain the treatment plan text after the doctor's confirmation;

[0012] Report generation and interaction: Generate a draft report based on the treatment plan text, show doctors the results of medical terminology standardization, and receive feedback from doctors on version differences;

[0013] Output generation and verification: After the large model generates the final diagnostic report, it displays the integrity verification results to the doctor and outputs standardized medical documents after the doctor's confirmation.

[0014] Furthermore, the medical information includes electronic medical record text data T, doctor-patient dialogue voice signals S, and medical image data D;

[0015] The information collection and preprocessing includes:

[0016] Using a medical acoustic model, the doctor-patient dialogue voice signal S is converted into text S. t ;

[0017] Using OCR technology, medical image data (D) is recognized as text (D). t ;

[0018] The converted text is displayed in real time for doctors to preview and correct. When a clinical description conflict is detected between multimodal medical information, the doctor is prompted to confirm the main data source through the interactive interface.

[0019] The final output is a union of multimodal medical data. This provides standardized input for subsequent diagnostic analysis.

[0020] Furthermore, the intelligent guided completion includes:

[0021] Based on medical information, the patient's preliminary diagnosis type is obtained, and a medical consultation questionnaire is generated through the clinical interaction matrix G. Where y represents the specific report type index selected by the physician, X is the current clinical data, and f g Generate networks for medical problems. This indicates a clinical feature splicing operation;

[0022] By analyzing the doctor's input, the logical conflict score s of two consecutive symptom descriptions is calculated:

[0023] s=σ(W c [x t |x t-1 ])∈[0,1]

[0024] In the formula, W c For medical logic verification weights, [·|·] represents the concatenation of symptom features, x t x represents the symptom feature vector input by the doctor in the current round (time t). t-1 This represents the symptom feature vector recorded in the previous round (time t-1), where σ is the standardization function; when s < θ, a medical clarification request is automatically generated; θ is the clinical consistency threshold.

[0025] The multi-round consultation process maintains the treatment status through a medical memory matrix M, with the following update rules:

[0026] M t+1 =f r (M t ,x t )

[0027] In the formula, f r To update the network for recursive states, when |M t+1 The consultation process is terminated when |0≥l, ensuring that the large model can continuously guide doctors to improve key symptom information until the clinical data completeness standard required for diagnosis is reached; l is the clinical information completeness threshold.

[0028] Furthermore, the feature extraction and interaction include:

[0029] Based on the input medical information, the medical feature extraction function f is used. e Extract the entity set ε of symptoms and signs, and construct a medical knowledge graph. Among the edge sets It includes R types of medical relationships;

[0030] Show the doctor the results of the symptom progression timeline verification:

[0031]

[0032] In the formula, ei ,e j For the physical manifestations of symptoms and signs, τ j τ i Indicates a medical entity type label;

[0033] When α ij =1 indicates that symptom i should appear after symptom j but in reverse order, prompting the doctor to check for contradictions; it receives the doctor's instructions to adjust the symptom timing relationship and updates the medical knowledge graph in real time.

[0034] Furthermore, the diagnostic criteria and interactions include:

[0035] Through medical knowledge graphs To achieve disease diagnosis matching; among which For the disease diagnosis node set, ε m The relationships between diseases;

[0036] The patient's clinical feature vector is:

[0037] v p =f e (X)

[0038] In the formula, f e X is the medical feature extraction function; X represents the current clinical data. (m represents the number of symptom indicators collected); potential disease diagnoses are calculated using the following similarity:

[0039]

[0040] In the formula, For the disease coding function, s i ∈[-1,1] represents the clinical match between the patient's symptoms and the i-th disease;

[0041] Diagnostic validity verification is achieved through a guideline timeliness detection function:

[0042] φ i =f v (dis i )∈{0,1}

[0043] In the formula, f v For the discriminant function of clinical guidelines, dis i It is the i-th candidate disease in the disease set, when φ i A value of 0 indicates that the diagnostic criteria have been updated; the final output is a set of candidate diagnoses confirmed by the doctor.

[0044]

[0045] In the formula, γ is the clinical matching threshold;

[0046] Doctors adjust via interactive interface The content and priority order are determined to ensure that diagnostic recommendations are in line with the latest clinical practice guidelines and individual patient characteristics.

[0047] Furthermore, the treatment recommendations and interactions include:

[0048] Using the medical text generation function f g To achieve automatic generation of personalized treatment plans, the generation formula is as follows:

[0049]

[0050] In the formula, ε is the clinical feature matrix. The diagnostic basis matrix; T0 represents the text vector representation of the initially generated treatment plan;

[0051] During the doctor-interactive verification phase, the modified treatment plan is verified using the medical coding function f. e Convert to a format that can be processed according to clinical rules:

[0052] R = f e (T u )

[0053] In the formula, T u The treatment plan is represented by a vector text after the doctor's interactive modifications.

[0054] Clinical rule validation passes medical decision function f d →0,1 p Implementation, where p is the preset number of clinical rules:

[0055] a = f d (R)

[0056] The function outputs a vector of verification results, a = (a1, ..., a2). p Each component a) j ∈0,1 represents the satisfaction state of the j-th clinical rule, when At that time, the large model automatically identifies specific violations and generates medical correction suggestions M;

[0057] The final output is the treatment plan text that has passed all clinical rule validations:

[0058] T f =T u ⊙(1-W m M)

[0059] In the formula, T f It is the text vector representation of the treatment plan that has finally passed clinical rule validation, W mMedical corrected weight matrix, where ⊙ represents element-wise multiplication.

[0060] Furthermore, the report generation and interaction include:

[0061] Mapping function f using medical report templates t To achieve standardized report generation, input content is mapped to a medical report template space:

[0062]

[0063] In the formula, ε is the clinical feature matrix. As a diagnostic basis matrix, Treatment plan matrix; M s This represents the preset medical report template structure vector, where s is the number of standard report paragraphs; This is a segment-level feature splicing operation for medical procedures.

[0064] Medical terminology standardization is achieved through the clinical language standardization function f. n accomplish:

[0065] D1 = f n (D0)⊙W n

[0066] In the formula, W n Let be the medical terminology transformation matrix, where ⊙ represents a linear transformation of the medical feature space;

[0067] The report version comparison function uses the medical difference detection operator Δ to achieve visual annotation:

[0068] V=Δ(D1,D u )

[0069] Where each position v of the output V i D represents the intensity of medical content differences in RGB color encoding. u This represents the interactively modified version of the report. The final output after interactively adjusting the paragraph order (π) is:

[0070] D f =Permute(D u ,π)

[0071] In the formula, Permute(·) is the permutation operator.

[0072] Furthermore, the output generation and verification include:

[0073] Mapping function using medical report templates To achieve standardized report generation, The clinical feature space output by the modular interaction layer. A collection of pre-set medical report templates;

[0074] First, load the medical template T corresponding to the specialty type k selected by the doctor. k And verify the completeness of clinical elements:

[0075]

[0076] In the formula, X is the clinical element matrix to be verified, Q is the number of necessary medical elements, and m j For medical element matching vectors, I() means return 1 when the condition is met and 0 when the condition is not met; τ is the clinical similarity threshold.

[0077] Report format standardization is achieved through medical compound functions:

[0078]

[0079] In the formula, g is a medical format conversion function that maps medical semantic features to a three-dimensional medical format space (font size, line spacing, paragraph indentation); ° indicates that function f is executed first. T The result is then passed to function g for further processing; f T This is a medical report template mapping function used to generate standardized reports;

[0080] The final output is a standardized diagnostic report that conforms to medical standards:

[0081]

[0082] In the formula, b is the medical format compliance marker vector, F is the number of medical document standard items, and L is the number of standard paragraphs in the corresponding report.

[0083] The second technical solution adopted in this invention is:

[0084] An electronic device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set, or an instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement a medical diagnostic report generation method based on a large model and interaction as described above.

[0085] The third technical solution adopted in this invention is:

[0086] A computer-readable storage medium storing at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by a processor to implement a medical diagnostic report generation method based on a large model and interaction as described above.

[0087] The fourth technical solution adopted in this invention is:

[0088] A computer program product or computer program includes computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium and execute the computer instructions, causing the computer device to perform the aforementioned method for generating medical diagnostic reports based on a large model and interaction.

[0089] The beneficial effects of this invention are as follows: By employing multimodal data acquisition and dynamic consultation path adjustment, this invention ensures the collection of comprehensive clinical information and reduces the omission of key signs. The introduction of symptom-examination result consistency verification and diagnostic basis sufficiency assessment mechanisms guarantees the logical rigor of medical judgments. Through the construction of a medical knowledge graph and evidence-based matching technology, relevant treatment guidelines are automatically provided, avoiding omissions and erroneous citations inherent in manual review. The automatic generation of initial drafts of differential diagnoses and treatment recommendations using a large-scale language model, combined with review by a clinical rule engine, improves the professionalism and security of the report content. The application of a specialist template engine ensures standardized and uniform output report formats, and terminology standardization makes medical expressions more rigorous and accurate. Clinicians participate in the modification process through multiple rounds of interaction, retaining their professional judgment while fully utilizing the intelligent support provided by AI, thus ensuring that the final diagnostic report conforms to actual clinical conditions and meets medical standards and specialist requirements. In summary, this invention achieves an intelligent upgrade of the medical diagnostic report generation process, significantly improving the accuracy, standardization, and clinical applicability of the reports. Attached Figure Description

[0090] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following description is provided with accompanying drawings of the relevant technical solutions in the embodiments of the present invention or the prior art. It should be understood that the accompanying drawings described below are only for the purpose of clearly illustrating some embodiments of the technical solutions of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0091] Figure 1 This is a schematic diagram of the architecture of the medical diagnostic report generation method based on a large model and interaction in an embodiment of the present invention;

[0092] Figure 2 This is a flowchart illustrating the medical diagnostic report generation method based on a large model and interaction in an embodiment of the present invention;

[0093] Figure 3 This is a flowchart illustrating the steps of a medical diagnostic report generation method based on a large model and interaction in an embodiment of the present invention. Detailed Implementation

[0094] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application. The step numbers in the following embodiments are set only for ease of explanation, and there is no limitation on the order between the steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0095] The terminology used in the embodiments of this application is for the purpose of describing specific embodiments only and is not intended to limit the embodiments of this application. The singular forms "a," "described," and "the" used in the embodiments of this application and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise. Furthermore, unless otherwise expressly limited, terms such as "set," "install," and "connect" should be interpreted broadly, and those skilled in the art can reasonably determine the specific meaning of the above terms in this invention in conjunction with the specific content of the technical solution.

[0096] In the description of this application, it should be understood that the orientation descriptions, such as up, down, front, back, left, right, etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application.

[0097] In the description of this application, "several" means one or more, "more than" means two or more, "greater than," "less than," and "exceeding" are understood to exclude the stated number, while "above," "below," and "within" are understood to include the stated number. The use of "first" and "second" in the description is merely for distinguishing technical features and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or implicitly indicating the order of the indicated technical features.

[0098] In the description of this application, "and / or" describes the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the related objects before and after it are in an "or" relationship.

[0099] Terminology Explanation:

[0100] LLM: an abbreviation for Large Language Model, is a deep learning-based artificial intelligence technique.

[0101] The technical problem this invention aims to solve is to improve the intelligence and efficiency of medical diagnostic report generation, avoiding problems such as missing information, logical contradictions, and improper citation of medical evidence that may occur during manual drafting. It also overcomes the limitations of traditional electronic medical record systems, such as rigid templates, inaccurate extraction of key features, and insufficient understanding of clinical context. Specifically, the core issue addressed by this invention is how to combine the natural language processing capabilities of large-scale models with multi-turn interactions with clinicians to provide a method that dynamically guides doctors to improve patient information, automatically extracts key clinical features, matches relevant medical evidence, and generates standardized diagnostic reports.

[0102] To address the aforementioned technical challenges, this invention provides a medical diagnostic report generation method based on a large model and multi-round user interaction. The method employs a three-layer architecture: an input layer, a modular interaction layer, and an output layer. First, in the input layer, a multimodal data acquisition system acquires patient complaints, examination and test results, and medical history. An intelligent guided questionnaire dynamic generation mechanism ensures the completeness and accuracy of clinical information. Then, in the modular interaction layer, the report generation task is decomposed into four core modules: a clinical feature extraction module identifies symptom and sign entities and verifies temporal relationships; a diagnostic basis module integrates a medical knowledge graph and matches evidence-based medicine; a treatment suggestion module generates personalized plans and examines drug interactions; and a report structuring module intelligently determines the document type and adapts to a specialty template. Leveraging the capabilities of a large language model, the system provides automatically generated suggestions for each module. Clinicians can perform multi-round interactive revisions of the content in each module, progressing from one module to the next after successful verification, gradually refining the report content. Finally, in the output layer, the results from each module are synthesized, processed with standardized terminology, and formatted to generate a professional diagnostic report that conforms to medical standards.

[0103] Compared with existing technologies, this invention significantly improves the efficiency and quality of medical diagnostic report generation. Through multimodal data acquisition and dynamic consultation path adjustment, it ensures comprehensive collection of clinical information and reduces the omission of key signs. The introduction of symptom-examination result consistency verification and diagnostic basis sufficiency assessment mechanisms guarantees the logical rigor of medical judgments. By constructing a medical knowledge graph and using evidence-based matching technology, relevant treatment guidelines are automatically provided, avoiding oversights and erroneous citations inherent in manual review. Utilizing a large-scale language model to automatically generate initial drafts of differential diagnoses and treatment recommendations, combined with review by a clinical rule engine, improves the professionalism and security of the report content. The application of a specialist template engine ensures standardized and consistent output report formats, and terminology standardization makes medical expressions more rigorous and accurate. Clinicians participate in the modification process through multiple rounds of interaction, retaining their professional judgment while fully utilizing the intelligent support provided by AI, thus ensuring that the final diagnostic report conforms to actual clinical conditions and meets medical standards and specialist requirements. In summary, this invention achieves an intelligent upgrade of the medical diagnostic report generation process, significantly improving the accuracy, standardization, and clinical applicability of the reports.

[0104] Example 1

[0105] like Figure 3 As shown, this embodiment provides a method for generating medical diagnostic reports based on a large model and interaction, including the following steps:

[0106] S1. Information Collection and Preprocessing: Acquire patients' medical information through multimodal data collection methods.

[0107] In some embodiments, medical information input provided by the patient is received through a multimodal interface, including electronic medical record text data T, doctor-patient dialogue voice signal S, and medical image data D.

[0108] The medical-specific acoustic model is used to convert the doctor-patient dialogue voice signal S into text S. t Medical images are converted into text using deep learning OCR. t The final output is multimodal medical data and collection. The system displays a real-time preview of the converted text for doctors to review. When there are clinical description conflicts between multimodal medical information, the system prompts doctors to confirm the primary data source through an interactive interface. For example, the system here refers to software capable of running large models and a device with display and input functions, such as a computer.

[0109] S2. Intelligent Guided Completion: The large model generates a consultation questionnaire to allow patients to supplement information, and adds medical information based on the information provided by the patients; it detects medical inconsistencies in the medical information and prompts doctors to review it in real time to ensure the accuracy and consistency of the clinical description.

[0110] In some embodiments, a medical consultation questionnaire is generated through a clinical interaction matrix G based on the patient's initial diagnosis type. Where y represents the selected specific report type index (1≤y≤n), and doctors can supplement or revise the questionnaire content. The system calculates the clinical logical conflict score s=σ(W) based on the doctor's feedback. c [x t |x t-1 When s < θ (clinical consistency threshold), a medical clarification request is generated, requiring the doctor to confirm the contradictions. Simultaneously, the system maintains the consultation interaction state through a medical memory matrix M, terminating the guidance process when the completeness of clinical information meets diagnostic requirements.

[0111] S3. Feature Extraction and Interaction: The large model identifies symptom and sign entities based on medical information and constructs a medical knowledge graph; it detects contradictions between signs and signs and prompts doctors to review and confirm.

[0112] In some embodiments, through the medical feature extraction function f e Extract the entity set ε of symptoms and signs to construct a clinical knowledge graph. The system displays the symptom progression timeline verification results to the doctor. It receives instructions from doctors to adjust the temporal relationship of symptoms and updates the clinical knowledge graph structure in real time.

[0113] S4. Diagnostic Criteria and Interaction: Based on the constructed medical knowledge graph, disease diagnoses are matched to form diagnostic criteria for doctors to confirm and adjust.

[0114] In some embodiments, through medical knowledge graphs Match potential diseases and calculate the patient's clinical feature vector v p Disease feature vector Medical similarity s i The system recommends a set of candidate diagnoses validated by the latest clinical guidelines to doctors. It also allows doctors to manually adjust diagnostic priorities and confidence levels.

[0115] S5. Treatment suggestions and interaction: Generate a draft of a personalized treatment plan based on the diagnosis results, and support doctors to make interactive modifications, and finally obtain the treatment plan text after the doctor's confirmation.

[0116] In some embodiments, through the medical text generation function f g Generate initial treatment plan For doctors to edit. The system will then display the edited treatment text (T) by the doctor. u The code is in the clinical rule format R=f e (T uThe system performs medical compliance checks, generates a correction suggestion matrix M when drug interactions or treatment risks are detected, and finally outputs a safe treatment plan text T confirmed by the doctor. f .

[0117] S6. Report Generation and Interaction: Generate a draft report based on the treatment plan text, show doctors the results of medical terminology standardization, and receive feedback from doctors on version differences.

[0118] In some embodiments, the specialist report template mapping function f t Generate a draft diagnostic report D0; after performing standardized medical terminology processing, present the version differences V = Δ(D1, D2) to the physician. u It supports adjusting the report paragraph structure and content order through a visual interface to generate a final structured medical diagnostic report (D). f .

[0119] S7. Output Generation and Verification: After the large model generates the final diagnostic report, it displays the integrity verification results to the doctor and outputs standardized medical documents after the doctor's confirmation.

[0120] In some embodiments, the essential elements of the report are checked using the clinical diagnostic element integrity verification formula Φ(X). When key clinical information is detected as missing, an intelligent completion mechanism is triggered. The medical document format compliance verification results are shown to the doctor, and a standardized diagnostic report is output after the doctor's final confirmation. The final medical documents, in DOCX or PDF format, conforming to hospital standards, are generated through the medical report rendering interface Render.

[0121] The above method will be explained in detail below with reference to the accompanying drawings and specific embodiments.

[0122] like Figure 1 and Figure 2 As shown in the figure, this embodiment provides a medical diagnostic report generation method based on a large model and interaction. This method adopts a three-layer system architecture of input layer, modular interaction layer and output layer to realize the intelligent generation of medical diagnostic reports. The specific implementation steps are as follows:

[0123] Step 1: Input Layer - Information Acquisition and Preprocessing: The system receives patient information, including chief complaint text, examination reports, and medical images, through a multimodal data acquisition system. Voice input is converted to text using speech recognition technology, and editable text is extracted from paper reports using OCR technology. The system standardizes all types of input information to form structured clinical data for use by subsequent modules.

[0124] As an optional implementation, in step 1, the system receives patient medical information input, including text input, through a multimodal interface. (where L is the sequence length and d is the word vector dimension), voice input (Where T is the time step and F is the number of Mel-frequency bands) and medical image input (Where H×W is the image resolution). Voice input is processed through an acoustic transformation matrix. (where V is the size of the phoneme set) and language conversion matrix (where Σ is a character set) is processed into text format:

[0125]

[0126] Where argmax(·) represents taking the maximum probability along the character dimension. Medical images are processed by convolutional kernel groups. (where K is the number of convolution channels) and character decoding matrix Convert to text:

[0127]

[0128] Where Conv2D(·) represents a two-dimensional convolution operation, and W′ is the width of the feature map after convolution. The system's final output is the union of multimodal medical data.

[0129] Step 2: Input Layer - Intelligent Guided Completion: Based on the patient's initial symptoms, the system dynamically generates a clinical questionnaire, ensuring the completeness of key vital signs information through multiple rounds of interaction. Simultaneously, the system automatically detects medical inconsistencies in the input information, providing real-time prompts for doctor review to ensure the accuracy and consistency of the clinical description.

[0130] As an optional implementation, in step 2, the system uses a clinical feature matrix. (where n is the total number of diagnostic report types and d is the medical feature dimension) to achieve intelligent consultation guidance. After the doctor selects a report type, the system uses current clinical data... (m represents the number of symptom indicators collected) Dynamically generate medical consultation questionnaires:

[0131]

[0132] in A network is generated for medical questions, where y represents the index of a specific report type selected (1 ≤ y ≤ n). This represents the clinical feature concatenation operation, where k is the number of medical questions generated. The consultation content is decoded using the medical knowledge decoding function f. d accomplish:

[0133]

[0134] in Identify the clinical indicator index set that needs improvement. The system completes the conversion of medical features into natural language. It analyzes doctor input through a clinical consistency detection module, calculating logical conflict scores between two consecutive symptom descriptions.

[0135] s=σ(W c [x t |x t-1 ])∈[0,1]

[0136] in For medical logic verification weights, [·|·] represents the concatenation of symptom features, x t x represents the symptom feature vector input by the doctor in the current round (time t). t-1 This represents the symptom feature vector recorded in the previous round (time t-1). This is a standardized function. When s < θ (clinical consistency threshold θ), the system automatically generates a medical clarification request. The multi-round consultation process utilizes a medical memory matrix. (T represents the consultation round) Maintain the treatment status, and update the following rules:

[0137] M t+1 =f r (M t ,x t )

[0138] in To update the network for recursive states, when |M t+1 The consultation process is terminated when |0≥l (l is the clinical information completeness threshold), ensuring that the system can continuously guide doctors to improve key symptom information until the clinical data completeness standard required for diagnosis is reached.

[0139] Step 3: Modular Interaction Layer - Clinical Feature Extraction Module: The system automatically identifies symptom and sign entities and constructs a symptom time-series relationship diagram. Medical logic verification ensures the rationality of the symptom development process and detects inconsistencies between signs, generating prompts for doctors to review and confirm.

[0140] As an optional implementation, in step 3, the system uses a medical entity recognition function. (Where L is the length of the clinical text, d is the dimension of the medical word vector, and k is the number of medical entities) Key clinical features are automatically extracted from patient medical records. The core formula for this process is:

[0141]

[0142] Where ε represents the extracted set of medical entities, each entity Includes clinical feature vectors, t i ∈1,…,T represents medical entity type labels (T is the preset total number of clinical entity types).

[0143] The system stores medical entities and their relationships in a medical knowledge graph. In, the edge set R predefined medical relation types are defined. Symptom temporal validation is performed using a time feature extraction function. To achieve this, the key decision-making formula is:

[0144]

[0145] Where τ i and τ j They represent symptoms e respectively i and e j The time feature vector is passed through g t The time series values ​​obtained after function mapping, α ij When the value is 1 (indicating that symptom i should appear after symptom j but in reverse chronological order), the system will prompt the doctor to check for contradictions. The core calculation formula of the clinical contradiction detection module is:

[0146]

[0147] in Let e ​​be the medical contradiction feature mapping function. p and e q Let represent the symptom pairs to be detected, and δ be the clinical judgment threshold. For the set... For each symptom in the list, the system will generate specific medical review recommendations.

[0148] Through the above technical solution, the system realizes a complete process from clinical feature extraction to medical contradiction detection. Doctors can adjust the relationship between symptoms through a visual interface, and the system updates in real time. And Φ and recalculate This module establishes a closed-loop clinical verification process. It transforms traditional medical feature verification into a quantifiable calculation process, significantly improving the clinical accuracy of diagnostic reports.

[0149] Step 4: Modular Interaction Layer - Diagnostic Basis Module: Based on the patient's clinical presentation, the system automatically matches relevant diseases in the medical knowledge graph, calculates the similarity of clinical features, and filters the most likely diagnosis. Simultaneously, it verifies the timeliness of the diagnostic criteria, excludes outdated treatment guidelines, and forms reliable diagnostic basis for doctors to confirm and adjust.

[0150] As an optional implementation, in step 4, the system uses a medical knowledge graph. Achieve disease diagnosis matching (wherein) For disease diagnosis node set, (This represents the association between diseases). The vectorized encoding process of patient clinical characteristics is represented as follows:

[0151]

[0152] in This is a medical feature extraction function. The system calculates and matches potential disease diagnoses using the following similarity metrics:

[0153]

[0154] in For the disease coding function, s i ∈[-1,1] represents the clinical match between the patient's symptoms and the i-th disease. Diagnostic validity is validated using a guideline timeliness detection function.

[0155] φ i =f v (dis i )∈{0,1}

[0156] Where f v For the discriminant function of clinical guidelines, dis i It is the i-th candidate disease in the disease set, when φ i A value of 0 indicates that the diagnostic criteria have been updated. The final output is a set of candidate diagnoses confirmed by the doctor.

[0157]

[0158] Where γ is the clinical matching threshold, which doctors can adjust through the interactive interface. The content and priority order should be determined to ensure that diagnostic recommendations are in line with the latest clinical practice guidelines and individual patient characteristics.

[0159] Step 5: Modular Interaction Layer - Treatment Recommendation Module: The system automatically generates a draft of a personalized treatment plan based on the diagnosis results, supporting interactive modifications by doctors. The drug interaction checking engine reviews the rationality of the treatment plan and provides corrective suggestions for risky drug combinations.

[0160] As an optional implementation, in step 5, the system uses a medical text generation function. This process enables the automatic generation of personalized treatment plans (where m is the number of clinical features, n is the number of diagnostic criteria, and L is the length of the generated text). This process is based on the output of the clinical feature extraction module. and the output of the diagnostic basis module The core generation formula is:

[0161]

[0162] Where T0 represents the initially generated text vector representation of the treatment plan. During the doctor's interactive verification phase, the modified treatment plan is processed through a medical coding function. Converted to a clinically rule-processable format (k is the dimension of medical rule features):

[0163] R = f e (T u )

[0164] Clinical rule validation passes medical decision function Implementation (p is the preset number of clinical rules):

[0165] a = f d (R)

[0166] The function outputs a vector of verification results, a = (a1, ..., a2). p ), each component a j ∈0,1 represents the satisfaction state of the j-th clinical rule, when At that time, the system automatically identifies specific violations and generates medical correction suggestions. The final output is the treatment plan text that has passed all clinical rule validations:

[0167] T f =T u ⊙(1-W m M)

[0168] Where T f It is the text vector representation of the treatment plan that has finally passed clinical rule validation, T u It is a text vector representation of the treatment plan after the doctor's interaction and modification. The medical correction weight matrix, where ⊙ represents element-wise multiplication, ensures that the treatment plan is both tailored to the individual characteristics of the patient and complies with the latest clinical treatment guidelines and drug safety standards.

[0169] Step 6: Modular Interaction Layer - Report Structure Module: The system organizes the processing results from each module in a structured manner according to medical document standards, automatically applying the corresponding specialty templates. The system ensures accurate professional terminology through medical terminology standardization and provides a version comparison function to facilitate doctors' review of modifications.

[0170] As an optional implementation, in step 6, the system maps medical report templates using a function. This process enables standardized report generation (where m is the number of clinical features, n is the number of diagnostic criteria, p is the number of treatment plan entries, and L is the total report length). The process first maps the input content to a medical report template space:

[0171]

[0172] in This is a clinical feature matrix. As a diagnostic basis matrix, A treatment plan matrix, This represents the preset medical report template structure vector (s is the number of standard report paragraphs). This is a segment-level feature concatenation operation for medical terminology. Medical terminology standardization is achieved through the clinical language standardization function. accomplish:

[0173] D1 = f n (D0)⊙W n

[0174] in This is a medical terminology transformation matrix, where ⊙ represents a linear transformation of the medical feature space. The report version comparison function uses a medical difference detection operator. Achieve visual annotation:

[0175] V=Δ(D1,D u )

[0176] Where each position of output V D represents the intensity of medical content differences in RGB color encoding. u This represents the interactively modified version of the report. The final output after interactively adjusting the paragraph order (π) is:

[0177] D f =Permute(D u ,π)

[0178] The permutation operator Permute(·) maintains the clinical semantic coherence between medical paragraphs, ensuring that the generated diagnostic report not only conforms to medical documentation standards but also accurately reflects the individualized diagnosis and treatment process of the patient.

[0179] Step 7: Output Layer - Report Generation and Verification: The system performs integrity verification on the generated diagnostic report to ensure that all necessary clinical elements are included. Through format standardization processing, it outputs standardized diagnostic reports that conform to medical standards, supporting export in multiple formats such as DOCX and PDF.

[0180] As an optional implementation, in step 7, the system maps medical report templates using a function. To achieve standardized report generation, The clinical feature space output by the modular interaction layer. This is a pre-defined set of medical report templates (N is the number of specialty template types). The system first loads the medical templates corresponding to the specialty types k∈1,…,N selected by the doctor. (L is the number of standard medical report paragraphs), and perform clinical element integrity verification:

[0181]

[0182] Among them is the clinical element matrix to be verified (Q is the number of necessary medical elements), is the medical element matching vector, and τ is the clinical similarity threshold. When Φ(X) < Q, the system automatically triggers the medical information completion mechanism. The report format normalization is achieved through a medical composite function:

[0183]

[0184] Among them is the medical format conversion function, which maps medical semantic features to a three-dimensional medical format space (font specification, line spacing, paragraph indentation). The final output is a standardized diagnostic report that meets medical specifications:

[0185]

[0186] where b ∈ 0, 1 F is the medical format compliance marking vector (F is the number of medical document specification items), and it satisfies |b - b0|0 = 0 (b0 is the preset medical specification standard). The system uses the medical report rendering function to generate a printable medical document that meets hospital standards, completing the full-process intelligent processing from clinical data collection to standardized report output, ensuring that the generated diagnostic report not only meets the requirements of medical document specifications but also completely and accurately reflects the entire process of the patient's diagnosis and treatment.

[0187] Example 2

[0188] The embodiment of the present invention also provides an electronic device, the electronic device includes a processor and a memory, and at least one instruction, at least one program, a code set or an instruction set is stored in the memory, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement as Figure 3 shown in a medical diagnostic report generation method based on a large model and interaction.

[0189] It is understood that the memory may include random access memory (RAM) or read-only memory. Optionally, the memory may include non-transitory computer-readable storage medium. The memory can be used to store instructions, programs, code, code sets, or instruction sets. The memory may include a stored program area and a stored data area, wherein the stored program area may store instructions for implementing an operating system, instructions for at least one function, instructions for implementing the various method embodiments described above, etc.; the stored data area may store data created according to the use of the server, etc.

[0190] A processor may include one or more processing cores. The processor connects to various parts of the server via various interfaces and lines, executing instructions, programs, code sets, or instruction sets stored in memory, and accessing data stored in memory to perform various server functions and process data. Optionally, the processor may be implemented using at least one of the following hardware forms: Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), and Programmable Logic Array (PLA). The processor may integrate one or more of the following: Central Processing Unit (CPU) and Modem. The CPU primarily handles the operating system and applications; the modem handles wireless communication. It is understood that the modem may also be implemented as a separate chip without being integrated into the processor.

[0191] Since this electronic device is an electronic device corresponding to a medical diagnostic report generation method based on a large model and interaction according to an embodiment of the present invention, and the principle of solving the problem by this electronic device is similar to that of this method, the implementation of this electronic device can refer to the implementation process of the above method embodiment, and the repeated parts will not be described again.

[0192] Example 3

[0193] This invention also provides a computer-readable storage medium storing at least one instruction, at least one program, a code set, or an instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by a processor to achieve the following: Figure 3 This illustrates a method for generating medical diagnostic reports based on a large model and interaction.

[0194] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0195] Since this storage medium is the storage medium corresponding to a medical diagnostic report generation method based on a large model and interaction in an embodiment of the present invention, and the principle of the storage medium in solving the problem is similar to that of the method, the implementation of this storage medium can refer to the implementation process of the above method embodiment, and the repeated parts will not be described again.

[0196] Example 4

[0197] In some possible implementations, various aspects of the methods of the embodiments of the present invention can also be implemented as a program product comprising program code that, when run on a computer device, causes the computer device to perform the steps of a large-model-based and interactive medical diagnostic report generation method according to various exemplary embodiments of this application as described above. The executable computer program code or "code" for performing the various embodiments can be written in high-level programming languages ​​such as C, C++, Python, Smalltalk, Java, JavaScript, Visual Basic, Structured Query Language (e.g., Transact-SQL), Perl, or in various other programming languages.

[0198] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0199] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0200] The above embodiments are merely illustrative of the technical concept and features of the present invention, and are intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly. They should not be construed as limiting the scope of protection of the present invention. All equivalent changes or modifications made based on the essence of the content of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for generating medical diagnostic reports based on large models and interaction, characterized in that, The following steps are involved: Information collection and preprocessing: Acquire patients' medical information through multimodal data collection methods; Intelligent guided completion: The large model generates a consultation questionnaire to allow patients to supplement information, and adds medical information based on the information supplemented by the patients; Detect medical inconsistencies in medical information and prompt doctors to review them; Feature extraction and interaction: The large model identifies symptom and sign entities based on medical information and constructs a medical knowledge graph; it detects contradictions between signs and signs and prompts doctors to review and confirm. Diagnostic criteria and interaction: The large model matches disease diagnoses based on the constructed medical knowledge graph, forming diagnostic criteria for doctors to confirm and adjust. Treatment suggestions and interaction: The large model generates a draft of a personalized treatment plan based on the diagnosis results, and supports doctors to make interactive modifications, and finally obtain the treatment plan text after the doctor's confirmation; Report generation and interaction: The large model generates a draft report based on the treatment plan text, shows the doctor the results of medical terminology standardization, and receives feedback from the doctor on the differences between the versions; Output generation and verification: After the large model generates the final diagnostic report, it displays the integrity verification results to the doctor and outputs standardized medical documents after the doctor's confirmation.

2. The method for generating medical diagnostic reports based on a large model and interaction according to claim 1, characterized in that, The medical information includes electronic medical record text data T, doctor-patient dialogue voice signals S, and medical image data D; The information collection and preprocessing includes: Using a medical acoustic model, the doctor-patient dialogue voice signal S is converted into text S. t ; Using OCR technology, medical image data (D) is recognized as text (D). t ; The converted text is displayed in real time for doctors to preview and correct. When a clinical description conflict is detected between multimodal medical information, the doctor is prompted to confirm the main data source through the interactive interface. The final output is a union of multimodal medical data. This provides standardized input for subsequent diagnostic analysis.

3. The method for generating medical diagnostic reports based on a large model and interaction according to claim 1, characterized in that, The intelligent guided autocomplete includes: Based on the patient's initial diagnosis, a medical consultation questionnaire Q = f is generated using the clinical interaction matrix G. g (G[y]⊕X); where y represents the specific report type index selected by the physician, X is the current clinical data, and f g Generate networks for medical problems; ⊕ indicates a clinical feature splicing operation. By analyzing the doctor's input, the logical conflict score s of two consecutive symptom descriptions is calculated: s=σ(W c [x t |x t-1 ])∈[0,1] In the formula, W c For medical logic verification weights, [·|·] represents the concatenation of symptom features, x t x represents the symptom feature vector input by the doctor in the current round (time t). t-1 This represents the symptom feature vector recorded in the previous round (time t-1), where σ is the standardization function; when s < θ, a medical clarification request is automatically generated; θ is the clinical consistency threshold. The multi-round consultation process maintains the treatment status through a medical memory matrix M, with the following update rules: M t+1 =f r (M t ,x t ) Where, f r To update the network for recursive states, when |M t+1 The consultation process is terminated when |0≥l, ensuring that the large model can continuously guide doctors to improve key symptom information until the clinical data completeness standard required for diagnosis is reached; l is the clinical information completeness threshold.

4. The method for generating medical diagnostic reports based on a large model and interaction according to claim 1, characterized in that, The feature extraction and interaction include: Based on the input medical information, the medical feature extraction function f is used. e Extract the entity set ε of symptoms and signs, and construct a medical knowledge graph. Among the edge sets Includes R types of medical relationships; Show the doctor the results of the symptom progression timeline verification: In the formula, e i ,e j For the physical manifestations of symptoms and signs, τ j τ i Indicates a medical entity type label; When α ij =1 indicates that symptom i should appear after symptom j but in reverse order, prompting the doctor to check for contradictions; it receives the doctor's instructions to adjust the symptom timing relationship and updates the medical knowledge graph in real time.

5. The method for generating medical diagnostic reports based on a large model and interaction according to claim 1, characterized in that, The diagnostic criteria and interactions include: Through medical knowledge graphs To achieve disease diagnosis matching; among which For disease diagnosis node set, The relationships between diseases; The patient's clinical feature vector is: Where, f e Here, X represents the medical feature extraction function; X represents the current clinical data. Potential disease diagnoses are matched using the following similarity calculations: In the formula, For the disease coding function, s i ∈[-1,1] represents the clinical match between the patient's symptoms and the i-th disease; Diagnostic validity verification is achieved through a guideline timeliness detection function: φ i =f v (dis i )∈{0,1} Where, f v For the discriminant function of clinical guidelines, dis i It is the i-th candidate disease in the disease set, when φ i A value of 0 indicates that the diagnostic criteria have been updated; the final output is a set of candidate diagnoses confirmed by the doctor. In the formula, γ is the clinical matching threshold; Doctors adjust via interactive interface The content and priority order are determined to ensure that diagnostic recommendations are in line with the latest clinical practice guidelines and individual patient characteristics.

6. The method for generating medical diagnostic reports based on a large model and interaction according to claim 1, characterized in that, The treatment recommendations and interactions include: Using the medical text generation function f g To achieve automatic generation of personalized treatment plans, the generation formula is as follows: In the formula, ε is the clinical feature matrix. The diagnostic basis matrix is ​​represented by T0; T0 represents the initially generated text vector representation of the treatment plan; during the doctor's interactive verification phase, the modified treatment plan is processed through the medical coding function f. e Convert to a format that can be processed according to clinical rules: R=f e (T u ) In the formula, T u The treatment plan is represented by a vector text after the doctor's interactive modifications. Clinical rule validation passes medical decision function f d →0,1 p Implementation, where p is the preset number of clinical rules: a=f d (R) The function outputs a vector of verification results, a = (a1, ..., a2). p ), each component a j ∈0,1 represents the satisfaction state of the j-th clinical rule, when At that time, the large model automatically identifies specific violations and generates medical correction suggestions M; The final output is the treatment plan text that has passed all clinical rule validations: T f =T u ⊙(1-W m M) In the formula, T f It is the text vector representation of the treatment plan that has finally passed clinical rule validation, W m Medical corrected weight matrix, where ⊙ represents element-wise multiplication.

7. The method for generating medical diagnostic reports based on a large model and interaction according to claim 1, characterized in that, The report generation and interaction include: Mapping function f using medical report templates t To achieve standardized report generation, input content is mapped to a medical report template space: In the formula, ε is the clinical feature matrix. As a diagnostic basis matrix, Treatment plan matrix; M s This represents the preset medical report template structure vector, where s is the number of standard paragraphs in the report; ⊕ represents the medical paragraph-level feature splicing operation; Medical terminology standardization is achieved through the clinical language standardization function f. n accomplish: D1=f n (D0)⊙W n In the formula, W n Let be the medical terminology transformation matrix, where ⊙ represents a linear transformation of the medical feature space; The report version comparison function uses the medical difference detection operator Δ to achieve visual annotation: V=Δ(D1,D u ) Where each position v of the output V i D represents the intensity of medical content differences in RGB color encoding. u This represents the interactively modified version of the report. The final output after interactively adjusting the paragraph order (π) is: D f =Permute(D u ,π) In the formula, Permute(·) is the permutation operator.

8. The method for generating medical diagnostic reports based on a large model and interaction according to claim 1, characterized in that, The output generation and verification include: Mapping function using medical report templates To achieve standardized report generation, The clinical feature space output by the modular interaction layer. A collection of pre-set medical report templates; First, load the medical template T corresponding to the specialty type k selected by the doctor. k And verify the completeness of clinical elements: In the formula, Let Q be the clinical element matrix to be validated, where Q represents the number of necessary medical elements, and m represents the number of elements required. j For medical element matching vectors, I() means return 1 when the condition is true and 0 when the condition is false; τ is the clinical similarity threshold. Report format standardization is achieved through medical compound functions: In the formula, g is a medical format conversion function that maps medical semantic features to a three-dimensional medical format space; This means that function f will be executed first. T The result is then passed to function g for further processing; f T This is a medical report template mapping function used to generate standardized reports; The final output is a standardized diagnostic report that conforms to medical standards: In the formula, b is the medical format compliance marker vector, F is the number of medical document standard items, and L is the number of standard paragraphs in the corresponding report.

9. An electronic device, characterized in that, The electronic device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set, or an instruction set, and the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement the method as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The storage medium stores at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by a processor to implement the method as described in any one of claims 1 to 8.

Images