Automatic orthodontic risk assessment method and device
By using deep learning and large models to process patient clinical data, the system automatically assesses orthodontic risks, solving the problem of relying on doctors' subjective judgment in existing technologies and enabling more accurate risk assessment and personalized treatment plan selection.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-01-26
- Publication Date
- 2026-04-10
AI Technical Summary
Existing orthodontic risk assessment methods mainly rely on doctors' subjective judgment and lack objective quantitative tools, which leads to biases in the assessment results and affects the accuracy of orthodontic treatment and plan selection.
By employing deep learning and large models, and acquiring patients' clinical data, we use neural networks, knowledge graph networks, or language large models for data analysis and processing to generate risk parameters, train orthodontic risk prediction language large models, and achieve automated orthodontic risk assessment.
It improves the accuracy of orthodontic risk assessment, reduces reliance on physician experience, and enables dynamic risk assessment based on individual differences and treatment plans, providing personalized assessment results.
Smart Images
Figure CN121839092A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of orthodontic technology, and more specifically, to an automated orthodontic risk assessment method. Background Technology
[0002] Malocclusion is a common oral disease that affects maxillofacial development, oral health, oral function, and facial aesthetics. Severe malocclusion can impair pronunciation and appearance, leading to psychosomatic illnesses; it can also severely reduce chewing function, causing indigestion and gastrointestinal disorders. Orthodontic treatment has emerged to address malocclusion. Orthodontic treatment aligns teeth, adjusts occlusion, and harmonizes the face by moving teeth and reshaping alveolar bone. However, during or after orthodontic treatment, there is a potential for damage to the hard and soft tissues of the maxillofacial region. These risks include: tooth demineralization / caries, root resorption, tooth loosening, incomplete space closure, bone fenestration / cracking, alveolar bone resorption, periodontitis, braces-like facial deformity, temporomandibular joint disorder, black triangle, and relapse after treatment. Therefore, objectively and quantitatively assessing the risks of individual orthodontic treatment and balancing the risks and benefits is beneficial for treatment planning and decision-making, improving medical safety and satisfaction, and providing guidance for clinical medical quality control.
[0003] Currently, internationally used orthodontic risk assessment methods primarily rely on the subjective judgment of physicians, lacking suitable objective quantitative tools. Treatment risk is closely related to patient diagnosis, treatment plan, and compliance, and is influenced by patient condition, physician skill level, and medical environment. Patient-related factors include genetic factors, dental anatomy, demographic characteristics, malocclusion factors, root canal treatment, systemic medical history, short root abnormalities, and periodontal disease. Treatment-related factors include biomechanical factors, appliances and auxiliary devices, accelerated tooth movement, treatment timing, tooth movement distance and direction, treatment duration, and regular follow-up visits. Physician-patient cooperation includes oral hygiene, regular follow-up visits, and physician skill level. Due to the complexity of influencing factors and the subjectivity of physician judgment, assessment results from different physicians can vary, leading to insufficient objectivity and making it difficult to provide an effective reference standard for orthodontic treatment and plan selection.
[0004] Meanwhile, in orthodontics, treatment needs, treatment difficulty, and treatment outcomes are all assessed using corresponding indices. However, regarding the risks of orthodontic treatment, there are no reports, either domestically or internationally, on a systematic assessment of orthodontic risks. Summary of the Invention
[0005] The purpose of this application is to overcome the shortcomings of existing technologies and provide an automated orthodontic risk assessment method. By using deep learning and large models, the orthodontic risk assessment is automated and can be personalized based on individual patient differences, disease conditions and different treatment plans, thereby effectively improving the accuracy of orthodontic risk assessment.
[0006] The objective of this application is achieved through the following technical solution:
[0007] Firstly, this application proposes an automated orthodontic risk assessment method, the method comprising:
[0008] Acquire clinical data of the patient to be evaluated and optimize the clinical data to obtain first optimized data;
[0009] The second data is obtained by analyzing the first optimized data using neural networks, knowledge graph networks, or large language models;
[0010] Risk parameters are generated by processing the second data using knowledge graph networks or large language models, respectively.
[0011] A large language model for orthodontic risk prediction was trained using the first optimized data, the second data, and risk parameters to predict the orthodontic risk level.
[0012] In one possible implementation, the clinical data includes clinical examination information, questionnaire survey information, and imaging data. The step of acquiring the clinical data of the patient to be evaluated and optimizing the clinical data to obtain first optimized data includes:
[0013] Record clinical examination information and questionnaire survey information, and use database comparison to extract the first abnormal value in the clinical examination information and questionnaire survey information;
[0014] The neural network model is used to locate key anatomical points, segment key anatomical structures and lesions from imaging data. The imaging data includes two-dimensional images and three-dimensional images. The two-dimensional images include lateral radiographs, panoramic radiographs, facial photographs and intraoral photographs. The three-dimensional images include intraoral scanning models, CBCT, MRI and facial scanning data.
[0015] The key anatomical points were measured and analyzed, and the analysis results were compared with the database to obtain the second abnormal value.
[0016] Disease types can be identified by comparing key anatomical structures and lesions with databases or using neural network classification.
[0017] The first optimized data is obtained by cross-validating the data based on the first abnormal value, the second abnormal value, and the disease type.
[0018] In one possible implementation, the second data includes diagnostic results and treatment plans. The step of analyzing the first optimized data using neural networks, knowledge graph networks, or large language models to obtain the second data includes:
[0019] The diagnostic results are obtained by comprehensively analyzing the first optimized data using a neural network;
[0020] The diagnostic results can be processed using knowledge graph networks or large language models to generate multiple treatment plans.
[0021] In one possible implementation, the step of processing the second data using a knowledge graph network or a large language model to generate risk parameters includes:
[0022] Multiple treatment options are processed using knowledge graph networks or large language models to generate a first risk parameter that corresponds one-to-one with each treatment option;
[0023] The diagnostic results are processed using knowledge graph networks or large language models to generate a second risk parameter;
[0024] A third risk parameter is generated by processing the basic information of the patient to be assessed using a knowledge graph network or a language large model. The basic information of the patient to be assessed includes, but is not limited to, gender, age, race, education level, height, and weight.
[0025] In one possible implementation, the step of training a large language model for orthodontic risk prediction using first optimized data, second data, and risk parameters to predict orthodontic risk levels includes:
[0026] The orthodontic risk prediction language model was trained using the first optimized data, the second data, and risk parameters.
[0027] Predict orthodontic risk levels using a trained orthodontic risk prediction language model.
[0028] Secondly, this application proposes an automated orthodontic risk assessment device, the device comprising:
[0029] The acquisition module is used to acquire the clinical data of the patient to be evaluated and optimize the clinical data to obtain the first optimized data;
[0030] The analysis module is used to analyze the first optimized data using neural networks, knowledge graph networks, or large language models to obtain the second data;
[0031] The processing module is used to process the second data using knowledge graph networks or large language models to generate risk parameters.
[0032] The training module is used to train a large language model for orthodontic risk prediction using the first optimized data, the second data, and risk parameters to predict the orthodontic risk level.
[0033] In one possible implementation, the acquisition module is further configured to:
[0034] Record clinical examination information and questionnaire survey information, and use database comparison to extract the first abnormal value in the clinical examination information and questionnaire survey information;
[0035] The neural network model is used to locate key anatomical points, segment key anatomical structures and lesions from imaging data. The imaging data includes two-dimensional and three-dimensional images. The two-dimensional images include lateral radiographs, panoramic radiographs, facial radiographs and intraoral radiographs. The three-dimensional images include intraoral scanning models, CBCT, MRI and facial scanning data.
[0036] The key anatomical points were measured and analyzed, and the analysis results were compared with the database to obtain the second abnormal value.
[0037] Disease types can be identified by comparing key anatomical structures and lesions with databases or using neural network classification.
[0038] The first optimized data is obtained by cross-validating the data based on the first abnormal value, the second abnormal value, and the disease type.
[0039] In one possible implementation, the analysis module is further configured to:
[0040] The diagnostic results are obtained by comprehensively analyzing the first optimized data using a neural network;
[0041] The diagnostic results can be processed using knowledge graph networks or large language models to generate multiple treatment plans.
[0042] In one possible implementation, the processing module is further configured to:
[0043] Multiple treatment options are processed using knowledge graph networks or large language models to generate a first risk parameter that corresponds one-to-one with each treatment option;
[0044] The diagnostic results are processed using knowledge graph networks or large language models to generate a second risk parameter;
[0045] A third risk parameter is generated by processing the basic information of the patient to be assessed using a knowledge graph network or a language large model. The basic information of the patient to be assessed includes, but is not limited to, gender, age, race, education level, height, and weight.
[0046] In one possible implementation, the training module is further configured to:
[0047] The orthodontic risk prediction language model was trained using the first optimized data, the second data, and risk parameters.
[0048] Predict orthodontic risk levels using a trained orthodontic risk prediction language model.
[0049] The main solution and its various further alternatives described above can be freely combined to form multiple solutions, all of which are solutions that can be adopted and are claimed in this application; furthermore, the (non-conflicting alternatives) can also be freely combined with each other and with other alternatives. Those skilled in the art, after understanding the solution of this application, will realize from the prior art and common general knowledge that there are many combinations, all of which are technical solutions to be protected by this application, and will not be exhaustively listed here.
[0050] This application discloses an automated orthodontic risk assessment method and device. First, clinical data of the patient to be assessed is acquired and optimized to obtain first optimized data. Then, the first optimized data is analyzed using a neural network, knowledge graph network, or language big data model to obtain second data. Next, the second data is processed using the knowledge graph network or language big data model to generate multiple risk parameters. Finally, the first optimized data, the second data, and the risk parameters are used to train an orthodontic risk prediction language big data model to predict the orthodontic risk level. This method can comprehensively assess the disease itself and the risk of treating the disease. By automating the orthodontic risk assessment through deep learning and language big data models, it can dynamically assess orthodontic risk according to different treatment plans, effectively improving the accuracy of orthodontic risk assessment and reducing reliance on physician experience. Attached Figure Description
[0051] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.
[0052] Figure 1 A flowchart illustrating an automated orthodontic risk assessment method proposed in an embodiment of this application is shown.
[0053] Figure 2 A schematic diagram illustrating the basic patient information presented in an embodiment of this application is shown.
[0054] Figure 3 The illustration shows the intended representation of a record according to an embodiment of this application.
[0055] Figure 4 This paper illustrates another flowchart of the automated orthodontic risk assessment method proposed in the embodiments of this application. Detailed Implementation
[0056] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. This application can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, unless otherwise specified, the following embodiments and features in the embodiments can be combined with each other.
[0057] Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0058] In current technologies, commonly used orthodontic risk assessment methods primarily rely on the doctor's subjective judgment, lacking suitable objective quantitative tools. Treatment risk is closely related to patient diagnosis, treatment plan, and compliance, and is influenced by patient condition, doctor's skill level, and medical environment. Patient-related factors include genetic factors, dental anatomy, demographic characteristics, malocclusion factors, root canal treatment, systemic medical history, short root abnormalities, and periodontal disease. Treatment-related factors include biomechanical factors, appliances and auxiliary devices, accelerated tooth movement, treatment timing, tooth movement distance and direction, treatment duration, and regular follow-up visits. Doctor-patient cooperation includes oral hygiene, regular follow-up visits, and doctor's skill level. Due to the complexity of influencing factors and the subjectivity of doctor's judgment, assessment results from different doctors can vary, leading to insufficient objectivity and making it difficult to provide an effective reference for orthodontic treatment and plan selection.
[0059] However, in clinical practice, oral clinical examination data typically includes patient questionnaires, clinical examinations, intraoral 3D scans, CBCT scans, panoramic radiographs, lateral radiographs, and facial and intraoral photographs. This data consists of text, two-dimensional images, and 3D data. While the abundance of data is beneficial for the accurate description of patient characteristics, the different data dimensions increase the processing difficulty. Therefore, reducing the data dimensionality and preprocessing patient data is necessary.
[0060] Therefore, this application proposes an automated orthodontic risk assessment method and apparatus. By using deep learning and a large language model to automate the orthodontic risk assessment, it can dynamically assess orthodontic risks according to different treatment plans, effectively improving the accuracy of orthodontic risk assessment. The following is a detailed description of the method.
[0061] Please refer to Figure 1 , Figure 1 The diagram illustrates a flowchart of an automated orthodontic risk assessment method proposed in an embodiment of this application. The method includes the following steps:
[0062] Step S1: Obtain the clinical data of the patient to be evaluated and optimize the clinical data to obtain the first optimized data.
[0063] Step S2: Analyze the first optimized data using neural networks, knowledge graph networks, or large language models to obtain the second data.
[0064] Step S3: Use knowledge graph networks or language large models to process the second data to generate risk parameters.
[0065] Step S4: Use the first optimized data, the second data, and the risk parameters to train a large language model for orthodontic risk prediction in order to predict the orthodontic risk level.
[0066] The clinical data includes questionnaires, clinical examination data, record sheets, and imaging data. It may also include examination results, intraoral scan models, and X-rays. The automated orthodontic risk assessment method proposed in this application first acquires the clinical data of the patient to be assessed and optimizes it to obtain first optimized data. Then, it analyzes the first optimized data using neural networks, knowledge graph networks, or language large models to obtain second data. The second data is then processed using knowledge graph networks or language large models to generate risk parameters. Finally, the first optimized data, the second data, and the risk parameters are used to train an orthodontic risk prediction language large model to predict the orthodontic risk level.
[0067] It can comprehensively assess the disease itself and the risks of treating the disease. Through deep learning and large models, it automates the orthodontic risk assessment and can dynamically assess orthodontic risks according to different treatment plans. This effectively improves the accuracy of orthodontic risk assessment, reduces reliance on doctors' experience, and is conducive to large-scale promotion and application.
[0068] Step S1, which involves acquiring clinical data of the patient to be evaluated and optimizing the clinical data to obtain first optimized data, includes:
[0069] Record clinical examination information and questionnaire survey information, and use database comparison to extract the first abnormal value in the clinical examination information and questionnaire survey information;
[0070] The neural network model is used to locate key anatomical points, segment key anatomical structures and lesions from imaging data. The imaging data includes two-dimensional and three-dimensional images. Two-dimensional images include lateral radiographs, panoramic radiographs, facial radiographs and intraoral radiographs. Three-dimensional images include intraoral scanning models, CBCT, MRI and facial scanning data.
[0071] The key anatomical points were measured and analyzed, and the results were compared with the database to obtain the second abnormal value;
[0072] The disease type is identified by comparing key anatomical structures and lesions with a database or classifying them using neural networks.
[0073] The first optimized data is obtained by cross-validating the data based on the first abnormal value, the second abnormal value, and the disease type.
[0074] The first optimized data is generated based on the patient's multi-dimensional clinical examination data. The method of converting multi-dimensional data into one-dimensional text data can be achieved using a neural network with fixed points, or other methods. This application embodiment does not limit this approach. The first optimized data mainly involves converting multi-dimensional data into text / numerical data through various processing steps, so that graph networks or large language models can be used subsequently.
[0075] In addition, further cross-validation of the first optimized data is required. Since the same disease will exhibit characteristics in different examinations, the first optimized data obtained from different clinical examination data usually needs to be cross-validated, and contradictions may occur in the process. Therefore, it is necessary to verify and analyze the first abnormal value, key dissection points, key anatomical structures and lesions, the second abnormal value, and the disease type, extract feature data to form a diagnosis and match treatment plans. Post-processing can be performed using existing relevant data feature methods or the implementation methods of other embodiments of this application, ultimately obtaining the first optimized data.
[0076] Clinical data includes clinical examination information, questionnaire data, and imaging data. Structured diagnostic questionnaires and structured clinical examination record forms were designed, and basic information of the patients to be evaluated was also included. Data was entered through a selection process; the database retrieved risk-related examination results from a structured database. Basic information of the patients to be evaluated included: patient age, gender, race, history of oral diseases, family history of genetic diseases, orthodontic needs, oral hygiene, dental and oral soft tissue health status, oral function and movement, facial aesthetics, and other indicators. Figure 2 This illustration shows a schematic diagram of the patient's basic information as presented in an embodiment of this application. Figure 3 This illustrates the intended representation of a record according to an embodiment of this application. For example, a questionnaire survey is shown below:
[0077] Chief complaint: The issue that the patient is most concerned about, but the patient as a whole should not be ignored;
[0078] General / Dental medical history: trauma, allergies, long-term medications (glucocorticoids, bisphosphonates), systemic diseases (diabetes, rheumatoid arthritis, heart disease, blood disorders, epilepsy, contraindications to tooth extraction, mental illness, metabolic diseases, hyperthyroidism, rickets);
[0079] Growth and development assessment: height and weight, secondary sexual characteristics, lateral cervical spine radiographs (CVMs cervical spine staging);
[0080] Social and behavioral evaluation: whether the parent or child requests treatment, level of cooperation, and dietary habits (e.g., smoking / drinking?).
[0081] In one possible embodiment, Table 1 shows the clinical examination data presented in the embodiments of this application:
[0082] Table 1
[0083]
[0084] In one possible embodiment, the imaging data may include 2D images, which are converted into textual data of measurement indicators using neural network analysis of panoramic radiographs, lateral radiographs, facial and intraoral photographs. Cascaded neural networks or other neural networks for key point localization are used to locate key anatomical landmarks in the lateral radiographs and photographs. Then, corresponding measurement items are calculated based on the coordinates of the key points, and the issues reflected in the lateral radiographs and lateral photographs are converted into textual descriptions based on the clinical significance of each measurement item.
[0085] Panoramic radiographs: dental diseases (impacted teeth, supernumerary teeth, missing teeth, caries, crown restorations, removable dentures, implants), pulp and periapical diseases (pulpitis, periapical periodontitis, post-root canal treatment, pulp resorption), periodontal diseases (gingivitis, periodontitis), joint diseases, maxillary sinusitis, abnormal bone density, tumors and cystic lesions.
[0086] Lateral radiographs: skeletal indices, dental indices, and soft tissue indices.
[0087] Facial photos: three courts and five eyes, facial symmetry, smile aesthetics evaluation, and profile aesthetics indicators.
[0088] Intraoral photographs: changes in tooth alignment, color, and texture; frenulum of the lips / tongue; and mucosal lesions.
[0089] A cascaded classification network was used to identify oral problems presented in panoramic and intraoral radiographs and to convert them into textual descriptions.
[0090] The imaging data may also include 3D images, which use neural network analysis to convert intraoral 3D scans and CBCT into text data; 3D key point localization algorithms can be used to calculate relevant indicators, and then the measurement results are compared with normal values to determine whether they are normal.
[0091] Intraoral 3D scan: midline, overbite, canine-molar relationship, crowding, Bolton index, Spee curve.
[0092] CBCT: lateral width, relationship between tooth root and alveolar bone, dental caries-pulp-periapical diseases, jawbone diseases, joint diseases, maxillary sinus diseases, and airway analysis.
[0093] Step S2, which involves analyzing the first optimized data using neural networks, knowledge graph networks, or large language models to obtain the second data, includes:
[0094] The diagnostic results are obtained by comprehensively analyzing the first optimized data using a neural network;
[0095] The diagnostic results can be processed using knowledge graph networks or large language models to generate multiple treatment plans.
[0096] The second set of data includes diagnostic results and treatment plans. First, the measurement items, values, and clinical significance are extracted from the first set of optimized data. The measurement items are derived from commonly used measurements in the orthodontic industry, and the clinical significance comes from comparisons between the measured values and standard values. Next, the measured values, measurement items, and clinical significance are analyzed to form feature data. Based on the clinical significance, positive or negative analysis results are extracted.
[0097] A knowledge base is then built based on the feature data, and the feature data is post-processed to generate a list of clinical diagnoses and problems. Different feature data should corroborate each other in the diagnosis of the same disease. However, due to limitations in data collection, analysis, or the data itself, different conclusions may occur. Therefore, it is necessary to comprehensively consider and determine the correct results, prioritize data from different sources, and automatically exclude erroneous data to build the list of diagnoses and problems.
[0098] Finally, based on the list of clinical diagnoses and problems, a corresponding treatment plan is matched for each diagnosis and problem. Since different diseases have different treatment methods, one disease may correspond to multiple treatment approaches, each of which is associated with risk factors. This implementation example exhaustively lists the treatment approaches corresponding to each diagnosis. The knowledge graph network automatically selects a treatment plan based on the inherent logic of the diagnostic results and the treatment goals of doctors and patients, allowing doctors and patients to adjust the treatment plan according to individual needs.
[0099] For each clinical diagnosis and problem list, a corresponding treatment plan is matched. Since different diseases have different treatments, a single disease may correspond to multiple treatment methods, and each treatment method is associated with risk factors, a knowledge graph network is used to automatically select a treatment plan based on the internal logic of the diagnostic results and the treatment goals of both doctors and patients. This allows doctors and patients to adjust the treatment plan according to individual needs.
[0100] Step S3, which involves processing the second data using a knowledge graph network or a large language model to generate risk parameters, includes:
[0101] Multiple treatment options are processed using knowledge graph networks or large language models to generate a first risk parameter that corresponds one-to-one with each treatment option;
[0102] The diagnostic results are processed using knowledge graph networks or large language models to generate a second risk parameter;
[0103] Knowledge graph networks or language large models are used to process the basic information of patients to be assessed and generate third risk parameters.
[0104] The basic information of the patient to be evaluated includes, but is not limited to, gender, age, race, education level, height, and weight. First, the diagnostic results and treatment plan are extracted from the secondary data. Then, a knowledge graph network or language model is used to automatically analyze the relevant risks of the diagnostic results and treatment plan. These risks are divided into three parts, each related to the disease diagnosis and the implementation of the treatment plan. These include: tooth demineralization / caries, root resorption, tooth loosening, space closure insufficiency, bone fenestration / cracking, alveolar bone resorption, periodontitis, braces face, temporomandibular joint disorder, black triangle, relapse after treatment, and psychological disorders.
[0105] Because the risk is related to the patient's basic information, such as the risk of the same disease being different in patients of different ages or genders, and the risk is also related to the patient's compliance, a third risk parameter was introduced.
[0106] Finally, the corresponding risk factors (first risk parameter, second risk parameter, and third risk parameter) are analyzed. Risk factors are private factors that affect the probability of risk occurrence, including patient-related factors: patient genetic factors, dental anatomy, demographic characteristics, malocclusion factors, root canal treatment, systemic medical history, short root abnormalities, and periodontal disease; treatment-related factors: biomechanical factors, orthodontic appliances and auxiliary devices, accelerated tooth movement, timing of treatment, distance and direction of tooth movement, and duration of treatment; and doctor-patient cooperation factors: oral hygiene, regular follow-up visits, and doctor's skill level.
[0107] Step S4, which involves training a large-scale language model for orthodontic risk prediction using the first optimized data, the second data, and risk parameters to predict the orthodontic risk level, includes:
[0108] The orthodontic risk prediction language model was trained using the first optimized data, the second data, and risk parameters.
[0109] Predict orthodontic risk levels using a trained orthodontic risk prediction language model.
[0110] Based on the second data, further analysis of orthodontic risks and associated risk factors is conducted, and orthodontic risk assessment is performed based on the first optimized data, the second data, and risk parameters. It should be understood that orthodontic risk assessment has not yet undergone systematic evaluation. The automated orthodontic risk assessment method proposed in this application can effectively integrate various clinical indicators, improve the accuracy of orthodontic risk assessment, and does not rely on the doctor's experience judgment, which is conducive to large-scale promotion and application.
[0111] Based on risk parameters, since each risk is related to a risk factor, and the risk factor determines the probability of the risk occurring, a knowledge graph and expert system are built based on the risk factors and their probabilities. The probability of each risk is determined based on the risk factors. For the risk factor related to doctor-patient cooperation, manual adjustment of parameters is allowed. The probability of each risk is determined using neural network learning. The specific learning method follows common neural network training methods, using large datasets and large models for multi-parameter training.
[0112] Various risk parameters are superimposed to form a total risk parameter. The neural network integrates these risk parameters using a large neural network model. The training method follows the same method as the large model training method. The total risk parameter is combined with the first and second optimized data to train a large prediction language model. The trained orthodontic risk prediction language model is then used to predict the orthodontic risk level and assess orthodontic risk.
[0113] The risk prediction algorithm is implemented by collecting multimodal big data on orthodontic patients (basic information, clinical examination, psychological state, treatment plan, and treatment results), analyzing and acquiring the data, and having experts assess the risk level. The risk levels related to the re-diagnosis and treatment plan, along with the risk levels assessed by the patients (based on basic patient information and expert evaluation), are used to train a large-scale language model for orthodontic risk prediction (this large-scale language model is an improved version of large models such as GPT, PaLM, and Claude, after pre-training). Finally, the language model predicts the orthodontic risk level.
[0114] In one possible embodiment, please refer to Figure 4 , Figure 4 This paper illustrates another flowchart of the automated orthodontic risk assessment method proposed in this application. Addressing the shortcomings of existing automated orthodontic risk assessment systems, it employs graph network prediction based on diagnostic results and a problem list. It uses structured data tables to collect treatment goals from questionnaires and clinical examinations. A 2D cascaded neural network is used to locate key anatomical points, and a cascaded classification network identifies diseases to obtain basic information. A 3D neural network locates key anatomical landmarks from CBCT and intraoral scans. Through key point measurement and analysis, image information from different dimensions is converted into textual information to obtain diagnostic results and a problem list.
[0115] In diagnosis and treatment planning, misdiagnosis or missed diagnosis often occurs when a single examination method is used to diagnose a disease. To avoid this, multiple different examinations can interpret the same disease from different perspectives. Since the different focuses of different examinations may lead to contradictory conclusions, a relevant evidence prioritization strategy was developed to resolve diagnostic conflicts. Orthodontic problems are categorized into 20 major groups, each group being diagnosed by 1-4 different examinations. In case of conflict, the diagnostic priority strategy is used to determine the order: evidence priority ① > ② > ③ > ④, where ① scores 4 points, ② scores 3 points, ③ scores 2 points, and ④ scores 1 point. When multiple diagnostic results conflict, the diagnosis with the higher total score is considered valid. Table 2 shows the evidence priorities for different imaging data reflecting the same clinical problem:
[0116] Table 2
[0117]
[0118] Diagnosis is the foundation of treatment plan design. After confirming the diagnosis, a treatment plan is designed based on the patient's complaints. This process is usually designed by the doctor based on experience, but it can also be automated using artificial neural networks, with the doctor making modifications and confirmations. The treatment plan includes treatment goals, treatment duration, orthodontic appliances and accessories used, and expected results. Multiple treatment options may exist; for example, crowding requires creating space, which can be achieved through interproximal reduction, tooth extraction, molar repositioning, arch expansion, and other methods. Therefore, both the patient and the doctor need to understand the risks and benefits of each method to make a suitable choice.
[0119] When predicting risks in orthodontic treatment, the risks are not only related to the patient's own disease status (diagnosis), but also closely related to the patient's level of cooperation and the choice of treatment plan. Therefore, it is necessary to analyze the diagnosis, treatment plan and the patient's level of cooperation to extract risk factors.
[0120] Each risk factor is quantitatively assessed based on the problem list, treatment plan, and level of cooperation in the dental data. The assessment result is a risk coefficient. The quantitative assessment method is assigned by the doctor or uses deep learning. Convolutional neural networks can easily fit a non-linear curve to evaluate the quantitative coefficient of each risk factor.
[0121] When conducting the analysis and assessment, each risk factor is quantitatively evaluated. An orthodontic risk prediction algorithm is designed by combining prior dental knowledge. The method is: quantitative assessment result of the problem list × treatment plan × level of cooperation. It should be understood that orthodontic risk assessment has not yet been systematically evaluated. In this application embodiment, a novel risk assessment method is proposed, effectively integrating various clinical indicators, improving the accuracy of orthodontic risk assessment, and not relying on the doctor's experience judgment, which is conducive to large-scale promotion and application.
[0122] This method is applicable to personalized risk prediction for patients of different ages and types, and can update risk assessment results based on different treatment plans, changes in treatment goals, and changes in compliance. It provides a reference for judging the merits of orthodontic treatment plans.
[0123] Compared with the prior art, the embodiments of this application have the following beneficial effects:
[0124] First, it fills the gap in current orthodontic risk assessment methods.
[0125] Second, the problem of automatic orthodontic risk assessment is transformed into problems of automatic diagnosis, automatic treatment plan design, automatic risk item assessment, and risk factor association; it allows doctors and patients to adjust non-fixed items such as treatment plan design and doctor-patient cooperation, making the design flexible and solving the problem of orthodontic risk prediction to a certain extent in cases of different treatment plans for the same disease and different patient cooperation levels for the same treatment;
[0126] Third, in terms of data processing, it proposes to automatically analyze multidimensional data through neural networks and convert it into one-dimensional structured text data, which reduces the difficulty of subsequent data processing; and to convert one-dimensional structured text data into graph structure, using knowledge graphs to solve complex interrelationship problems of "diagnosis-treatment plan-risk item-risk factor" and predict the final risk.
[0127] Fourth, the impact of different risk factors on orthodontic treatment risks was quantified through neural networks and big data.
[0128] The following describes a possible implementation of an automated orthodontic risk assessment device, which performs the various execution steps and corresponding technical effects of the automated orthodontic risk assessment method shown in the above embodiments and possible implementations. The device includes:
[0129] The acquisition module is used to acquire the clinical data of the patient to be evaluated and optimize the clinical data to obtain the first optimized data;
[0130] The analysis module is used to analyze the first optimized data using neural networks, knowledge graph networks, or large language models to obtain the second data;
[0131] The processing module is used to process the second data using knowledge graph networks or large language models to generate risk parameters.
[0132] The training module is used to train a large language model for orthodontic risk prediction using the first optimized data, the second data, and risk parameters to predict the orthodontic risk level.
[0133] In one possible implementation, the acquisition module is further configured to:
[0134] Record clinical examination information and questionnaire survey information, and use database comparison to extract the first abnormal value in the clinical examination information and questionnaire survey information;
[0135] The neural network model is used to locate key anatomical points, segment key anatomical structures and lesions from imaging data. The imaging data includes two-dimensional and three-dimensional images. Two-dimensional images include lateral radiographs, panoramic radiographs, facial radiographs and intraoral radiographs. Three-dimensional images include intraoral scanning models, CBCT, MRI and facial scanning data.
[0136] The key anatomical points were measured and analyzed, and the analysis results were compared with the database to obtain the second abnormal value.
[0137] Disease types can be identified by comparing key anatomical structures and lesions with databases or using neural network classification.
[0138] The first optimized data is obtained by cross-validating the data based on the first abnormal value, the second abnormal value, and the disease type.
[0139] In one possible implementation, the analysis module is further configured to:
[0140] The diagnostic results are obtained by comprehensively analyzing the first optimized data using a neural network;
[0141] The diagnostic results can be processed using knowledge graph networks or large language models to generate multiple treatment plans.
[0142] In one possible implementation, the processing module is further configured to:
[0143] Multiple treatment options are processed using knowledge graph networks or large language models to generate a first risk parameter that corresponds one-to-one with each treatment option;
[0144] The diagnostic results are processed using knowledge graph networks or large language models to generate a second risk parameter;
[0145] The basic information of the patients to be assessed is processed using knowledge graph networks or language large models to generate third risk parameters. The basic information of the patients to be assessed includes, but is not limited to, gender, age, race, education level, height, and weight.
[0146] In one possible implementation, the training module is further configured to:
[0147] The orthodontic risk prediction language model was trained using the first optimized data, the second data, and risk parameters.
[0148] Predict orthodontic risk levels using a trained orthodontic risk prediction language model.
[0149] Therefore, the automated orthodontic risk assessment method and device disclosed in this application can comprehensively assess the disease itself and the risks of treating the disease. By automating the orthodontic risk assessment through deep learning and language large models, it can dynamically assess orthodontic risks according to different treatment plans, effectively improving the accuracy of orthodontic risk assessment and reducing reliance on doctors' experience.
[0150] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.
Claims
1. An automated orthodontic risk assessment method, characterized in that, The method includes: Acquire clinical data of the patient to be evaluated and optimize the clinical data to obtain first optimized data; The second data is obtained by analyzing the first optimized data using neural networks, knowledge graph networks, or large language models; Risk parameters are generated by processing the second data using knowledge graph networks or large language models respectively. A large language model for orthodontic risk prediction was trained using the first optimized data, the second data, and risk parameters to predict the orthodontic risk level.
2. The automated orthodontic risk assessment method as described in claim 1, characterized in that, The clinical data includes clinical examination information, questionnaire survey information, and imaging data. The steps of obtaining the clinical data of the patient to be evaluated and optimizing the clinical data to obtain the first optimized data include: Record clinical examination information and questionnaire survey information, and use database comparison to extract the first abnormal value in the clinical examination information and questionnaire survey information; The neural network model is used to locate key anatomical points, segment key anatomical structures and lesions from imaging data. The imaging data includes two-dimensional images and three-dimensional images. The two-dimensional images include lateral radiographs, panoramic radiographs, facial photographs and intraoral photographs. The three-dimensional images include intraoral scanning models, CBCT, MRI and facial scanning data. The key anatomical points were measured and analyzed, and the analysis results were compared with the database to obtain the second abnormal value. Disease types can be identified by comparing key anatomical structures and lesions with databases or using neural network classification. The first optimized data is obtained by cross-validating the data based on the first abnormal value, the second abnormal value, and the disease type.
3. The automated orthodontic risk assessment method as described in claim 1, characterized in that, The second data includes diagnostic results and treatment plans. The steps of analyzing the first optimized data using neural networks, knowledge graph networks, or large language models to obtain the second data include: The diagnostic results are obtained by comprehensively analyzing the first optimized data using a neural network; The diagnostic results can be processed using knowledge graph networks or large language models to generate multiple treatment plans.
4. The automated orthodontic risk assessment method as described in claim 3, characterized in that, The steps for processing the second data using knowledge graph networks or large language models to generate risk parameters include: Multiple treatment options are processed using knowledge graph networks or large language models to generate a first risk parameter that corresponds one-to-one with each treatment option; The diagnostic results are processed using knowledge graph networks or large language models to generate a second risk parameter; A third risk parameter is generated by processing the basic information of the patient to be assessed using a knowledge graph network or a language large model. The basic information of the patient to be assessed includes, but is not limited to, gender, age, race, education level, height, and weight.
5. The automated orthodontic risk assessment method as described in claim 1, characterized in that, The steps for training a large-scale language model for orthodontic risk prediction using the first optimized data, the second data, and risk parameters to predict the level of orthodontic risk include: The orthodontic risk prediction language model was trained using the first optimized data, the second data, and risk parameters. Predict orthodontic risk levels using a trained orthodontic risk prediction language model.
6. An automated orthodontic risk assessment device, characterized in that, The device includes: The acquisition module is used to acquire the clinical data of the patient to be evaluated and optimize the clinical data to obtain the first optimized data; The analysis module is used to analyze the first optimized data using neural networks, knowledge graph networks, or large language models to obtain the second data; The processing module is used to process the second data using knowledge graph networks or large language models to generate risk parameters. The training module is used to train a large language model for orthodontic risk prediction using the first optimized data, the second data, and risk parameters to predict the orthodontic risk level.
7. The automated orthodontic risk assessment device as described in claim 6, characterized in that, The acquisition module is also used for: Record clinical examination information and questionnaire survey information, and use database comparison to extract the first abnormal value in the clinical examination information and questionnaire survey information; The neural network model is used to locate key anatomical points, segment key anatomical structures and lesions from imaging data. The imaging data includes two-dimensional images and three-dimensional images. The two-dimensional images include lateral radiographs, panoramic radiographs, facial photographs and intraoral photographs. The three-dimensional images include intraoral scanning models, CBCT, MRI and facial scanning data. The key anatomical points were measured and analyzed, and the analysis results were compared with the database to obtain the second abnormal value. Disease types can be identified by comparing key anatomical structures and lesions with databases or using neural network classification. The first optimized data is obtained by cross-validating the data based on the first abnormal value, the second abnormal value, and the disease type.
8. The automated orthodontic risk assessment device as described in claim 6, characterized in that, The analysis module is also used for: The diagnostic results are obtained by comprehensively analyzing the first optimized data using a neural network; The diagnostic results can be processed using knowledge graph networks or large language models to generate multiple treatment plans.
9. The automated orthodontic risk assessment device as described in claim 8, characterized in that, The processing module is also used for: Multiple treatment options are processed using knowledge graph networks or large language models to generate a first risk parameter that corresponds one-to-one with each treatment option; The diagnostic results are processed using knowledge graph networks or large language models to generate a second risk parameter; A third risk parameter is generated by processing the basic information of the patient to be assessed using a knowledge graph network or a language large model. The basic information of the patient to be assessed includes, but is not limited to, gender, age, race, education level, height, and weight.
10. The automated orthodontic risk assessment device as described in claim 6, characterized in that, The training module is also used for: The orthodontic risk prediction language model was trained using the first optimized data, the second data, and risk parameters. Predict orthodontic risk levels using a trained orthodontic risk prediction language model.