Orthodontic scheme intelligent auxiliary decision-making system based on multi-source data

Abstract

The invention discloses an intelligent auxiliary decision-making system for an orthodontic scheme based on multi-source data, and relates to the technical field of orthodontics, and the system comprises three core units: a multi-source data platform construction unit realizes structured storage and efficient interaction of data; the problem list and diagnosis intelligent generation unit analyzes tooth, jaw and soft tissue data based on three-dimensional space dimensions, automatically compares standard reference values to generate a problem list and a structured diagnosis conclusion, and provides a quantitative basis; the intelligent recommendation unit for the orthodontic treatment scheme fuses multi-modal data features, generates a preliminary scheme by using a deep learning model, combines similar case retrieval and knowledge graph optimization logic, and continuously improves recommendation accuracy through a clinical feedback closed loop. Three-dimensional fusion and automatic diagnosis of multi-source data are realized, subjective experience deviation is reduced, diagnosis and treatment efficiency is improved, and standardized and personalized auxiliary decision support is provided for orthodontics doctors.

Description

Technical Field

[0001] The present invention relates to the technical field of oral orthodontics, and in particular to an intelligent auxiliary decision-making system for orthodontic plans based on multi-source data. Background Art

[0002] Dentofacial deformities refer to deformities of the teeth, jaws, and craniofacial area caused by genetic and environmental factors, which affect the growth and development of the dentofacial area, oral health and function, facial aesthetics, and can even cause systemic diseases such as indigestion. In recent years, with the development of social economy and the improvement of people's requirements for facial aesthetics, more and more patients choose to improve maxillofacial deficiencies through orthodontic treatment. In order to obtain satisfactory orthodontic treatment results, it is particularly important to carefully formulate a diagnosis and treatment plan before the treatment process begins. However, traditional orthodontic diagnosis and treatment methods have many shortcomings. Therefore, it is particularly important to invent an intelligent auxiliary decision-making system for orthodontic plans based on multi-source data.

[0003] The existing technology also has the following technical defects, which are specifically reflected in: 1. In the existing technology, in traditional orthodontic diagnosis and treatment, doctors' judgment of patients' maxillofacial deformity conditions relies largely on subjective experience judgment and evaluation. Different doctors may come to different conclusions due to factors such as personal experience and judgment differences, and the data interpretation may be subjective and biased.

[0004] 2. In existing orthodontic treatment technologies, patient intraoral scan data (STL format), imaging data (DICOM format), clinical examination records (text tables), and 3D facial photography (OBJ format) are scattered across different systems, resulting in a fragmented and isolated state. While each system carries valuable information, it lacks effective integration, creating information silos. Doctors must manually compare data across platforms, which can easily lead to missing key information, inaccurate treatment plans, and even misjudgments. Summary of the Invention

[0005] The purpose of the present invention is to provide an intelligent auxiliary decision-making system for orthodontic plans based on multi-source data, which solves the problems existing in the background technology.

[0006] In order to solve the above technical problems, the present invention adopts the following technical solutions: The present invention provides an intelligent auxiliary decision-making system for orthodontic solutions based on multi-source data, including: a multi-source data platform construction unit: used to build a structured data management platform, classify and collect multi-source data, manage hierarchical labels and modularly store them, and realize efficient data calling, querying and visual interaction through the MySQL database and front-end and back-end architecture.

[0007] Question list and diagnosis intelligent generation unit: used to automatically compare standard reference values ​​based on multi-source data, analyze the teeth, jaws, and soft tissue structures from a three-dimensional spatial dimension, generate question lists and structured diagnostic conclusions, and provide quantitative basis for clinical review.

[0008] Orthodontic treatment plan intelligent recommendation unit: used to integrate multimodal data features, use deep learning models to generate preliminary treatment plans, combine similar case retrieval with knowledge graph optimization plan logic, and continuously improve recommendation accuracy through a clinical feedback loop.

[0009] Preferably, the multi-source data platform construction unit includes: a data collection and structured storage module, a multi-source data hierarchical classification module, a data layer design module, and a case data query and presentation module.

[0010] The data collection and structured storage module is used to adopt multiple input and storage strategies for different types of data.

[0011] The multi-source data hierarchical classification module is used to finely classify patient multi-source data based on a hierarchical medical record labeling system.

[0012] The data layer design module is used to build a three-level structured database architecture using MySQL to optimize data management and application to meet complex clinical business needs.

[0013] The case data query and presentation module is the same as querying case data and presenting it.

[0014] Preferably, the data collection and structured storage module is specifically as follows: for data with fixed options, the system adopts a radio button or multiple-choice box input method, and after input, the data is stored in a system database in text format.

[0015] When faced with data with different options and whose subsequent input content will change, a combination of "selection box + dynamic input box" is used for input. First, the general category of the data is determined through the selection box, and then the dynamic input box is used to allow the user to enter specific content. When storing, it is also saved in text form.

[0016] To improve data quality and utilization efficiency for model measurement, cephalometric measurement and other data, the system directly parses the standardized files exported by the hospital's existing dolphin system and enters the data into the system.

[0017] For the specific handwritten content in the medical records, the quadrant coding method is used to convert it into a standardized text form and then store it.

[0018] Preferably, the multi-source data hierarchical classification module is specifically: "Gold" label: a set of key core indicators in diagnosis, covering the chief complaint, key indicators of model analysis, and imaging analysis results, which plays a direct and decisive role in diagnosis and treatment decisions. It is usually collected by doctors based on standardized diagnosis and treatment processes and has the highest weight.

[0019] Important tags: Contains some secondary data for image, facial, and model analysis. Although not required for all cases, they can serve as alternative data to "gold" tags in specific diagnosis and treatment scenarios, providing key diagnosis and treatment basis. They have high clinical value and will be given priority in data retrieval and analysis.

[0020] Ordinary tags: mainly collect common information from daily clinical examinations, which have limited impact on the diagnosis of individual cases. However, in large-scale data analysis, they can provide important statistical information for diagnosis and treatment, and assist in the formulation of diagnosis and treatment plans.

[0021] Supplementary tags: Used to record unstructured extended information such as patient oral information and doctor input, such as unconventional situations. They usually do not constitute the core basis for diagnosis and treatment decisions, but they have considerable value in formulating refined and personalized treatment plans.

[0022] Preferably, the problem list and diagnosis intelligent generation unit includes: a data acquisition module, a standard comparison module, a diagnosis analysis module, a result aggregation module and a result display module.

[0023] The data acquisition module is used to call the patient's complete case information from the database, covering multi-source heterogeneous data including oral hygiene status, facial analysis, model analysis, CBCT images and soft tissue three-dimensional facial features.

[0024] The standard comparison module is used to automatically screen out the standard reference range related to the diagnostic indicators at the corresponding stage according to the patient's dentition development stage. After completing the indicator screening, a specific comparison function is set within the system to determine whether the indicator is in the normal range, marginal range or abnormal range.

[0025] The diagnostic analysis module is used to analyze all the data under the module by calling the comparison function item by item for the data of different structures or systems in the case. The comparison results of each data point will be spliced ​​into the module-level diagnostic description using an explanatory language, thereby generating a problem list and preliminary diagnostic conclusion for the module.

[0026] The result aggregation module is used to integrate the diagnosis results of the module into the global diagnosis report, and then cyclically analyze the data of the next module and perform comparison, judgment and diagnosis result splicing according to the same logic.

[0027] The result display module is used to present multi-dimensional diagnostic results in a structured form. For the three major anatomical structures of teeth, jaws, and soft tissues, the diagnostic conclusions in the three spatial dimensions of sagittal, vertical, and horizontal are output respectively. At the same time, based on the patient's specific condition, the system intelligently filters key data and standard control information from the complete diagnostic basis library, and selectively displays the basis content that is strongly related to the current diagnostic conclusion.

[0028] Preferably, the orthodontic treatment plan intelligent recommendation unit includes: a multi-source data integration module, a preliminary plan generation module, a plan optimization module and a continuous optimization module.

[0029] The data integration and preprocessing module is used to clean case data, remove noise, outliers and missing values, and ensure data quality;

[0030] The multi-source data integration module is used to integrate multi-source data such as patient medical records, images, oral scan data, etc., perform cleaning, standardization processing and feature extraction, and construct a unified feature representation.

[0031] The preliminary treatment plan generation module is used to fuse the extracted features, combine deep learning technology and pre-trained medical knowledge models to build a treatment plan generation model, and generate a preliminary treatment plan.

[0032] The solution optimization module is used to introduce a similar case retrieval function, find cases with similar conditions to the current patient through multimodal feature similarity, optimize the treatment plan in combination with reference cases, and improve personalization and accuracy.

[0033] The continuous optimization module is used to integrate the treatment plan recommendation module into the orthodontic auxiliary diagnosis and treatment system, continuously optimize the model through clinical feedback, and achieve efficient and accurate treatment plan recommendations.

[0034] The beneficial effects of the present invention are: 1. In the present invention, multi-source data is integrated, covering data such as three-dimensional dental models, facial hard tissue morphology, facial soft tissue morphology, etc., to establish an objective evaluation model. It can systematically analyze the three anatomical structures of teeth, jaws and soft tissues from the three spatial dimensions of sagittal, vertical and horizontal directions to draw a diagnosis, comprehensively analyze the patient's oral functional status, facial aesthetic characteristics and orthodontic risk factors, and help formulate a more scientific, reasonable and personalized orthodontic treatment plan.

[0035] 2. In this invention, based on authoritative guidelines and clinical experience, patient case data is analyzed, problems are automatically identified from multiple dimensions and levels, and diagnostic conclusions and evidence are generated, thereby improving the scientific nature and credibility of auxiliary diagnosis. Based on deep learning and multimodal data fusion technology, a system is constructed to generate an orthodontic treatment plan. It can generate personalized orthodontic treatment plans based on the unique data characteristics of each patient. A similar case retrieval function is introduced to screen out cases similar to the current patient's condition from a large number of historical cases, and the preliminary plan is optimized in combination with reference cases to make the treatment plan more suitable for the individual patient's condition and improve the accuracy and effectiveness of the treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0037] Figure 1 This is a schematic diagram of the system structure connection of the present invention. DETAILED DESCRIPTION

[0038] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0039] Reference Figure 1 As shown, the present invention provides an intelligent auxiliary decision-making system for orthodontic solutions based on multi-source data, including: a multi-source data platform construction unit: used to build a structured data management platform, classify and collect multi-source data, manage hierarchical labels and modularly store them, and realize efficient data calling, querying and visual interaction through MySQL database and front-end and back-end architecture.

[0040] In a specific embodiment, the multi-source data platform construction unit includes: a data collection and structured storage module, a multi-source data hierarchical classification module, a data layer design module, and a case data query and presentation module.

[0041] The data collection and structured storage module is used to adopt multiple input and storage strategies for different types of data.

[0042] The multi-source data hierarchical classification module is used to finely classify patient multi-source data based on a hierarchical medical record labeling system.

[0043] The data layer design module is used to build a three-level structured database architecture using MySQL to optimize data management and application to meet complex clinical business needs.

[0044] In the present invention, multi-source data, including three-dimensional dental models, facial hard tissue morphology, facial soft tissue morphology and other data, are integrated to establish an objective evaluation model. This model can systematically analyze the three anatomical structures of teeth, jaws and soft tissues from the three spatial dimensions of sagittal, vertical and horizontal to draw a diagnosis, and comprehensively analyze the patient's oral functional status, facial aesthetic characteristics and orthodontic risk factors, to help formulate a more scientific, reasonable and personalized orthodontic treatment plan.

[0045] It should be noted that the data layer design module is specifically as follows: core data modular storage: modular storage design is carried out based on clinical dimensions, specifically including: patient basic information, oral hygiene status, facial analysis results, oral model analysis data, treatment plan, CBCT (cone beam computed tomography) imaging information, oral scan data and soft tissue three-dimensional facial data and other core dimensions, which improves the efficiency of targeted data calling.

[0046] The index management layer realizes multi-module collaboration: In order to achieve efficient data retrieval and associated query, the system has designed a separate medical record index table to store the unique identifiers (IDs) of various medical record modules, realizing unified indexing and module navigation of the complete medical record structure.

[0047] Auxiliary data table independent storage: The system creates a data table specifically for storing doctor information, which comprehensively and meticulously records all key information about doctors, including their basic personal information and professional titles. This facilitates subsequent plan review and other research.

[0048] The case data query and presentation module is the same as querying case data and presenting it.

[0049] It should be noted that, for the case data query and presentation module, the system builds a business logic layer based on the Spring framework and the MyBatis-Plus persistence tool. For each type of data table in the database, corresponding entity classes are created, and supporting Mapper interfaces, Controller controller classes and Service service classes are further constructed to achieve standardized access to data and business logic encapsulation. When querying a case, the system first finds the unique identifier (ID) of the data of each module related to the medical record based on the medical record index table, and then uses the corresponding Mapper to read the data in the corresponding table in module order, and then aggregates and assembles the data of each module in the Service layer, and finally encapsulates the assembled medical record data into a unified JSON format object for use by the front-end system.

[0050] After receiving JSON-formatted medical record data from the backend, the frontend first parses the data structure and logically divides it into pre-set modules. The specific content in each module is mapped to the corresponding frontend display component through a data binding mechanism. This enables diverse interactive interface functions such as displaying basic patient information, previewing CBCT images, and visualizing facial analysis charts, thereby enabling dynamic display and interactive operation of medical record data. This design ensures decoupling of frontend and backend data, a clear structure, and smooth interaction, effectively improving the system's user experience and maintainability.

[0051] In a specific embodiment, the data collection and structured storage module is specifically as follows: for data with fixed options, such as the patient's gender, whether or not they are allergic, and other information, the system uses a radio button or multiple-choice box entry method. After entry, the data is stored in the system database in text format.

[0052] When faced with data with different options and whose subsequent input content will change, such as patients' bad habits, family medical history, etc., a combination of "selection box + dynamic input box" is used for input. First, the general category of the data is determined through the selection box, and then the dynamic input box is used to allow the user to enter specific content. When storing, it is also saved in text form.

[0053] To improve data quality and utilization efficiency for model measurement, cephalometric measurement and other data, the system directly parses the standardized files exported by the hospital's existing dolphin system and enters the data into the system.

[0054] For specific handwritten content in medical records, such as special examination results and diagnostic opinions written by doctors, the object coding method is used to convert them into standardized text form and then store them.

[0055] In a specific embodiment, the multi-source data hierarchical classification module is specifically: "Gold" label (core diagnostic label): a set of key core indicators in diagnosis, covering chief complaints, key indicators of model analysis, and imaging analysis results, which plays a direct and decisive role in diagnosis and treatment decisions. It is usually collected by doctors based on standardized diagnosis and treatment processes and has the highest weight.

[0056] Important tags (key auxiliary diagnostic tags): Contains some secondary data for image, facial, and model analysis. Although they are not required for all cases, they can be used as alternative data to "gold" tags in specific diagnosis and treatment scenarios to provide key diagnosis and treatment basis. They have high clinical value and will be given priority in data retrieval and analysis.

[0057] Ordinary labels (routine clinical records): mainly collect common information from daily clinical examinations, such as basic patient information, general oral examination items, facial analysis reference data, etc., which have limited impact on the diagnosis of individual cases, but in large-scale data analysis, they can provide important statistical information for diagnosis and treatment, and assist in the formulation of diagnosis and treatment plans.

[0058] Supplementary tags (extended information): Used to record unstructured extended information stated by patients or entered by doctors, such as unconventional situations. They usually do not constitute the core basis for diagnosis and treatment decisions, but they have considerable value in formulating refined and personalized treatment plans.

[0059] Question list and diagnosis intelligent generation unit: used to automatically compare standard reference values ​​based on multi-source data, analyze the teeth, jaws, and soft tissue structures from a three-dimensional spatial dimension, generate question lists and structured diagnostic conclusions, and provide quantitative basis for clinical review.

[0060] In a specific embodiment, the problem list and diagnosis intelligent generation unit includes: a data acquisition module, a standard comparison module, a diagnosis analysis module, a result aggregation module and a result display module.

[0061] The data acquisition module is used to call the patient's complete case information from the database, covering multi-source heterogeneous data including oral hygiene status, facial analysis, model analysis, CBCT images and soft tissue three-dimensional facial features.

[0062] The standard comparison module is used to automatically screen out the standard reference range related to the diagnostic indicators of the corresponding stage according to the patient's dentition development stage (such as the deciduous dentition stage, the replacement dentition stage or the permanent dentition stage). After completing the indicator screening, a specific comparison function is set within the system to determine whether the indicator is in the normal range, the marginal range or the abnormal range. This function can be labeled according to the numerical range (such as "normal", "high", "low") to provide structured information for subsequent diagnosis.

[0063] The diagnostic analysis module is used to analyze the data of different structures or systems (such as teeth, jaws, soft tissues, etc.) in the case. The system calls the comparison function item by item to analyze all the data under the module. The comparison results of each data point will be spliced ​​into the module-level diagnostic description using an explanatory language, thereby generating a problem list and preliminary diagnostic conclusion for the module.

[0064] The result aggregation module is used to integrate the diagnostic results of each module into the global diagnostic report. It then loops through the data of the next module and performs comparison, judgment, and diagnostic result splicing according to the same logic. This process repeats on a module-by-module basis until all modules are analyzed, achieving a comprehensive diagnosis of the case data.

[0065] After obtaining diagnostic results from all modules and completing diagnostic analysis for a module, the system further searches and correlates the definitions and clinical explanations of each diagnosis in standard literature or guidelines based on the medical evidence involved. The system also searches the patient's problem list and original case data to ensure that the diagnosis is consistent with the actual situation. All diagnostic results and corresponding evidence are centrally stored and encapsulated as JSON data objects in a separate module structure, providing a sound data foundation for front-end display and physician interpretation.

[0066] The result display module is used to present multi-dimensional diagnostic results in a structured form. For the three major anatomical structures of teeth, jaws, and soft tissues, the diagnostic conclusions in the three spatial dimensions of sagittal, vertical, and horizontal are output respectively. At the same time, based on the patient's specific condition, the system intelligently filters key data and standard control information from the complete diagnostic basis library, and selectively displays the basis content that is strongly related to the current diagnostic conclusion.

[0067] Orthodontic treatment plan intelligent recommendation unit: used to integrate multimodal data features, use deep learning models to generate preliminary treatment plans, combine similar case retrieval with knowledge graph optimization plan logic, and continuously improve recommendation accuracy through a clinical feedback loop.

[0068] In a specific embodiment, the orthodontic treatment plan intelligent recommendation unit includes: a multi-source data integration module, a preliminary plan generation module, a plan optimization module and a continuous optimization module.

[0069] The data integration and preprocessing module is used to clean case data, remove noise, outliers and missing values, and ensure data quality.

[0070] The multi-source data integration module is used to integrate multi-source data such as patient medical records, images, oral scan data, etc., perform cleaning, standardization processing and feature extraction, and construct a unified feature representation.

[0071] Convert the patient's photos to a standard format, align them to a standard template, and normalize the pixel values to the range of [0, 1]. Align the oral scan model to the reference model using the ICP algorithm. For the medical record text, use cTAKES for word segmentation, filter out meaningless words such as "de" and ",", retain medical terms, use ClinicalBERT and BioWordVec to obtain word vectors, and generate sentence vectors through average pooling or [CLS] tags. For CT images and facial photos, use a pre-trained convolutional neural network (CNN) to extract two-dimensional features. For oral scan data, use a 3D convolutional neural network (3DCNN) to process three-dimensional point cloud data and extract three-dimensional structural features of teeth and gums. The extracted features are reduced in dimension through global average pooling (GAP) or adaptive average pooling (AAP) to retain the most important information and mapped into a unified feature space.

[0072] For text data, first preprocess the medical record text, including word segmentation, stop word removal, and stemming. Then, use pre-trained word embedding models such as ClinicalBERT or BioWordVec to convert medical terms into word vectors. These models have been pre-trained on medical literature and can capture the semantic information of medical concepts. For each sentence, use average pooling or [CLS] tags to generate sentence vectors, which will serve as the feature representation of the text data.

[0073] Adopt a method based on the Transformer architecture for feature fusion. Use 6 layers of stacking, with each layer containing 12 heads of self-attention mechanism: the first 3 layers focus on fine-grained intra-modal associations through local window attention, and the last 3 layers use global attention to establish cross-modal high-order logic. Insert a cross-attention module in the fourth layer, use text features as Query, and image / oral scan features as Key-Value to calculate semantic-guided visual weights and strengthen the explicit alignment between clinical descriptions and anatomical structures. Finally, adopt dynamic feature gating: use the text [CLS] token to generate a mixed weight α ∈ [0, 1] through two layers of MLP, and perform weighted fusion according to F_fused = α × (image pooled features) + (1 - α) × (oral scan pooled features), and concatenate with text features.

[0074] The preliminary plan generation module is used to fuse the extracted features, combine deep learning technology and pre-trained medical knowledge models to construct a treatment plan generation model, and generate a preliminary treatment plan.

[0075] The fused feature vector is passed as input to the treatment plan generation model. This model is a deep learning network that is trained by learning from a large amount of labeled case data. During the model training process, a preliminary treatment plan is generated by combining an open source pre-trained large language model with a small model. By learning from massive amounts of medical literature and clinical data, the open source pre-trained large language model can accurately analyze the pathological mechanisms, treatment methods, and prognosis assessments of dentofacial diseases, and then output preliminary treatment plans. Since the plans generated by the large model may contain logical contradictions or unreasonable aspects, the small model TabPFN is introduced to screen and optimize the data it generates, effectively eliminating unreasonable or low-quality plans, and significantly improving the overall quality of treatment plan generation.

[0076] The solution optimization module is used to introduce a similar case retrieval function, find cases with similar conditions to the current patient through multimodal feature similarity, optimize the treatment plan in combination with reference cases, and improve personalization and accuracy.

[0077] The system uses a vector-based similarity search algorithm to retrieve the top three similar cases (cosine similarity > 0.85) from the historical case database. A knowledge graph-driven reranking module optimizes the priority of candidate cases. For example, when a patient's ANB value falls below a set threshold, the system automatically increases the weight of non-surgical treatment options. By comparing newly generated treatment plans with historical plans, the model is provided with rich clinical context and empirical references, allowing the generated plans to fully consider a variety of clinical situations.

[0078] The continuous optimization module is used to integrate the treatment plan recommendation module into the orthodontic auxiliary diagnosis and treatment system, continuously optimize the model through clinical feedback, and achieve efficient and accurate treatment plan recommendations.

[0079] To ensure the logical consistency and clinical feasibility of treatment plans, a knowledge graph-based verification method was adopted. This method leverages the rich medical knowledge within the pre-trained large language model and combines it with a custom constraint graph to perform contradiction detection and logical verification on the generated treatment plans.

[0080] During the verification process, the system first breaks down the generated treatment plan into specific treatment measures and steps, then maps these to corresponding nodes and relationships in the knowledge graph. Using a graph neural network, it traverses the paths connecting treatment steps and medical entities, identifying hard conflicts and soft constraint deviations. Using the TransE model, it generates interpretable correction recommendations.

[0081] In this invention, based on authoritative guidelines and clinical experience, patient case data is analyzed, problems are automatically identified from multiple dimensions and levels, and diagnostic conclusions and evidence are generated, thereby improving the scientific nature and credibility of auxiliary diagnosis. Based on deep learning and multimodal data fusion technology, a system is constructed to generate an orthodontic treatment plan. It can generate personalized orthodontic treatment plans based on the unique data characteristics of each patient. A similar case retrieval function is introduced to screen out cases similar to the current patient's condition from a large number of historical cases, and the preliminary plan is optimized in combination with reference cases to make the treatment plan more suitable for the individual patient's condition and improve the accuracy and effectiveness of the treatment.

[0082] The above contents are merely examples and explanations of the concept of the present invention. Those skilled in the art may make various modifications or additions to the described specific embodiments or replace them in a similar manner. As long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, they should all fall within the scope of protection of the present invention.

Claims

1. An intelligent auxiliary decision-making system for orthodontic plans based on multi-source data, characterized by: include: Multi-source data platform construction unit: used to build a structured data management platform, classify and collect multi-source data, manage hierarchical tags, and modularize storage. It uses the MySQL database and front-end and back-end architecture to achieve efficient data call, query, and visual interaction. Question list and diagnosis intelligent generation unit: Based on multi-source data, it automatically compares standard reference values, analyzes the tooth, jaw, and soft tissue structure from a three-dimensional spatial dimension, generates a question list and structured diagnosis conclusions, and provides quantitative evidence for clinical review; Orthodontic treatment plan intelligent recommendation unit: used to integrate multimodal data features, use deep learning models to generate preliminary treatment plans, combine similar case retrieval with knowledge graph optimization plan logic, and continuously improve recommendation accuracy through a clinical feedback loop.

2. The orthodontic plan intelligent auxiliary decision-making system based on multi-source data according to claim 1, characterized in that: The multi-source data platform construction unit includes: a data collection and structured storage module, a multi-source data hierarchical classification module, a data layer design module, and a case data query and presentation module; The data collection and structured storage module is used to adopt multiple input and storage strategies for different types of data; The multi-source data hierarchical classification module is used to finely classify multi-source patient data based on a hierarchical medical record labeling system; The data layer design module is used to build a three-level structured database architecture using MySQL to optimize data management and application to meet complex clinical business needs; The case data query and presentation module is the same as querying case data and presenting it.

3. The orthodontic plan intelligent auxiliary decision-making system based on multi-source data according to claim 2, characterized in that: The data collection and structured storage module is specifically: For data with fixed options, the system uses radio button or multiple-choice box input. After input, the data is stored in the system database in text format. When faced with data with different options and whose subsequent input content may change, the system uses a combination of "selection box + dynamic input box" for input. First, the selection box determines the general category of the data, and then the dynamic input box allows the user to enter specific content. When storing, it is also saved in text format. To improve data quality and utilization efficiency for model measurement, cephalometric measurement, and other data, the system directly parses the standardized files exported by the hospital's existing Dolphin system and enters the data into the system; For specific handwritten content in medical records, the system uses quadrant coding method to convert it into standardized text form and then stores it.

4. The orthodontic plan intelligent auxiliary decision-making system based on multi-source data according to claim 2, characterized in that: The multi-source data hierarchical classification module is specifically: "Gold" label: A collection of key core indicators in diagnosis, including the chief complaint, key indicators of model analysis, and imaging analysis results. These indicators play a direct and decisive role in diagnosis and treatment decisions. They are usually collected by doctors based on standardized diagnosis and treatment processes and have the highest weight. Important tags: Contains some secondary data for image, facial, and model analysis. Although not required for all cases, they can serve as alternative data to "golden" tags in specific diagnosis and treatment scenarios, providing key diagnostic and treatment evidence. They have high clinical value and are prioritized during data retrieval and analysis. Common tags: These are mainly information collected during routine clinical examinations. They have limited impact on the diagnosis of individual cases, but in large-scale data analysis, they can provide important statistical information for diagnosis and treatment, assisting in the development of diagnosis and treatment plans. Supplementary tags: Used to record unstructured extended information such as patient oral information and doctor input, such as unconventional situations. They usually do not constitute the core basis for diagnosis and treatment decisions, but they have considerable value in formulating refined and personalized treatment plans.

5. The orthodontic plan intelligent auxiliary decision-making system based on multi-source data according to claim 1, characterized in that: The problem list and diagnosis intelligent generation unit includes: a data acquisition module, a standard comparison module, a diagnosis analysis module, a result aggregation module and a result display module; The data acquisition module is used to retrieve the patient's complete case information from the database, including multi-source heterogeneous data such as oral hygiene status, facial analysis, model analysis, CBCT images, and soft tissue three-dimensional facial features; The standard comparison module is used to automatically screen out the standard reference ranges related to diagnostic indicators at the corresponding stage according to the patient's dentition development stage. After completing the indicator screening, a specific comparison function is set within the system to determine whether the indicator is within the normal range, marginal range or abnormal range; The diagnostic analysis module is used to analyze all the data in the module by calling the comparison function item by item for the data of different structures or systems in the case. The comparison results of each data point will be spliced ​​into the module-level diagnostic description using an explanatory language, thereby generating a list of problems and a preliminary diagnostic conclusion for the module; The result aggregation module is used to integrate the diagnosis results of the module into the global diagnosis report, and then cyclically analyze the data of the next module and perform comparison, judgment and diagnosis result splicing according to the same logic; The result display module is used to present multi-dimensional diagnostic results in a structured form. For the three major anatomical structures of teeth, jaws, and soft tissues, the diagnostic conclusions in the three spatial dimensions of sagittal, vertical, and horizontal are output respectively. At the same time, based on the patient's specific condition, the system intelligently filters key data and standard control information from the complete diagnostic basis library, and selectively displays the basis content that is strongly related to the current diagnostic conclusion.

6. The orthodontic plan intelligent auxiliary decision-making system based on multi-source data according to claim 1, characterized in that: The orthodontic treatment plan intelligent recommendation unit includes: a multi-source data integration module, a preliminary plan generation module, a plan optimization module and a continuous optimization module; The data integration and preprocessing module is used to clean case data, remove noise, outliers and missing values, and ensure data quality; The multi-source data integration module is used to integrate multi-source data such as patient medical records, images, oral scan data, etc., perform cleaning, standardization and feature extraction, and construct a unified feature representation; The preliminary treatment plan generation module is used to fuse the extracted features, combine deep learning technology and pre-trained medical knowledge models to build a treatment plan generation model, and generate a preliminary treatment plan; The solution optimization module is used to introduce a similar case retrieval function, find cases with similar conditions to the current patient through multimodal feature similarity, and optimize the treatment plan based on reference cases to improve personalization and accuracy; The continuous optimization module is used to integrate the treatment plan recommendation module into the orthodontic auxiliary diagnosis and treatment system, continuously optimize the model through clinical feedback, and achieve efficient and accurate treatment plan recommendations.

Images