Intelligent pre-diagnosis form generation method and system for children's tumors

Abstract

The application provides a kind of intelligent pre-consultation form generation method and system for children's tumors, relating to the technical field of intelligent medical consultation, first identify the scene type of children's tumor pre-consultation and extract the scene demand features to form a set, obtain the multiple source input data such as parents' preliminary symptom description text, children's past health records and examination results, and process them in association. Then, based on the scene demand feature set, adjust the parameters of the children's tumor large model, input the associated data set to generate the consultation item semantic prototype. According to the parents' expression habit sample, optimize the expression logic to form the preliminary pre-consultation form, collect the parents' feedback to form a multi-dimensional feedback data set. Finally, input the feedback data set into the model to generate the form field optimization scheme and the preliminary diagnosis suggestion draft, and adjust the preliminary pre-consultation form to obtain the final children's tumor intelligent pre-consultation form containing the preliminary diagnosis direction. The application improves the scene adaptability and accuracy of pre-consultation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent medical diagnosis, in particular to an intelligent pre-diagnosis form generation method and system for children's tumors. BACKGROUND

[0002] In the field of children's tumor treatment, pre-diagnosis is a key preliminary link in the disease diagnosis and treatment process. Currently, children's tumor pre-diagnosis mainly relies on medical personnel to collect information through face-to-face communication or parents filling out simple questionnaires. However, there are differences in different pre-diagnosis scenarios. Initial diagnosis pre-diagnosis needs to collect children's basic information, symptom manifestations, etc.; follow-up scenarios focus on understanding disease progression, treatment effect, etc.; and special symptom screening scenarios mainly ask in-depth questions about specific symptoms. The existing pre-diagnosis method cannot be flexibly adjusted according to different scenarios, and cannot accurately meet the diversified needs of different scenarios.

[0003] At the same time, when parents describe children's symptoms, the expression methods are various and may be inaccurate and incomplete, and the data sources such as children's past health records and examination result sheets are scattered, lacking effective association and integration, resulting in low information utilization efficiency. In addition, the existing pre-diagnosis form is of a fixed format, without fully considering the expression habits of parents, with many professional terms, which are difficult for parents to understand, and the form content and diagnosis suggestions cannot be optimized in a timely manner according to the feedback of parents, affecting the accuracy and efficiency of pre-diagnosis, and being not conducive to the early detection and precise treatment of children's tumors. SUMMARY

[0004] In view of the above-mentioned problems, in combination with the first aspect of the present application, the embodiments of the present application provide an intelligent pre-diagnosis form generation method for children's tumors, which comprises:

[0005] Recognize the scene type of children's tumor pre-diagnosis, extract the scene demand characteristics corresponding to each scene type, form a scene demand characteristic set, and the scene type includes initial diagnosis pre-diagnosis scene, follow-up scene and special symptom screening scene;

[0006] Obtain pre-diagnosis related multi-source input data, and perform association processing on the multi-source input data to obtain an association data set, wherein the multi-source input data includes parent initial symptom description text, children's past health records and parent uploaded examination result sheets;

[0007] Adjust the parameter configuration of the children's tumor large model based on the scene demand characteristic set to obtain an adjusted children's tumor large model adapted to the scene, and the parameter configuration includes case report form structure generation parameters, examination result association weights and initial diagnosis suggestion generation thresholds;

[0008] The association data set is input into the adjusted child tumor large model to generate an inquiry entry semantic prototype containing case report form standard fields, and the inquiry entry semantic prototype is associated with examination result verification items and preliminary diagnosis direction prompts;

[0009] The expression logic of the inquiry entry semantic prototype is optimized according to the parent expression habit sample, the professional case report form fields are converted into expressions understandable by parents, the association guide sentences of symptoms and examination results are supplemented, the preliminary pre-consultation form is formed, the preliminary pre-consultation form is pushed to the parent interactive terminal, the form content filled by the parents, the symptom description and the supplementary explanation of the examination results are collected, and the form filling time, the field modification times and the skipping behavior of the parents are recorded, and a multi-dimensional feedback data set is formed;

[0010] The multi-dimensional feedback data set is input into the adjusted child tumor large model to analyze the association between the form content and the examination results and the causes of parent interaction obstacles, generate a form field optimization scheme and a preliminary diagnosis suggestion draft, adjust the field order and expression content of the preliminary pre-consultation form based on the form field optimization scheme, supplement the diagnosis related verification items in combination with the preliminary diagnosis suggestion draft, and obtain a final intelligent pre-consultation form for child tumors containing a preliminary diagnosis direction.

[0011] In still another aspect, the embodiments of the present application also provide an intelligent pre-consultation form generation system for child tumors, comprising:

[0012] A processor; a machine readable storage medium for storing machine executable instructions of the processor; wherein the processor is configured to execute the machine executable instructions to perform the intelligent pre-consultation form generation method for child tumors described above.

[0013] In still another aspect, the embodiments of the present application also provide a computer program product, the computer program product comprising machine executable instructions stored in a computer readable storage medium, and the processor of the intelligent pre-consultation form generation system for child tumors reads the machine executable instructions from the computer readable storage medium, and the processor executes the machine executable instructions, so that the intelligent pre-consultation form generation system for child tumors executes the intelligent pre-consultation form generation method for child tumors described above.

[0014] Based on the above aspects, by identifying different scene types of child tumor pre-diagnosis and extracting corresponding scene demand features to form a set, the pre-diagnosis focus in each scene can be known, and the scene adaptability of pre-diagnosis can be effectively improved. By obtaining multi-source input data and performing correlation processing, scattered information can be integrated, data utilization efficiency can be improved, and the information on which the pre-diagnosis is based can be more comprehensive and accurate. Based on the scene demand feature set, the parameters of the child tumor large model are adjusted to obtain a model adapted to the scene, which can generate question entry semantic prototypes that meet the scene demand and enhance the pertinence of pre-diagnosis. The professional expressions are converted into language that parents can understand and supplemented with associated guiding sentences to form a preliminary pre-diagnosis form, improving the convenience and accuracy of parents filling out the form. By collecting multi-dimensional feedback data sets and inputting them into the model for analysis, the form fields and preliminary diagnosis suggestion drafts can be continuously optimized, and finally the final child tumor intelligent pre-diagnosis form containing the preliminary diagnosis direction is obtained, realizing the dynamic optimization of the pre-diagnosis form and the accurate generation of the diagnosis suggestion, and improving the efficiency, accuracy and intelligent level of the child tumor pre-diagnosis. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 is the execution flow diagram of the intelligent pre-diagnosis form generation method for child tumors provided by an embodiment of the present application.

[0016] Figure 2 is a schematic diagram of exemplary hardware and software components of the intelligent pre-diagnosis form generation system for child tumors provided by an embodiment of the present application. DETAILED DESCRIPTION

[0017] The present application will be described in detail below with reference to the accompanying drawings of the specification, Figure 1 is a flow diagram of the intelligent pre-diagnosis form generation method for child tumors provided by an embodiment of the present application, and the intelligent pre-diagnosis form generation method for child tumors will be described in detail below.

[0018] Step S110: Identify the scene type of child tumor pre-diagnosis, extract the scene demand features corresponding to each scene type, and form a scene demand feature set.

[0019] This embodiment takes the initial diagnosis pre-diagnosis scene of a three-year-old child suspected of having neuroblastoma as an example. In this scene, the parents take the child to the doctor for the first time, and need to collect information through the pre-diagnosis system to assist in preliminary diagnosis.

[0020] Step S111: Obtain the scene identifier selected by the parents through the scene selection portal of the parent interaction terminal, or identify the scene tendency based on the expression content initially input by the parents, and determine the current pre-diagnosis scene type.

[0021] The parent sees scene selection options on the home page of the interaction terminal, including "initial diagnosis pre-consultation", "follow-up visit", and "special symptom screening". The parent clicks on the "initial diagnosis pre-consultation" option, thereby determining that the current scene type is an initial diagnosis pre-consultation scene. If the parent does not directly select, but inputs "child first discovered a lump in the stomach" in the text box, the initial diagnosis scene tendency is identified through analysis of expressions such as "first discovered".

[0022] Step S112: If the scene type is an initial diagnosis pre-consultation scene, the demand characteristics of the initial diagnosis pre-consultation scene are extracted, including complete medical history collection demand, basic physiological index coverage demand, examination result preliminary verification demand, and preliminary diagnosis direction guidance demand. The core attention content corresponding to each demand characteristic is extracted and recorded.

[0023] After determining the initial diagnosis pre-consultation scene, for the complete medical history collection demand, the core attention content includes the gestational age of the child at birth, the mode of delivery, the birth weight, the height and weight growth during the growth process, the teething and language and motor development time, whether there has been a history of repeated infection, allergy, or surgery, and whether there are tumor or genetic disease patients in the family. The core attention content of the basic physiological index coverage demand includes vital signs such as body temperature, heart rate, respiratory rate, blood pressure, and blood oxygen saturation, as well as the current weight, height, and head circumference. The examination result preliminary verification demand focuses on the completeness of the examination report uploaded by the parent, whether there are abnormal indicators, and the qualification of the examination institution. The preliminary diagnosis direction guidance demand is to suggest possible tumor types based on existing information, suggest further examination items, and pay attention to using a statement that will not cause excessive panic to the parents. The above core attention content is recorded in detail.

[0024] Step S113: If the scene type is a follow-up visit scene, the demand characteristics of the follow-up visit scene are extracted, including treatment after symptom change record demand, recent examination result comparison demand, follow-up conclusion generation demand, and next follow-up time suggestion demand. The time range limitation corresponding to each demand characteristic is extracted and recorded.

[0025] When the scene is a follow-up visit scene, the time range limitation of the treatment after symptom change record demand is from the last visit to the current follow-up period, and the frequency, duration, and severity of the change of the symptoms need to be recorded. The time range of the recent examination result comparison demand is to compare the examination reports in the recent months with the previous baseline reports. The time range of the follow-up conclusion generation demand is the current follow-up day, and the stage evaluation is generated based on the current symptoms and examination results. The time range of the next follow-up time suggestion demand is determined according to the treatment stage, such as suggesting a follow-up in two weeks during the maintenance treatment period, or suggesting a follow-up in three months during the rehabilitation period. The time range limitation corresponding to each demand characteristic is recorded.

[0026] Step S114: If the scene type is a special symptom screening scene, extract the demand characteristics of the special symptom screening scene, including the target symptom detail description demand, the symptom associated part examination demand, the specific tumor index inquiry demand, and the screening conclusion prompt demand, extract the tumor type associated range corresponding to the target symptom and record.

[0027] For the special symptom screening scene, assume that the target symptom is unexplained fever with joint pain. The target symptom detail description demand records the onset time of fever, the type of fever, the highest body temperature, the way of defervescence and the effect, the part of joint pain, the nature, and the degree of activity limitation. The associated range of the symptom associated part examination demand includes bone marrow puncture, joint ultrasound, whole body PET-CT, and other examinations. The specific tumor index inquiry demand covers blood routine, inflammation markers, tumor markers, and other indicators. The associated range of the screening conclusion prompt demand is limited to tumor types highly related to fever and joint pain. By querying the medical knowledge base, the tumor type associated range corresponding to the target symptom is extracted and recorded.

[0028] Step S115: Perform a textual description of the demand characteristics of each scene type and mark the priority order of the demand characteristics.

[0029] The four demand characteristics of the initial diagnosis pre-consultation scene are described in words. The complete medical history collection demand is described as "collecting comprehensive health-related information of the child from birth to the present, providing basic data support for diagnosis"; the basic physiological indicator coverage demand is described as "obtaining the child's current vital signs and growth and development indicators to assess overall health status"; the examination result preliminary verification demand is described as "verifying the effectiveness of the examination report uploaded by the parents and the relevance of abnormal indicators to symptoms"; and the preliminary diagnosis direction guidance demand is described as "prompting possible diagnosis directions and next step examination suggestions without causing panic to the parents". According to the clinical diagnosis logic, the complete medical history collection demand and the basic physiological indicator coverage demand are marked as higher priority, the examination result preliminary verification demand is marked as medium priority, and the preliminary diagnosis direction guidance demand is marked as lower priority.

[0030] Step S116: Organize the demand characteristic descriptions of different scene types by scene type, demand characteristic, priority, and core content, remove repetitive demand characteristic descriptions between different scenes, and form a scene demand characteristic set.

[0031] According to the format of scene type, demand characteristic, priority, and core content, the demand characteristic descriptions of each scene type are organized. For example, in the initial diagnosis pre-consultation scene, the complete medical history collection demand corresponds to the corresponding priority and core content. By comparing the demand characteristic descriptions of different scenes, repetitive content is removed, such as "examination result verification" which is present in both initial diagnosis and special screening scenes. The feature description of the high priority scene is retained, and finally a scene demand characteristic set is formed.

[0032] Step S120: Obtain pre-diagnosis related multi-source input data, and perform association processing on the multi-source input data to obtain an association data set.

[0033] In the above initial diagnosis scenario, the multi-source input data includes a preliminary symptom description text input by the parent, such as "The child has had recurrent abdominal pain for nearly a month, a lump can be felt in the right lower abdomen, and weight loss", the child's past health records, such as full-term normal delivery without asphyxia history, once hospitalized due to bronchitis, and timely vaccination, and the parent uploaded external hospital abdominal ultrasound report and blood routine examination results.

[0034] Step S121: Collect the preliminary symptom description text input by the parent through the text input box of the parent interactive terminal, format the preliminary symptom description text, remove preset symbols and repetitive expressions, and retain the description of symptom occurrence time, manifestation form, duration, and relieving or aggravating factors.

[0035] The parent inputs "The child has had abdominal pain for more than a month, right of the navel, cries when pressed, recently dislikes eating, loses weight, and cannot sleep at night, and feels better in the morning?" in the text input box. After formatting the text, removing preset symbols such as "!" and "?", and removing repetitive expressions such as "abdominal pain" and "abdominal pain on the right of the navel", the latter is retained. The symptom occurrence time is more than a month, the manifestation form is right lower abdominal pressing pain, decreased appetite, weight loss, the duration is persistent and aggravated at night, the relieving factor is morning relief, and the aggravating factor is pressing.

[0036] Step S122: Obtain the child's past health records through the history record import interface of the interactive terminal, sort the child's past health records in chronological order, and extract key health events in each child's past health record, which includes past medical records, treatment plan records, and past examination reports.

[0037] The parent authorizes to obtain the child's past health records through the history record import interface of the interactive terminal. The obtained records include hospitalization in a community hospital due to acute bronchitis at a certain time, treatment plan of anti-infection and aerosol inhalation; physical examination in the child care department at a certain time, normal height and weight, and no abnormalities in blood routine and urine routine; emergency treatment for febrile convulsion at a certain time, normal electroencephalogram. The above records are sorted in chronological order, and key health events such as "certain time acute bronchitis hospitalization treatment anti-infection + aerosol" are extracted.

[0038] Step S123: Receive the examination result sheet uploaded by the parent through the file upload entrance of the interactive terminal, analyze the text content and numerical data of each examination result sheet, and extract index information related to the child's tumor diagnosis, which includes image examination report, laboratory test report, and pathological examination report.

[0039] In the initial diagnosis scenario of suspected neuroblastoma, the parents uploaded three test result sheets: an abdominal ultrasound report from an external hospital (PDF format), a blood test report (JPG format), and a pathology consultation report (TXT format).

[0040] For example, step S1231: receiving the test result sheets uploaded by the parents through the file upload portal of the parent interaction terminal, identifying the format of the uploaded test result sheets, distinguishing between image examination reports, laboratory test reports, and pathology examination reports, and supporting test result sheet formats including image formats, portable document formats, and text formats.

[0041] After the file upload module receives the file, it first reads the file binary header information: the abdominal ultrasound report is identified as a portable document format with the "%PDF-" mark; the blood test report is determined to be an image format through the "FFD8FF" file header; and the pathology consultation report is determined to be a text format through ASCII character stream detection. Then, a medical document classification model is called, and the file names "ultrasound", "blood test", and "pathology" are combined with keywords to classify the three reports as image examination reports, laboratory test reports, and pathology examination reports, respectively. The classification results are stored in the file metadata for subsequent parsing process routing.

[0042] Step S1232: If the test result sheet is an image examination report, extract the text content of the image examination report using optical character recognition technology, locate the examination site, image performance, and diagnosis opinion fields in the text, extract the specific organ or tissue name corresponding to the examination site, extract the lesion description content in the image performance, and extract the preliminary judgment result in the diagnosis opinion.

[0043] For the abdominal ultrasound report (image examination report), the parsing engine calls the PDF text extraction interface to obtain the text layer and scans the page using optical character recognition technology (medical special-purpose OCR model). The BERT named entity recognition model is used to locate the fields: the specific organ name "adrenal gland area (right)" is extracted from "examination site: whole abdomen (focus on adrenal gland area)"; the lesion description (location, echo characteristics, boundary, internal structure, and blood flow signal) is extracted from "image performance: a low echo mass is seen in the right adrenal gland area, with unclear boundary and uneven internal echo, and visible rich blood flow signal"; and the preliminary judgment result "neuroblastoma possible" is extracted from "diagnosis opinion: consider neuroblastoma possible, suggest enhanced CT". The extracted information is stored in the "examination type-site-feature-conclusion" structure.

[0044] Step S1233: If the check result sheet is a laboratory test report, extract the test item name, test result value, reference range, and unit field in the laboratory test report, and screen out the test items related to the diagnosis of childhood tumors; for the screened test items, record the test result value, reference range, and unit of each test item, mark the test items with test result values exceeding the reference range as abnormal indicators, and record the abnormal direction.

[0045] After the blood routine test sheet (laboratory test report) is corrected in picture format, the four columns of data of "test item, result, unit, reference range" are extracted by table detection algorithm. The relevant items are screened out by calling the childhood tumor core index library (containing 58 indexes such as NSE and LDH): hemoglobin (result, unit, reference range), platelet count (normal), white blood cell count (normal), lactate dehydrogenase (higher than the upper limit of the reference range). The numerical comparison algorithm determines that hemoglobin is reduced and lactate dehydrogenase is increased, which are marked as abnormal indicators and the abnormal direction is recorded. The normal indicators are marked as "normal", and the results are stored in the structure of "test item-result-unit-reference range-abnormal marker".

[0046] Step S1234: If the check result sheet is a pathological examination report, extract the specimen type, pathological diagnosis, and immunohistochemical result fields in the pathological examination report, record the tissue source corresponding to the specimen type, extract the conclusive description in the pathological diagnosis, and extract the positive and negative indicators in the immunohistochemical result.

[0047] The pathological consultation report (pathological examination report) extracts the tissue source "adrenal gland area (right)" by regular expression positioning field: "specimen type: right adrenal gland area puncture biopsy tissue"; "pathological diagnosis: small round blue cell tumor, combined with immunohistochemistry, consistent with neuroblastoma" extracts the conclusive description "neuroblastoma"; "immunohistochemistry: Syn (+), CgA (+), NSE (+), Ki-67 (70% +), LCA (-), CK (-)" is split by semicolon, identifies "(+)" as positive indicators (Syn, CgA, NSE, Ki-67), and "(-)" as negative indicators (LCA, CK), and Ki-67 synchronously records the positive percentage. Extract the index and verify the consistency with the neuroblastoma marker library.

[0048] Step S1235: The check site, lesion description, and diagnosis opinion extracted from the image examination report, the tumor-related test items, values, and abnormal markers extracted from the laboratory test report, and the specimen type, pathological diagnosis, and immunohistochemical results extracted from the pathological examination report are arranged according to the report type, extraction field, and specific content format.

[0049] The integrated module organizes data in a three-level structure of "report type-extraction field-specific content": the image examination report corresponds to "examination site: adrenal region (right); lesion description: hypoechoic mass, unclear boundary, rich blood flow; diagnostic opinion: possible neuroblastoma"; the laboratory test report corresponds to "tumor-related items: lactate dehydrogenase (elevated), hemoglobin (decreased)"; the pathology examination report corresponds to "specimen type: adrenal region puncture tissue; pathological diagnosis: neuroblastoma; immunohistochemistry: Syn (+), CgA (+), NSE (+), Ki-67 (70% +)". The data is stored in structured JSON format, with each report type as the top key and the extraction field as the secondary key.

[0050] Step S1236: Compare the organized information with the common indicators of childhood tumors, supplement the tumor-related explanations of the indicators, and form single-indicator data of the examination results.

[0051] Call the common indicators of childhood tumors, compare the organized indicators with the entries in the library: lactate dehydrogenase elevation is associated with "neuroblastoma, lymphoma, and other solid tumors common abnormalities"; immunohistochemistry Syn (+), CgA (+) is associated with "neuroendocrine tumor markers positive, supporting the diagnosis of neuroblastoma"; Ki-67 (70% +) is associated with "high cell proliferation activity, indicating high malignancy of the tumor". After supplementing the associated explanations, the single-indicator data of the examination results contains the original extraction information and tumor-related explanations, and is stored completely in the database for subsequent correlation analysis.

[0052] Step S124: Associate the symptom keywords in the preliminary symptom description text with the key health events in the child's past health records, identify the relevance of the current symptoms to the past health events, and extract and record the association results.

[0053] Extract the keywords "abdominal pain", "abdominal mass", and "weight loss" from the preliminary symptom description and map them to UMLS standard terms. The child's past health events include "hospitalization for bronchitis at 6 months", "febrile convulsions at 1 year", and "normal periodic physical examination", and the cosine similarity calculated by the symptom-disease association strength algorithm is below the threshold value (0.6), indicating no direct association. Time series analysis shows that the recent physical examination (3 months ago) had no abdominal abnormalities, indicating a new onset of symptoms, and the association result is recorded as "no significant association between current symptoms and past health events, recent new onset of abnormalities" and stored in the association analysis result table.

[0054] Step S125: Associate the symptom sites in the preliminary symptom description text with the indicator information in the single-indicator data of the examination results, match the corresponding examination indicators of the symptom sites, extract and record the associated indicator names and values.

[0055] The human anatomy part ontology library constructs a mapping relationship. "Right lower abdomen" is matched with "right adrenal region" through a spatial topology algorithm, and the coincidence degree is 0.82 (higher than the threshold value 0.7), and the associated part is determined. The corresponding indicators of the right adrenal region are extracted: image examination "low echo mass (5 cm x 4 cm)" "rich blood flow signal", laboratory examination "lactate dehydrogenase elevation" "hemoglobin reduction", pathological examination "neuroblastoma" diagnosis. The association result is stored according to "symptom site-examination site-matching degree-index list", and the index list contains name, value, unit and abnormal mark.

[0056] Step S126: Associate the key health events in the child's past health record with the recent indicators in the examination result sheet, analyze the trend of the indicators after treatment, extract the trend analysis result and record.

[0057] Extract the laboratory data in the past 1 year, and focus on comparing the same indicators: lactate dehydrogenase is normal in the physical examination 3 months ago, and is increased by more than 50% currently, the growth rate is calculated in combination with no special treatment history, and it is determined to be related to space-occupying lesions; hemoglobin decreased from the lower limit of normal to mild decrease, combined with weight loss, suggesting chronic consumption. The trend analysis result is recorded as "lactate dehydrogenase significantly increased (compared with baseline), hemoglobin slightly decreased, associated with right adrenal region space-occupying lesions", and the trend curve data is stored.

[0058] Step S127: Integrate the association results of symptoms and past records, the association results of symptoms and examination indicators, and the trend analysis results of past records and recent indicators, while adding the original preliminary symptom description text, the sorted child's past health record and the analyzed examination result sheet indicator data, to form an association data set.

[0059] The association data set adopts a nested JSON structure: the top layer contains "raw data" and "association analysis" modules; the "raw data" stores the formatted symptom description, the past health records in reverse chronological order, and the analyzed examination indicators; the "association analysis" contains symptom-past record, symptom site-examination indicator, and index trend analysis result. The data is associated with the pre-consultation session ID through UUID, and after integrity verification, the missing metadata is supplemented and stored in a distributed database.

[0060] Step S130: Adjust the parameter configuration of the child tumor large model based on the scene demand feature set to obtain an adjusted child tumor large model adapted to the scene.

[0061] According to the scene demand feature set of the initial diagnosis pre-consultation scene, the parameters of the child tumor large model are adjusted to adapt to the scene.

[0062] Step S131: Extract the demand features corresponding to the current scene type from the scene demand feature set, and determine the adjustment direction of each demand feature to the model parameters.

[0063] From the scene demand feature set, the complete medical history collection demand, the basic physiological index coverage demand, the examination result preliminary verification demand and the preliminary diagnosis direction guidance demand of the initial diagnosis pre-consultation scene are extracted. According to the corresponding relationship between the demand and the model parameter, the field integrity parameter in the complete medical history collection demand corresponding adjustment case report form structure generation parameter is determined, the direction is maximization coverage; the field type parameter corresponding to the basic physiological index coverage demand is adjusted, the direction is to contain the vital signs and growth index field; the examination result association weight corresponding to the examination result preliminary verification demand is adjusted, the direction is to improve the image examination and symptom association weight; the preliminary diagnosis suggestion generation threshold corresponding to the preliminary diagnosis direction guidance demand is adjusted, the direction is to reduce the threshold to allow the preliminary direction to be generated based on part of the data.

[0064] Step S132: If the current scene is the initial diagnosis pre-consultation scene, for the case report form structure generation parameter, set the complete case report form field generation mode, set the form to contain basic information field, medical history field, symptom field, examination result field and preliminary diagnosis prompt field, and adjust the field generation order parameter to arrange the fields according to the preset logic.

[0065] In the initial diagnosis pre-consultation scene, the case report form structure generation parameter is set to the complete mode, the form contains basic information fields such as child's name, gender, birth date, etc.; medical history fields such as birth history, growth and development history, etc.; symptom fields such as current main symptoms, accompanying symptoms, etc.; examination result fields such as image examination, laboratory test results, etc.; preliminary diagnosis prompt fields such as possible diagnosis direction, recommended examination items, etc. The field generation order is adjusted to be basic information, medical history, symptom, examination result, preliminary diagnosis prompt, so that the fields are arranged according to this preset logic.

[0066] Step S133: For the examination result association weight parameter, if the scene demand feature contains the examination result preliminary verification demand, the weight value of the examination result index associated with the inquiry item is adjusted to a value that meets the examination result preliminary verification demand, so that the generated inquiry item is preferentially associated with the uploaded examination result single index.

[0067] Since the initial diagnosis scene demand feature contains the examination result preliminary verification demand, the weight value of the image examination index associated with the inquiry item is increased, so that the generated inquiry item can preferentially associate the uploaded image examination index such as abdominal ultrasound report, which facilitates the preliminary verification of the examination result.

[0068] Step S134: Generate threshold parameters for preliminary diagnosis suggestions, set diagnosis suggestion generation thresholds that meet the preliminary diagnosis direction guidance requirements according to the preliminary diagnosis direction guidance requirements in the scene demand, so that the preliminary diagnosis direction prompt can be generated after obtaining the basic symptoms and preliminary examination results, and at the same time, limit the expression range of the diagnosis suggestions to avoid absolute expression.

[0069] According to the preliminary diagnosis direction guidance requirements, the preliminary diagnosis suggestion generation threshold is set to a lower level, so that the preliminary diagnosis direction prompt can be generated after obtaining the basic symptoms and preliminary examination results. At the same time, the expression of the diagnosis suggestions is limited to avoid using absolute words such as "definite diagnosis" and instead using expressions such as "may be considered" and "not ruled out".

[0070] Step S135: If the current scene is a follow-up visit scene, adjust the case report form structure generation parameters to a follow-up visit dedicated mode, set the form to retain the past treatment record field and the current symptom change field, add the treatment effect evaluation field and the next follow-up plan field, and adjust the field generation order parameters to arrange the fields according to the preset logic.

[0071] When the scene is a follow-up visit scene, the case report form structure generation parameters are adjusted to a follow-up visit dedicated mode, the form retains the past treatment record field, records the previous treatment plan, drug dosage, etc., and the current symptom change field records the change in symptoms compared with the last follow-up. The treatment effect evaluation field is added to evaluate the effect after treatment, and the next follow-up plan field determines the time and examination items of the next follow-up. The field generation order is adjusted to past treatment record, current symptom change, treatment effect evaluation, and next follow-up plan.

[0072] Step S136: Adjust the examination result association weight parameter to adjust the comparison weight between the recent examination result sheet and the past examination result sheet to a value that meets the examination result comparison requirements, so that the generated inquiry items contain examination result change reason inquiry content; adjust the preliminary diagnosis suggestion generation threshold parameter to a value that meets the complete data requirements, and only generate the follow-up conclusion suggestion after obtaining the treatment after symptom and examination result sheet comparison data.

[0073] In the follow-up visit scene, the weight of the comparison between the recent examination result sheet and the past examination result sheet is increased, so that the generated inquiry items contain inquiries about the reasons for the change in examination results. The preliminary diagnosis suggestion generation threshold is adjusted to a higher value, and only after obtaining the treatment after symptom and examination result sheet comparison data, the follow-up conclusion suggestion is generated.

[0074] Step S137: Input the adjusted case report form structure generation parameters, examination result association weight, and preliminary diagnosis suggestion generation threshold into the parameter adjustment interface of the pediatric tumor large model.

[0075] Input the adjusted case report form structure generation parameters, examination result association weights, and preliminary diagnosis suggestion generation threshold into the parameter adjustment interface of the pediatric tumor large model to complete the configuration of the model parameters.

[0076] Step S138: Call the model adaptability test program, input test data matching the current scenario, and obtain the output test inquiry items and diagnosis suggestion drafts.

[0077] Call the model adaptability test program, input test data matching the initial diagnosis scenario into the adjusted pediatric tumor large model, and obtain the output test inquiry items and diagnosis suggestion drafts.

[0078] Step S139: Compare the fit degree of the test output results and the scenario requirement characteristics, check whether the case report form fields cover the scenario requirements, whether the examination result single association meets the weight setting, and whether the diagnosis suggestion meets the threshold requirements.

[0079] Compare the fit degree of the test inquiry items and diagnosis suggestion drafts obtained by the test and the initial diagnosis scenario requirement characteristics, check whether the case report form fields cover complete medical history collection, basic physiological index coverage, and other requirements, whether the examination result single association meets the set weight, and whether the diagnosis suggestion meets the generation threshold requirements.

[0080] Step S1310: If the fit degree does not meet the expectation, adjust the corresponding parameters again and test again until the test output results completely fit the scenario requirement characteristics, and obtain the adjusted pediatric tumor large model adapted to the scenario.

[0081] If the fit degree of the test output results and the scenario requirement characteristics does not meet the expectation, analyze the reasons for the mismatch, adjust the corresponding model parameters such as the case report form field integrity parameters and the examination result association weights, and then test again. Repeat this process until the test output results completely meet the scenario requirement characteristics, and obtain the adjusted pediatric tumor large model adapted to the initial diagnosis scenario.

[0082] Step S140: Input the associated data set into the adjusted pediatric tumor large model to generate inquiry item semantic prototypes containing case report form standard fields, wherein the inquiry item semantic prototypes are associated with examination result verification items and preliminary diagnosis direction prompts.

[0083] Input the associated data set obtained above into the adapted pediatric tumor large model. The model generates inquiry item semantic prototypes containing case report form standard fields based on the information in the associated data set. The semantic prototypes are associated with examination result verification items, such as verification of the size of the abdominal ultrasound report, and preliminary diagnosis direction prompts, such as a suggestion that it may be neuroblastoma.

[0084] Step S141: Extract the symptom keyword set in the preliminary symptom description text, the sorted key events in the child's health record, the parsed single indicator data of the test results, and the association results between each data from the association data set, and classify and organize them into input data groups according to data types.

[0085] From the association data set, the symptom keyword set such as "abdominal pain" and "weight loss", the sorted key events in the child's health record such as "acute bronchitis hospitalization" and "febrile convulsion emergency", the parsed single indicator data of the test results such as the size of the abdominal ultrasound and the value of lactate dehydrogenase in blood routine, and the association results between each data are classified and organized into input data groups according to data types, so as to input the model for processing.

[0086] Step S142: Input the input data group into the case report form field mapping layer of the adjusted child tumor large model, generate parameters based on the case report form structure, match the contents in the input data group with the standard case report form fields, and determine the type of case report form fields to be generated.

[0087] Input the input data group into the case report form field mapping layer of the model, match the contents in the input data group with the standard case report form fields according to the previously set case report form structure generation parameters, and determine the type of fields to be generated, such as basic information fields, medical history fields, etc.

[0088] Step S143: For each determined case report form field type, generate a basic expression framework, which includes a field name, an inquiry content guide, and an input format prompt.

[0089] For each determined case report form field type, a basic expression framework is generated. For example, the basic expression framework of the basic information field includes the field name "child's name", the inquiry content guide "please enter the child's name", and the input format prompt "please enter the Chinese name".

[0090] Step S144: Based on the test result association weight, identify the abnormal items of the single indicator data of the test results in the input data group, generate corresponding test result verification items for the abnormal indicators, and add them to the basic expression framework of the corresponding case report form field.

[0091] According to the test result association weight, identify the abnormal items of the single indicator data of the test results in the input data group, such as the abnormal indicator of elevated lactate dehydrogenase in blood routine, generate the test result verification item "the blood routine report you uploaded shows that lactate dehydrogenase is higher than the normal range, does the child have other discomforts?", and add the verification item to the corresponding case report form field basic expression framework.

[0092] Step S145: Based on the preliminary diagnosis suggestion generation threshold, analyze the association between the symptom keywords, the child's past health records and the single indicators of the test results in the input data set. If the association result reaches the diagnosis suggestion generation threshold, generate a preliminary diagnosis direction prompt, and add the prompt content to the corresponding case report table field according to the preset format.

[0093] According to the preliminary diagnosis suggestion generation threshold, analyze the association between the symptom keywords "abdominal pain" and "weight loss", the child's past health records and the single indicators of the test results such as right adrenal gland area occupation and lactate dehydrogenase elevation. When the association result reaches the set threshold, generate a preliminary diagnosis direction prompt, such as "considering the possibility of neuroblastoma in combination with symptoms and test results", and add it to the corresponding case report table field according to the preset format.

[0094] Step S146: Perform semantic integration on the generated content containing the case report table field basic expression framework, test result verification items and preliminary diagnosis direction prompts, remove expressions unrelated to the current scene requirements in the integrated content, arrange the integrated content according to the order of the case report table fields, each field corresponds to a complete expression containing basic inquiries, verification items and prompts, and form a semantic prototype of the inquiry item.

[0095] Perform semantic integration on the generated content containing the case report table field basic expression framework, test result verification items and preliminary diagnosis direction prompts, so that the expression is more coherent. Remove expressions unrelated to the initial diagnosis scene requirements, such as treatment effect evaluation content related to re-examination. Arrange the integrated content according to the order of the case report table fields, each field forms a complete expression containing basic inquiries, verification items and prompts, thereby forming a semantic prototype of the inquiry item.

[0096] Step S150: According to the parent expression habit sample, optimize the expression logic of the inquiry item semantic prototype, convert the professional case report table field into a parent understandable expression, supplement the association guide sentences of symptoms and test results, form a preliminary pre-inquiry form, push the preliminary pre-inquiry form to the parent interactive terminal, collect the form content filled by the parents, supplement the symptom description and the supplementary explanation of the test results, at the same time record the form filling time, field modification times and skipping behavior of the parents, and integrate to form a multi-dimensional feedback data set.

[0097] For example, in the initial diagnosis scenario of suspected neuroblastoma, after completing the optimization of the semantic prototype of the interview item, a preliminary pre-interview form containing 23 items is formed, which is pushed to the parents through the form display interface of the parent interaction terminal. The form uses a paging loading mode, displaying 5 items per page, with "previous / next" navigation buttons and "save draft" and "submit" function keys at the bottom. The parents start filling in from the "basic information" page, and then complete the content filling of the "medical history collection", "symptom description", "test result confirmation", and "initial diagnosis guidance" modules in turn. The system records the filling behavior data in real time, and finally integrates it into a multi-dimensional feedback data set containing form content, supplementary notes, and interaction behavior.

[0098] Step S151: Collect samples of expression habits of parents from different backgrounds, divide them into symptom description sample group, terminology understanding sample group, and examination result feedback sample group according to expression scenarios. The expression habit samples include parents' expressions of children's symptoms, understanding of medical terminology, and feedback of examination results.

[0099] For example, expression habit samples are collected through multi-center clinical research. The inclusion criteria are parents who have had children with tumors in the past 3 years. Stratified sampling is performed according to region (east / middle / west), education level (primary school to undergraduate and above), and tumor type of the child (solid tumor / hematoma). A total of 1200 valid samples were collected. The symptom description sample group contains 400 samples of parents' natural language descriptions of symptoms such as "abdominal pain", "fever", and "weight loss", such as "the child's stomach hurts like a needle, like a needle" and "the fever always starts in the afternoon, and taking fever-reducing medicine can reduce it a little". The terminology understanding sample group contains 400 samples of parents' understanding of professional terms such as "neuroblastoma" and "lactate dehydrogenase", such as "I thought 'occupancy' was a cyst" and "I don't understand what 'LDH' means". The examination result feedback sample group contains 400 samples of parents' interpretation of ultrasound and CT reports, such as "B ultrasound said there was a 5 cm thing, the doctor said it might be a tumor" and "the blood test sheet with the arrow pointing upwards is not normal". After all samples are de-identified, they are stored in the corpus in the format "sample ID-scenario type-original text-labeled terminology".

[0100] Step S152: Analyze the expression text of the symptom description sample group to count the common colloquial words used by parents when describing symptoms, and form a colloquial word list for symptoms; analyze the expression text of the terminology understanding sample group to extract professional case report table field names that parents have difficulty understanding and the corresponding colloquial conversion methods, and form a colloquial conversion table for case report table fields; analyze the expression text of the examination result feedback sample group to summarize the common expression logic used by parents when feeding back examination results, and form an examination result guided expression template.

[0101] The symptom description sample group analysis uses the term frequency-inverse document frequency (TF-IDF) algorithm to obtain the top 50 popular words such as "stomach pain" (corresponding to "abdominal pain"), "fever" (corresponding to "fever"), and "weight loss" (corresponding to "weight loss"), and to construct a symptom popular vocabulary table, including the fields of "professional term-popular term-frequency-geographical difference", such as "vomiting-vomiting-89.2%-northern use of 'dry' accounts for 12%". The term understanding sample group identifies difficult terms by calculating the perplexity, and for 32 professional fields such as "neuron-specific enolase" and "mesenchymal tumor", a three-level conversion structure of "full name-short name-popular explanation" is used to form a case report table field popular conversion table, such as "lactate dehydrogenase-LDH-a kind of enzyme in the body that helps cell energy metabolism, elevated may indicate cell damage". The examination result feedback sample group uses a sequence pattern mining algorithm to summarize the typical expression logic of "examination type+abnormal performance+parental concern", such as "did the B-ultrasound, said there was something on the kidney, we were very afraid that it was cancer", and accordingly designs an examination result guiding expression template containing three elements of "examination item confirmation+key indicator interpretation+emotion guidance", divided into three sub-templates of "imaging examination", "laboratory test", and "pathological examination".

[0102] Step S153: Extract the professional case report table field name in the semantic prototype of the inquiry item, and replace it with a popular expression by referring to the case report table field popular conversion table; extract the symptom-related expression in the semantic prototype of the inquiry item, and replace the professional symptom term with a popular vocabulary by referring to the symptom popular vocabulary table.

[0103] From the semantic prototype of the inquiry item, 38 professional field names and 27 symptom terms are extracted and converted into popular expressions. The case report table fields such as "family history of tumors" are replaced with "is there anyone in the family who has had cancer", and "previous history of surgery" is replaced with "has there been any previous open surgery"; the symptom terms such as "persistent dull pain" are replaced with "always aching, not very sharp pain", and "intermittent fever" is replaced with "fever in bursts, sometimes fever and sometimes not fever". During the replacement process, the bidirectional maximum matching algorithm is used to ensure the accuracy of term positioning, and long sentences containing multiple professional terms are segmented and converted, such as "the child has unexplained weight loss and loss of appetite" is converted to "the child has unexplained weight loss and loss of appetite". The term mapping relationship before and after conversion is stored in a version control table to support backtracking queries.

[0104] Step S154: Supplement the associated guiding sentences before and after the expression of the examination result verification item in the semantic prototype of the inquiry item, and refer to the examination result guiding expression template.

[0105] For the examination result verification item, supplementary related content is added in the current item front guide and back follow-up question. For example, for the "ultrasound examination result confirmation" item, the front guide is "The abdominal ultrasound report you uploaded shows that there is an abnormality in the right adrenal gland area (in simple terms: a growth on the kidney), please confirm the following information", and the back follow-up question is "If you have any questions about the report content, you can write down the questions you want to know here, and the doctor will answer them in the follow-up consultation". Referring to the examination result guide statement template, the image examination item focuses on supplementing the "examination site simple positioning" guide, such as "The 'right adrenal gland area' in the report is approximately located on the right side of the child's belly and slightly above the waist"; the laboratory test item focuses on supplementing the "index abnormality significance" guide, such as "The 'lactate dehydrogenase elevation' may indicate cell damage, which needs to be combined with other tests"; the pathological examination item focuses on supplementing the "diagnostic relevance" guide, such as "The'small round blue cell tumor' in the pathology report is a type of tumor that needs further diagnosis". All guide sentences are controlled within 20 characters, and are distinguished from the main content of the item by using bold font.

[0106] Step S155: The logical coherence of the optimized expression content is checked, the field logical order verification tool is used to compare the field order after the popular expression with the standard logic of the case report form, and the conflict content between the guide sentences and the verification items and diagnostic prompts is checked and corrected by the expression conflict detection module.

[0107] The field logical order verification tool is called, the standardized case report form logic template developed by the International Pediatric Oncology Cooperation Group (SIOP) is loaded, and the optimized form field order is compared with it. For example, the standard logic requires "birth history" before "growth and development history", and the tool detects that the current form places "growth and development history" first, and automatically marks it as "logical order abnormality" and suggests adjustment; for the order of "symptom duration" and "symptom relief factors", it is determined as "logical consistency" because it conforms to the "time-feature" description logic. The expression conflict detection module uses rule engine + deep learning dual detection: the rule engine presets 12 conflict rules such as "guide sentences must not contain diagnostic conclusions" and "verification items must be strongly associated with examination results", and if it finds an absolute expression such as "this result may be cancer", it is automatically replaced with "this result needs further examination to determine its nature"; the deep learning model based on the BERT conflict detection model identifies the implicit conflict between "the child is thin, which must be malnutrition" in the guide sentence of the "weight loss" item and "tumor may cause weight loss" in the follow-up, and suggests modifying it to "the child is thin, which may have multiple reasons, let's understand them one by one".

[0108] Step S156: Obtain the understanding degree of the parents of different backgrounds reading the optimized expression content and the questions raised, adjust the expression method again for the question points, arrange all the items in the order of the optimized case report form fields, each item contains the basic inquiry of popular expression, related guide sentence, examination result verification item and preliminary diagnosis direction prompt, and form a preliminary pre-diagnosis form.

[0109] Select 15 parents of different backgrounds (region, education level, and tumor type, each 3 layers, 5 people in each layer) to conduct cognitive tests, use "out-loud thinking method" to record the reading process: let the parents read the content of the items sentence by sentence and say the understanding meaning, and record the length of pause (> 3 seconds is regarded as understanding obstacle) and question points. Test found that "right adrenal space occupying" even converted to "something grows in the kidney", still 6 parents misunderstood as "something grows in the kidney", further adjusted to "something grows in the position of the right side of the belly and the upper part of the waist (called space occupying in medicine)"; "Ki-67 index 70%" is converted to "cell proliferation activity is higher", 8 parents cannot understand, change to "the higher this index, the faster the bad cells grow inside". According to the feedback, 12 expressions are modified, and the final preliminary pre-diagnosis form contains 4 modules and 23 items, each item consists of "basic inquiry (popular expression) + related guide (blue word) + verification item (editable text box / single selection button) + diagnosis prompt (gray word note)", such as item S15-08: "Does the child have the following conditions when the belly hurts? (Multiple selection) [Basic inquiry] -> The ultrasound report you uploaded shows that the mass is in the right abdomen, this position usually has these performances when it hurts [Related guide] -> □ Hurts directly sweating □ Doesn't want to walk □ Vomit □ Other (please fill in) [Verification item] -> These performances are helpful for doctors to judge whether the mass compresses the surrounding tissue [Diagnosis prompt]”.

[0110] Step S157: Load the items of the preliminary pre-diagnosis form to the form display interface of the parent interactive terminal according to the optimized order, set independent filling area for each item, and the filling area contains text input box, option selection button and supplementary explanation upload entrance.

[0111] The form display interface adopts responsive design, which is suitable for mobile phones (portrait / landscape), tablet devices, and the entry filling area is arranged in three rows of "question description area - input control area - auxiliary explanation area". The text input box is set to a single line input (maxlength=20) according to the expected input length, and the "detailed symptom description" is set to a multi-line text box (rows=4, maxlength=500), and supports voice input (long press the microphone button to speak, and automatically convert the text). The option selection button is divided into single selection (round button) and multiple selection (square checkbox), such as "abdominal pain nature" for single selection group (□ persistent □ intermittent □ paroxysmal), and "companion symptoms" for multiple selection group (□ fever □ vomiting □ fatigue). The supplementary explanation upload entry is set to a "+ upload picture / recording" icon button, which pops up a file selection menu after clicking, supporting the upload of JPG / PNG format pictures (≤10MB) or MP3 format recordings (≤60 seconds), and the upload progress is displayed in real time. Percentage, and after completion, a thumbnail / audio waveform preview is generated. All input controls are set with mandatory item identifiers (*), and the submit button is grayed out and prompts "please complete the red mandatory items" when not filled in.

[0112] Step S158: When the parent clicks the start filling button of the form display interface, the form filling record program of the interactive terminal is started, and the total start time of the form filling is recorded.

[0113] After the parent clicks the "start filling" button (blue gradient background, white text, located at the top center of the form home page), the front-end JavaScript triggers the startRecording() function, which obtains the current timestamp (accurate to milliseconds) through Date.now(), and stores it in localStorage as the total start time, and sends a WebSocket message to the back-end, records "user ID-form ID-start time" to the filling behavior log table. The form filling record program adopts a double-thread design: the main thread handles UI interaction (input / selection / upload), and the secondary thread handles page switching, input box focus / defocus, etc. through requestAnimationFrame(), to ensure that time recording is not affected by UI blocking. If the parent closes the page halfway, the last filling progress is automatically restored when opening next time, and the total start time is still the timestamp of the first click on the "start filling" button, avoiding repeated timing.

[0114] Step S159: For each entry, when the parent enters the filling area of the entry, the filling start time of the entry is recorded, and when the parent clicks the next step or skip button, the filling end time of the entry is recorded, and the difference between the filling end time and the filling start time is calculated as the filling duration of the entry.

[0115] Each entry sets an independent fill-in timer. When the parent enters the visible area of the entry by clicking or scrolling (i.e., the top of the entry is within 200px of the top of the viewport), the entryFocus() event is triggered, recording the start time of the fill-in (accurate to the millisecond). When the parent clicks the "Next" button or the "Skip" button, the entryBlur() event is triggered, recording the end time of the fill-in. Fill-in duration = end time - start time. If the fill-in duration < 3 seconds (a quick skim), the system automatically marks it as "Possible not read carefully"; if the fill-in duration > 300 seconds (5 minutes), it is marked as "Difficult to fill in". For example, the fill-in start time of entry S15-03 "What was the child's birth weight?" is 1692512345678, and the end time is 1692512346890, with a fill-in duration of 1212 milliseconds (about 1.2 seconds), marked as "Quick fill-in"; the fill-in start time of entry S15-17 "What questions do you have about the initial diagnosis of 'neuroblastoma possible'?" is 1692512456789, and the end time is 1692512678901, with a fill-in duration of 222112 milliseconds (about 3.7 minutes), marked as "Difficult to fill in". All time data is stored in the format "entry ID - start time - end time - duration - mark".

[0116] Step S1510: When the parent modifies the input content in the fill-in area and submits it, record the number of modification operations, save the content before and after each modification, and form a modification record for the entry.

[0117] The input control binds the input event to listen to the modification behavior, and the text input box uses a debounce process (trigger after 300ms without input), and the option button click triggers it. Each modification, the system generates a modification record object: {"modification ID": uuid, "entry ID": "S15-05", "content before modification": "38.5℃", "content after modification": "39.2℃", "modification time": 1692512345678, "modification type": "value adjustment"}. For text input boxes, a difference comparison algorithm based on Levenshtein distance is used to record the added, deleted, and modified content, such as "originally wrote'started hurting yesterday' and later changed to'started hurting 3 days ago'"; for option buttons, record the selected state changes, such as "from 'fever' unselected to selected". The number of modifications is cumulatively calculated, and multiple undo / redo of the same content is considered as one modification. All modification records are stored locally through IndexedDB and uploaded to the server when the form is submitted, for analysis of which content the parent has uncertainty about.

[0118] Step S1511: When the parent clicks the skip button, record the skip behavior of the entry, and at the same time, pop up a supplementary explanation window to prompt the parent to select the skip reason, which includes information cognitive deficiency, content irrelevance, expression understanding disorder, allows the parent to input supplementary explanation, records the skip reason and supplementary explanation content.

[0119] The "skip" button (gray text, hidden by default, displayed when mouse hovers / touches) is set on the right side of the entry. After clicking, the skipEntry() function is triggered: first, record "entry ID-skip time-current progress" in the behavior log; Then pop up a modal window (semi-transparent black background, white foreground frame, displayed in the center), the window title is "Why skip this question?", The content area contains three single selection options: "A. Can't understand what the question means (expression understanding disorder)" "B. Don't know how to answer (information cognitive deficiency)" "C. This question is not related to my child's situation (content irrelevance)", Each option is followed by a "please add explanation" text box (optional). After the parent selects and clicks the "OK" button to close the window, the system records "skip reason code (A / B / C)-supplementary explanation text-window dwell time", such as "B-children have not done surgery, don't know how to fill in-15 seconds". If the parent closes the window without selecting the reason, it is marked as "D. Other reasons" by default, and the supplementary explanation is empty. The skipped entry is displayed as a "skipped" gray label on the form page, allowing the parent to return to fill it in again.

[0120] Step S1512: When the parent completes all entry filling or clicks the submit button, collect the form content input by the parent in each entry filling area, including text input content and option selection result.

[0121] After the parent clicks the "Submit" button (green background, located at the bottom of the last page of the form), the front-end performs form validation: checking whether all required fields are filled (except for skipped items), whether the text input box content meets the format requirements (e.g., a mobile phone number is 11 digits), and whether the uploaded file is complete. After successful validation, all entered content is collected through a FormData object: text input box content is stored as "item ID-value" key-value pairs, such as "height=105cm"; option selection results are stored as "item ID-selected value array", such as "painCharacteristics=['paroxysmal','colic']"; unfilled optional fields are marked as "null", and skipped items are marked as "skipped". The form content is serialized in JSON format and sent to the back-end via an HTTPS POST request. The request header includes "Content-Type:application / json" and "Authorization:Bearertoken". Upon receiving the data, the back-end performs data integrity verification (e.g., required fields are not empty). If the verification passes, it returns "Submission Successful"; otherwise, it returns the specific error location (e.g., "Please enter the child's birth weight").

[0122] Step S1513: Collect additional symptom descriptions and supplementary explanations of the test results uploaded by parents through the supplementary explanation upload portal.

[0123] The supplementary information upload portal is displayed at the bottom of each entry as a "+Supplementary Information" link. Clicking it opens the file upload panel, supporting two types of supplementary information: text supplementation (within 200 characters), such as "When the child is in pain, he / she covers his / her right abdomen with his / her hand"; and multimedia supplementation (images / audio recordings), such as uploading photos of the child's facial expressions when in pain or audio recordings describing the symptoms. Uploaded files are compressed (image resolution ≤ 1920×1080, audio sampling rate 44.1kHz, bitrate 128kbps) and transmitted to the object storage service via a chunked upload algorithm (5MB per chunk), returning a unique resource identifier (URI). Supplementary information is stored in association with the corresponding entry ID. For example, the supplementary information for entry S15-09 "Specific location of abdominal pain" is "Uploaded a video of the child pointing out the location of the pain (URI: / uploads / 20230815 / xxx.mp4)". All supplementary content generates an associated index table when the form is submitted to ensure traceability.

[0124] Step S1514: Record the total end time of form filling, and calculate the difference between the total end time and the total start time as the total form filling time.

[0125] After the parent clicks the "Submit" button and the server returns a successful response, the front end obtains the total end timestamp by Date.now(), calculates the total fill duration = total end time - total start time, such as start time 1692512000000, end time 1692512360000, total duration = 360000 milliseconds (60 minutes). If the parent fills in multiple times (saves drafts along the way), the total duration is the sum of the fill durations of each time, such as 20 minutes for the first fill, 40 minutes for the second fill, total duration = 60 minutes. The total fill duration and the number of entries included in the form (23) are used to calculate the average fill duration = total duration / number of entries, which is used to evaluate the overall complexity of the form. If the average fill duration > 60 seconds per entry, it is prompted to simplify the expression.

[0126] Step S1515: Associate the fill duration, modification times, skipping behavior, skipping reasons, supplementary notes of each entry with the form content filled by the parent, supplementary symptom description, supplementary notes on the test results according to the entry number, fill information, behavior record, and supplementary content format. Integrate all associated data to form a multi-dimensional feedback data set.

[0127] The multi-dimensional feedback data set adopts a hybrid storage structure of relational + document type: the relational database (MySQL) stores structured data, including "form ID-entry ID-fill content-fill duration-modification times-skip flag-total duration"; the document database (MongoDB) stores unstructured data, including "entry ID-modification record array-skip reason details-supplementary note text-upload file URI". Both are associated by form ID and entry ID to support joint queries. For example, querying entry S15-08 can get: "fill content: □sweating from pain (selected), □don't want to walk (selected); fill duration: 180 seconds; modification times: 2 times (first selected '□vomiting', then canceled); skipping behavior: none; supplementary notes: uploaded a video of the child's pain (URI:...)". After the data set is generated, a data quality verification process is triggered to check for abnormal situations such as empty fill content but no skip flag, abnormally high modification times (>10 times), etc. Abnormal data is marked and entered into the manual review queue.

[0128] Step S160: Input the multi-dimensional feedback data set into the adjusted child tumor large model to analyze the relevance of form content and test results, the causes of parent interaction barriers, generate a form field optimization scheme and a preliminary diagnosis suggestion draft, adjust the field order and expression content of the preliminary pre-consultation form based on the form field optimization scheme, supplement diagnostic related verification entries based on the preliminary diagnosis suggestion draft, and obtain the final child tumor intelligent pre-consultation form containing the preliminary diagnosis direction.

[0129] Input the multi-dimensional feedback dataset into the adjusted child tumor large model, and the model analyzes the relevance between the form content and the examination results, such as whether the abdominal pain symptoms filled by the parents are closely related to the position of the mass in the ultrasound report, and the causes of the parents' interaction barriers, such as the fact that a certain field is modified multiple times due to complex expression. According to the analysis results, generate form field optimization schemes, such as adjusting the field order, simplifying the expression content, and preliminary diagnosis suggestion drafts, such as initially considering neuroblastoma and suggesting further CT examination. Based on the form field optimization scheme, adjust the preliminary pre-consultation form, such as moving the field with multiple modifications forward and simplifying the expression, and supplement diagnostic related verification items based on the preliminary diagnosis suggestion draft, such as "Does the child have a family history of neuroblastoma?", to finally obtain a child tumor intelligent pre-consultation form containing a preliminary diagnosis direction.

[0130] Step S161: Extract the form content filled by the parents, supplementary symptom description, and supplementary explanation of the examination results from the multi-dimensional feedback dataset, and compare them with the original examination result sheet and the child's past health records in the associated data set, to analyze the relevance between the form content and the examination result sheet.

[0131] Extract the form content filled by the parents from the multi-dimensional feedback dataset, such as the specific description of abdominal pain, supplementary symptom description such as cold sweats when abdominal pain, and supplementary explanation of the examination results such as hard mass texture. Compare the above content with the original examination result sheet in the associated data set, such as the abdominal ultrasound report, and the child's past health records, such as past medical examination, to analyze the relevance between the form content and the examination result sheet, and determine whether the symptoms filled by the parents are related to the abnormal indicators shown in the examination results.

[0132] Step S162: If any entry in the form content filled by the parents is not related to the abnormal indicators in the examination result sheet, mark the entry as needing to supplement the association guide and record the guide content direction that needs to be supplemented; if any entry in the form content filled by the parents is not consistent with the corresponding indicator value in the examination result sheet, mark the entry as needing to add a numerical verification prompt and record the verification prompt content that needs to be added.

[0133] During the analysis process, if it is found that a symptom filled in a certain field in the form, such as "the child has a cough", is not related to an abnormal indicator in the examination result sheet, such as a mass in the right adrenal gland area, mark the entry as needing to supplement the association guide and record the guide content direction as "explain that the cough is not related to the current examination results and does not need to be emphasized". If the value filled in a certain field in the form, such as the child's weight filled by the parents, is not consistent with the weight value recorded in the examination result sheet, mark the entry as needing to add a numerical verification prompt and record the verification prompt content as "the weight you filled is not consistent with the weight in the uploaded report, please check and fill in after verification".

[0134] Step S163: Extract the filling time, modification times, skipping behavior and skipping reasons of each item from the multi-dimensional feedback data set, and analyze the causes of parent interaction barriers.

[0135] From the multi-dimensional feedback data set, the filling time of each item is extracted, such as the filling of a certain item taking a long time, the number of modifications, such as modifying the filling content multiple times, whether there is a behavior of skipping the item, and the skipping reason, such as “not understanding the meaning of the question”. Comprehensive analysis of the causes of interaction barriers encountered by parents when filling out the form.

[0136] Step S164: If the filling time of any item exceeds the preset threshold of filling time and the number of modifications exceeds the preset threshold of modification times, and there is a statement understanding barrier in the skipping reason, record the optimization direction as simplifying the statement structure and splitting long sentences into short sentences; if the skipping reason of any item is information cognitive deficiency, record the optimization direction as adding example description or reducing the professional degree of the inquiry content.

[0137] When the filling time of a certain item exceeds the preset threshold, and the number of modifications also exceeds the preset threshold, and there is a statement understanding barrier in the skipping reason of the parent feedback, such as “the sentence is too long to understand”, record the optimization direction as simplifying the statement structure of the item and splitting long sentences into multiple short sentences to facilitate the understanding of parents. If the skipping reason of a certain item is information cognitive deficiency, such as “I don't know what lymph nodes are”, record the optimization direction as adding example description at the item, such as “lymph nodes are as small as soybean-sized bumps”, or reducing the professional degree of the inquiry content, asking in more popular language.

[0138] Step S165: Input the correlation analysis result and the interaction barrier cause analysis result into the adjusted child tumor large model, generate the parameter and examination result correlation weight based on the case report form structure, and generate the form field optimization scheme; the form field optimization scheme includes the item number to be modified, the modified expression content, the added correlation guide sentence or verification prompt, and the item order adjustment suggestion; at the same time, generate a threshold based on the preliminary diagnosis suggestion, combine the form content, supplementary symptom description, supplementary explanation of examination results and child's past health records, analyze the comprehensive correlation of symptoms, medical history and examination results, and generate a preliminary diagnosis suggestion draft; the preliminary diagnosis suggestion draft includes possible tumor type direction, further examination item suggestion, and preliminary treatment suggestion for current symptoms, and the expression contains prompt content that needs to be combined with further diagnosis of the doctor.

[0139] The correlation analysis results, such as which entries need to be supplemented with correlation guidance, and the interaction obstacle cause analysis results, such as which entries need to be simplified in expression, are input into the adjusted child tumor large model. The model generates parameter and examination result correlation weights based on the case report form structure, generates a form field optimization scheme that clearly indicates the entry number that needs to be modified, such as entry 5, the modified expression content, such as splitting long sentences into short sentences, the correlation guidance sentences or verification prompts that need to be added, such as adding “the symptom is related to the examination result, please describe in detail”, and the entry order adjustment suggestions, such as interchanging the order of entry 3 and entry 5. At the same time, the model generates a threshold based on the preliminary diagnosis suggestion, combines the form content, supplementary symptom description, supplementary explanation of examination results, and child health records, and comprehensively analyzes the correlation between symptoms, medical history, and examination result sheets to generate a preliminary diagnosis suggestion draft, which contains possible tumor type directions, such as neuroblastoma, further examination item suggestions, such as abdominal CT and tumor marker detection, and preliminary treatment suggestions for current symptoms, such as avoiding pressing the abdominal mass, with the prompt content “the above suggestions need to be combined with further diagnosis by the doctor” included in the expression.

[0140] Step S166: Reasonable verification is performed on the form field optimization scheme and the preliminary diagnosis suggestion draft. The form field optimization scheme is compared and corrected with the core logic of the case report form using a scheme logic consistency verification tool. Through the expression compliance detection module, absolute expressions in the preliminary diagnosis suggestion draft are identified and deleted to form the final form field optimization scheme and preliminary diagnosis suggestion draft.

[0141] Using the scheme logic consistency verification tool, the form field optimization scheme is compared with the core logic of the case report form to check for logical contradictions, such as whether the field order adjustment conforms to the diagnosis logic, and corrections are made for inconsistencies. Through the expression compliance detection module, the preliminary diagnosis suggestion draft is checked to identify and delete absolute expressions such as “definitely neuroblastoma” and modify them to “consider neuroblastoma possible” to form the final form field optimization scheme and preliminary diagnosis suggestion draft.

[0142] Step S167: Load the preliminary pre-consultation form and rearrange the order of the form entries according to the entry order adjustment suggestions in the form field optimization scheme. Move the entries that need to be supplemented with correlation guidance to the entries related to the corresponding examination result sheet indicators, and move the entries that need numerical verification prompts to the vicinity of the original examination result sheet related entries.

[0143] Load the preliminary pre-consultation form, adjust the order of the form items according to the suggested order in the form field optimization plan, and rearrange the order of the form items. Move the items that require supplementary association guidance to the positions after the corresponding examination result sheet indicators, such as placing the symptoms related to ultrasound examination results after the ultrasound report indicator. Move the items that require numerical verification prompts to the positions near the original examination result sheet, such as placing the body weight numerical verification item near the item containing the body weight examination result.

[0144] Step S168: For the items marked as requiring supplementary association guidance in the optimization plan, replace the original expression content of the item with the modified expression in the optimization plan and add the corresponding association guidance statement. For the items marked as requiring numerical verification prompts in the optimization plan, add the verification prompt content below the filling area of the item. For the items marked as having expression understanding barriers in the optimization plan, simplify the expression structure according to the optimization direction, split long sentences into short sentences, and remove complex modifiers. For the items marked as having information cognition deficiencies in the optimization plan, add example explanations.

[0145] For the items marked as requiring supplementary association guidance in the optimization plan, replace the original expression content of the item with the modified expression in the optimization plan and add the corresponding association guidance statement, such as "This symptom is related to the ultrasound report you uploaded, please fill it in according to the actual situation". For the items requiring numerical verification prompts, add the verification prompt content below the filling area of the item, such as "Please check the corresponding numerical value in the examination report you uploaded and fill it in". For the items with expression understanding barriers, simplify the expression structure according to the optimization direction, split long sentences into short sentences, and remove complex modifiers to make the expression more concise and easy to understand. For the items with information cognition deficiencies, add example explanations, such as "For example: Lymph node enlargement is manifested as small nodules that can be felt on the neck".

[0146] Step S169: Extract possible tumor type directions and further examination item suggestions from the preliminary diagnosis suggestion draft, generate corresponding diagnosis-related verification items for each tumor type direction, and generate corresponding examination guidance items for each further examination item suggestion. Insert the generated diagnosis-related verification items and examination guidance items into the adjusted form in logical order, with the insertion position being after the corresponding symptom or examination result sheet related items. Use the form logical chain detection tool to perform overall logical scanning on the inserted form, adjust the order of the items to eliminate logical breakpoints.

[0147] Extract the possible tumor type direction from the draft of the preliminary diagnosis suggestion, such as neuroblastoma, and the item suggestion for further examination, such as abdominal CT, tumor marker detection. Generate the corresponding diagnosis-related verification items for the tumor type direction of neuroblastoma, such as "Does the child have unexplained fever?" Generate the examination guide items for the abdominal CT examination suggestion, such as "Do you agree to perform abdominal CT examination to further clarify the mass condition?" Insert the above diagnosis-related verification items and examination guide items into the adjusted form in logical order, after the corresponding symptom or examination result single-related items, such as inserting the neuroblastoma verification items after the abdominal pain symptom items. Use the form logical chain detection tool to perform overall logical scanning on the inserted form, check whether there are logical breakpoints, such as incoherent connection between items, adjust the order of items to eliminate logical breakpoints.

[0148] Step S1610: Integrate all adjusted items, added guide sentences and verification items, use the core field integrity verification tool to check the integrity of the case report form core fields, examination result single-related items and preliminary diagnosis direction prompts in the form, and supplement the missing contents.

[0149] Integrate all adjusted items, added guide sentences and verification items. Use the core field integrity verification tool to check whether the case report form core fields such as basic information, medical history, symptoms, etc. are complete, whether the examination result single-related items such as the corresponding items of each examination result are complete, and whether the preliminary diagnosis direction prompts are included. Supplement the missing contents, such as supplementing the family medical history field.

[0150] Step S1611: Obtain the review results of medical personnel on the integrated form, match and verify the form content against the pediatric tumor pre-diagnosis clinical specification, evaluate the clinical reference value of the diagnosis-related verification items, and make final adjustments according to the review opinions to obtain the final pediatric tumor intelligent pre-diagnosis form containing the preliminary diagnosis direction.

[0151] Submit the integrated form to medical personnel for review, and the medical personnel match and verify the form content against the pediatric tumor pre-diagnosis clinical specification, evaluate the clinical reference value of the diagnosis-related verification items, such as whether a certain verification item is important for diagnosing neuroblastoma. Make final adjustments to the form according to the review opinions of the medical personnel, such as "need to add tumor marker examination related guide items", to obtain the final pediatric tumor intelligent pre-diagnosis form containing the preliminary diagnosis direction.

[0152] In an exemplary embodiment, an intelligent pre-diagnosis form generation system for pediatric tumors is provided, which can be a terminal, a server, etc., and its internal structure diagram can be as follows Figure 2The intelligent pre-diagnosis form generation system for children's tumors includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface, the display unit and the input device are connected to the system bus through the input / output interface. The processor of the intelligent pre-diagnosis form generation system for children's tumors is used to provide computing and control capabilities. The memory of the intelligent pre-diagnosis form generation system for children's tumors includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The input / output interface of the intelligent pre-diagnosis form generation system for children's tumors is used to exchange information between the processor and external devices. The communication interface of the intelligent pre-diagnosis form generation system for children's tumors is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be achieved through WIFI, mobile cellular network, near field communication or other technologies. The computer program is executed by the processor to implement an intelligent pre-diagnosis form generation method for children's tumors. The display unit of the intelligent pre-diagnosis form generation system for children's tumors is used to form a visually visible picture, which can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen. The input device of the intelligent pre-diagnosis form generation system for children's tumors can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the intelligent pre-diagnosis form generation system for children's tumors, or an external keyboard, touchpad or mouse, etc.

[0153] It should be noted that, in order to simplify the expression of the present disclosure and help understand one or more embodiments of the present application, in the foregoing description of embodiments of the present application, various features are sometimes combined into one embodiment, figure or description thereof.

Claims

1. An intelligent pre-consultation form generation method for children's tumors, characterized by, The method comprises: Identify the scene type of the child tumor pre-consultation, extract the scene demand characteristics corresponding to each scene type, form a scene demand characteristic set, and the scene type includes initial diagnosis pre-consultation scene, follow-up scene and special symptom investigation scene; Obtain pre-consultation related multi-source input data, and obtain an associated data set by associating the multi-source input data, the multi-source input data including parent initial symptom description text, child past health record and parent uploaded examination result sheet; Adjust the parameter configuration of the child tumor large model based on the scene demand characteristic set to obtain an adjusted child tumor large model adapted to the scene, and the parameter configuration includes case report form structure generation parameters, examination result association weight and initial diagnosis suggestion generation threshold; Input the associated data set into the adjusted child tumor large model to generate a consultation item semantic prototype including a case report form standard field, and the consultation item semantic prototype is associated with an examination result verification item and an initial diagnosis direction prompt; According to the parent expression habit sample, optimize the expression logic of the consultation item semantic prototype, convert the professional case report form field into a parent understandable expression, supplement the associated guide sentences of symptoms and examination results, form a preliminary pre-consultation form, push the preliminary pre-consultation form to the parent interactive terminal, collect the form content, supplementary symptom description and supplementary explanation of the examination results filled by the parents, and record the form filling time, field modification times and skipping behavior of the parents, and integrate to form a multi-dimensional feedback data set; Input the multi-dimensional feedback data set into the adjusted child tumor large model, analyze the association between the form content and the examination results and the causes of the parent interaction obstacles, generate a form field optimization scheme and an initial diagnosis suggestion draft, adjust the field order and expression content of the preliminary pre-consultation form based on the form field optimization scheme, and supplement the diagnosis related verification items combined with the initial diagnosis suggestion draft to obtain a final child tumor intelligent pre-consultation form including the initial diagnosis direction.

2. The intelligent pre-consultation form generation method for pediatric tumors according to claim 1, characterized in that, The method comprises: Obtain the scene identification of the parent selected through the scene selection entrance of the parent interactive terminal, or identify the scene tendency based on the expression content of the parent initial input to determine the current pre-consultation scene type; If the scene type is the initial diagnosis pre-consultation scene, extract the demand characteristics of the initial diagnosis pre-consultation scene, including complete medical history collection demand, basic physiological index coverage demand, examination result preliminary verification demand and initial diagnosis direction guidance demand, extract the core attention content corresponding to each demand characteristic and record it; If the scene type is the follow-up scene, extract the demand characteristics of the follow-up scene, including treatment after symptom change record demand, recent examination result comparison demand, follow-up conclusion generation demand and next follow-up time suggestion demand, extract the time range limit corresponding to each demand characteristic and record it; If the scene type is a special symptom screening scene, the demand characteristics of the special symptom screening scene are extracted, including target symptom detail description demand, symptom associated part examination demand, specific tumor index inquiry demand and screening conclusion prompt demand, the tumor type associated range corresponding to the target symptom is extracted and recorded; The demand characteristics of each scene type are described in words, and the priority of the demand characteristics is marked; The demand characteristic descriptions of different scene types are sorted by scene type, demand characteristic, priority and core content, repeated demand characteristic descriptions between different scenes are removed, and a scene demand characteristic set is formed. 3.The intelligent pre-consultation form generation method for pediatric tumors according to claim 1, wherein, The acquisition of the pre-consultation related multi-source input data, the association processing of the multi-source input data, and the obtaining of the associated data set, include: Collect the preliminary symptom description text input by the parents through the text input box of the parent interaction terminal, format the preliminary symptom description text, remove the preset symbols and repeated expressions, and retain the symptom occurrence time, manifestation form, duration and relieving or aggravating factor description; Obtain the child's past health records through the historical record import interface of the interaction terminal, sort the child's past health records in chronological order, and extract the key health events in each child's past health record, the child's past health records including past medical records, treatment plan records and past examination reports; Receive the examination result sheet uploaded by the parents through the file upload entrance of the interaction terminal, analyze the text content and numerical data of each examination result sheet, extract the index information related to the child's tumor diagnosis, and the examination result sheet includes image examination report, laboratory test report and pathological examination report; Associate the symptom keywords in the preliminary symptom description text with the key health events in the child's past health records, identify the relevance of the current symptoms and the past health events, extract the association results and record them; Associate the symptom sites in the preliminary symptom description text with the index information in the examination result sheet, match the examination index corresponding to the symptom site, extract the associated index name and value and record them; Associate the key health events in the child's past health records with the recent indicators in the examination result sheet, analyze the trend of the indicators after treatment, extract the trend analysis results and record them; Integrate the association results of symptoms and past records, the association results of symptoms and examination indicators, and the trend analysis results of past records and recent indicators, and add the original preliminary symptom description text, the sorted child's past health records and the analyzed examination result sheet index data to form an associated data set. 4.The intelligent pre-consultation form generation method for pediatric tumors according to claim 1, wherein, The parameter configuration of the child tumor large model is adjusted based on the scene demand characteristic set, and an adjusted child tumor large model adapted to the scene is obtained, including: Extract the demand characteristics corresponding to the current scene type from the scene demand characteristic set, and determine the adjustment direction of each demand characteristic to the model parameters; If the current scenario is an initial consultation scenario, set the complete case report form field generation mode for the form structure generation parameters, and set the form to include basic information field, medical history field, symptom field, examination result field and preliminary diagnosis prompt field. At the same time, adjust the field generation order parameter so that the fields are arranged according to the preset logic. Regarding the weight parameters associated with the examination results, if the scenario requirements include the requirement for preliminary verification of the examination results, the weight values ​​associated with the examination result indicators and the consultation items will be adjusted to values ​​that meet the requirements for preliminary verification of the examination results, so that the generated consultation items will be associated with the uploaded examination result single indicator first. For the generation of preliminary diagnostic suggestions, a threshold parameter is set according to the preliminary diagnostic direction guidance requirements in the scenario. The threshold for generating diagnostic suggestions that meets the preliminary diagnostic direction guidance requirements is set so that preliminary diagnostic direction prompts can be generated after obtaining basic symptoms and preliminary examination results. At the same time, the scope of the diagnostic suggestions is limited to avoid absolute statements. If the current scenario is a follow-up visit, adjust the form structure generation parameters of the case report form to the follow-up-specific mode, set the form to retain the fields of previous treatment records and current symptom changes, add the fields of treatment effect evaluation and next follow-up plan, and adjust the field generation order parameters so that the fields are arranged according to the preset logic; Adjust the weighting parameters for the correlation of examination results, adjusting the weighting of the comparison between recent and previous examination results to a value that meets the requirements for comparing examination results, so that the generated consultation items include inquiries about the reasons for changes in examination results; adjust the threshold parameter for generating preliminary diagnostic suggestions to a value that meets the requirements for complete data, and only generate follow-up conclusion suggestions after obtaining data comparing post-treatment symptoms with examination results. The adjusted case report form structure generates parameters, examination result association weights, and preliminary diagnostic suggestions are input into the parameter adjustment interface of the pediatric tumor model. Call the model adaptability testing program, input test data that matches the current scenario, and obtain the output test questions and draft diagnostic suggestions; Compare the test output results with the characteristics of the scenario requirements, check whether the fields of the case report form cover the scenario requirements, check whether the association of the result sheet meets the weight settings, and check whether the diagnostic suggestions meet the threshold requirements; If the fit does not meet expectations, readjust the corresponding parameters and test again until the test output fully matches the characteristics of the scenario requirements, thus obtaining an adjusted large model of pediatric tumors adapted to the scenario. 5.The intelligent pre-consultation form generation method for pediatric tumors according to claim 1, wherein, The process of inputting the associated data set into the adjusted pediatric tumor model to generate semantic prototypes of consultation entries containing standard fields from case report forms includes: Extract the set of symptom keywords from the preliminary symptom description text, the sorted key events in the child's past health records, the parsed single-indicator data of the examination results, and the correlation results between the data from the associated dataset, and organize them into input data groups according to data type. The input data set is input into the field mapping layer of the adjusted pediatric tumor large model case report form. Parameters are generated based on the case report form structure. The content in the input data set is matched with the fields of the standard case report form to determine the types of field to be generated in the case report form. generating a basic expression framework for each determined case report table field type, the basic expression framework containing a field name, an inquiry content guide, and an input format prompt; identifying abnormal items of single-index examination result data in the input data set based on the examination result correlation weight, generating corresponding examination result verification items for the abnormal indexes, and adding the items to the basic expression framework of the corresponding case report table field; generating a threshold value based on the preliminary diagnosis suggestion, analyzing the relevance of the symptom keywords, the child's past health records, and the single-index examination results in the input data set, generating a preliminary diagnosis direction prompt if the correlation result reaches the diagnosis suggestion generation threshold, and adding the prompt content to the corresponding case report table field in a predetermined format; performing semantic integration on the generated content containing the case report table field basic expression framework, the examination result verification items, and the preliminary diagnosis direction prompt, removing expressions unrelated to the current scene requirements from the integrated content, arranging the integrated content in the order of the case report table fields, each field corresponding to a complete expression containing the basic inquiry, the verification items, and the prompt, forming a semantic prototype of the inquiry item. 6.The intelligent pre-consultation form generation method for pediatric tumors according to claim 1, wherein, The expression logic for optimizing the semantic prototype of the inquiry item according to the parent expression habit sample converts the professional case report table field into an expression that parents can understand, supplements the correlation guide sentences of symptoms and examination results, and forms a preliminary pre-inquiry form, including: Collecting expression habit samples from parents of different backgrounds, dividing them into symptom description sample groups, terminology understanding sample groups, and examination result feedback sample groups according to the expression scene, the expression habit samples containing the parents' expressions of describing children's symptoms, understanding medical terminology, and feeding back examination results; Analyzing the expression texts of the symptom description sample group, counting the common colloquial words used by parents when describing symptoms, forming a symptom colloquial word table; analyzing the expression texts of the terminology understanding sample group, extracting the professional case report table field names that parents have difficulty understanding and the corresponding colloquial conversion methods, forming a case report table field colloquial conversion table; analyzing the expression texts of the examination result feedback sample group, summarizing the common expression logic used by parents when feeding back examination results, and forming an examination result guide expression template; Extracting the professional case report table field names in the semantic prototype of the inquiry item, and replacing them with colloquial expressions according to the case report table field colloquial conversion table; extracting the symptom-related expressions in the semantic prototype of the inquiry item, and replacing the professional symptom terminology with colloquial words according to the symptom colloquial word table; Supplementing the correlation guide sentences before and after the expression of the examination result verification item in the semantic prototype of the inquiry item, and referring to the examination result guide expression template; Performing a logical coherence check on the optimized expression content, using a field logic order verification tool to compare the field order after colloquial expression with the standard logic of the case report table, and checking and correcting the conflict content between the guide sentences and the verification items and diagnosis prompts through an expression conflict detection module. Obtaining the understanding degree of the parents representatives of different backgrounds reading the optimized expression content and the questions raised, adjusting the expression mode again according to the questions, arranging all the items in the order of the optimized case report form fields, each item containing the basic inquiry of popular expression, the associated guiding sentence, the examination result verification item and the preliminary diagnosis direction prompt, forming a preliminary pre-diagnosis form. 7.The intelligent pre-consultation form generation method for pediatric tumors according to claim 1, wherein, The preliminary pre-diagnosis form is pushed to the parent interactive terminal, the form content filled by the parents, the supplementary symptom description and the supplementary explanation of the examination result are collected, and the form filling time, the field modification times and the skipping behavior of the parents are recorded to form a multi-dimensional feedback data set, including: The items of the preliminary pre-diagnosis form are loaded to the form display interface of the parent interactive terminal in the optimized order, each item is provided with an independent filling area, and the filling area includes a text input box, an option selection button and a supplementary explanation upload entrance; When the parent clicks the start filling button of the form display interface, the form filling record program of the interactive terminal is started, and the total start time of the form filling is recorded; For each item, when the parent enters the filling area of the item, the filling start time of the item is recorded, when the parent clicks the next step or the skip button, the filling end time of the item is recorded, and the difference between the filling end time and the filling start time is calculated as the filling time of the item; When the parent modifies the input content in the filling area and submits, the number of modification operations is recorded, the content before modification and the content after modification are saved after each modification to form the modification record of the item; When the parent clicks the skip button, the skipping behavior of the item is recorded, and a supplementary explanation window is popped up to prompt the parent to select the skipping reason, the skipping reason includes information cognition deficiency, content irrelevance and expression understanding obstacle, the parent is allowed to input the supplementary explanation, and the skipping reason and the supplementary explanation content are recorded; When the parent completes the filling of all items or clicks the submit button, the form content input by the parent in each item filling area is collected, including the text input content and the option selection result; The supplementary symptom description text and the supplementary explanation of the examination result uploaded by the parent through the supplementary explanation upload entrance are collected; The total end time of the form filling is recorded, and the difference between the total end time and the total start time is calculated as the total filling time of the form; The filling time, the modification times, the skipping behavior, the skipping reason and the supplementary explanation of each item are associated with the form content, the supplementary symptom description and the supplementary explanation of the examination result filled by the parent according to the item number, the filling information, the behavior record and the supplementary content, all the associated data are integrated to form a multi-dimensional feedback data set. 8.The intelligent pre-consultation form generation method for pediatric tumors according to claim 1, wherein, The multi-dimensional feedback data set is input into the adjusted children tumor large model, the correlation between the form content and the examination result and the causes of the parent interaction obstacle are analyzed, and a form field optimization scheme and a preliminary diagnosis suggestion draft are generated, including: The form content filled by the parents, the supplementary symptom description and the supplementary explanation of the examination result are extracted from the multi-dimensional feedback data set, and are compared with the original examination result sheet and the children's past health records in the associated data set to analyze the correlation between the form content and the examination result sheet. If any one entry in the form content is filled with a symptom unrelated to the abnormal indicators in the examination result sheet, mark the entry as needing to supplement the association guidance and record the content direction of the guidance to be supplemented; If any one entry in the form content is filled with a value inconsistent with the corresponding indicator value in the examination result sheet, mark the entry as needing value verification prompt and record the verification prompt content to be added; Extract the filling duration, modification times, skipping behavior and skipping reasons of each entry from the multi-dimensional feedback data set to analyze the causes of parental interaction barriers; If the filling duration of any one entry exceeds the preset threshold of filling duration and the modification times exceeds the preset threshold of modification times, and there is a statement understanding barrier in the skipping reasons, record the optimization direction as simplifying the statement structure and splitting long sentences into short sentences; If the skipping reason of any one entry is information cognitive deficiency, record the optimization direction as adding example explanation or reducing the professional degree of the inquiry content; Input the association analysis results and the interaction barrier cause analysis results into the adjusted child tumor large model, generate the parameter and examination result association weight based on the form structure of the case report form, and generate the form field optimization scheme; the form field optimization scheme includes the entry number to be modified, the modified statement content, the association guidance sentence or verification prompt to be added, and the entry order adjustment suggestion; At the same time, generate a threshold based on the preliminary diagnosis suggestion, combine the form content, the supplemented symptom description, the supplemented explanation of the examination result and the child's past health records, analyze the comprehensive association of symptoms, medical history and examination result sheet, and generate a preliminary diagnosis suggestion draft; the preliminary diagnosis suggestion draft includes the possible tumor type direction, the item suggestion for further examination, the preliminary treatment suggestion for the current symptoms, and the prompt content to be combined with the doctor's further diagnosis in the statement; Reasonably check the form field optimization scheme and the preliminary diagnosis suggestion draft, use a scheme logic consistency checking tool to compare and correct the form field optimization scheme and the core logic of the case report form; through a statement compliance detection module, identify the absolute statement in the diagnosis suggestion draft and delete it, to form the final form field optimization scheme and the preliminary diagnosis suggestion draft. 9.The intelligent pre-consultation form generation method for pediatric tumors according to claim 1, wherein, Adjust the field order and statement content of the preliminary pre-diagnosis form based on the form field optimization scheme, supplement the diagnosis related verification entries based on the preliminary diagnosis suggestion draft, and obtain the final child tumor intelligent pre-diagnosis form containing the preliminary diagnosis direction, including: Load the preliminary pre-diagnosis form, adjust the order of the form entries according to the entry order adjustment suggestion in the form field optimization scheme, adjust the entries needing to supplement the association guidance to the entries related to the corresponding examination result sheet indicators, and adjust the entries needing value verification prompt to the entries near the original examination result sheet; For the entries marked as needing to supplement the association guide in the optimization scheme, replace the expression content of the entry with the modified expression in the optimization scheme, and add the corresponding association guide statement; for the entries marked as needing numerical verification prompts in the optimization scheme, add verification prompt content below the fill-in area of the entry; for the entries marked as filling duration exceeding the filling duration preset threshold and modification times exceeding the modification times preset threshold, and at the same time, the expression understanding disorder exists in the skip reason in the optimization scheme, simplify the expression structure according to the optimization direction, split long sentences into short sentences, and remove complex modifiers; for the entries marked as skip reason is information cognitive missing in the optimization scheme, add example explanation; Extract possible tumor type directions and further examination item suggestions from the preliminary diagnosis proposal draft, and generate corresponding diagnosis related verification entries for each tumor type direction; Generate corresponding examination guide entries for each further examination item suggestion, insert the generated diagnosis related verification entries and examination guide entries into the adjusted form in logical order, and insert the position behind the corresponding symptom or examination result single related entry, use the form logical chain detection tool to perform overall logical scanning on the inserted form, adjust the order of the entries to eliminate logical breakpoints; Integrate all adjusted entries, added guide statements and verification entries, use the core field integrity verification tool to check the integrity of the case report form core field, examination result single association item and preliminary diagnosis direction prompt in the form, and supplement the missing content; Obtain the review results of medical personnel on the integrated form, match and verify the form content against the pediatric tumor pre-diagnosis clinical specification, evaluate the clinical reference value of the diagnosis related verification entries, make final adjustments according to the review opinions, and obtain the final pediatric tumor intelligent pre-diagnosis form containing the preliminary diagnosis direction.

10. An intelligent pre-diagnosis form generation system for children's tumors, characterized by, Comprise: A processor; A machine-readable storage medium for storing machine-executable instructions of the processor; Wherein, the processor is configured to execute the machine-executable instructions to perform the intelligent pre-diagnosis form generation method for pediatric tumors in any one of claims 1 to 9.

Images