An oral health abnormality early warning and remote inquiry linkage system and use method

Abstract

The present application relates to the field of oral health management, and more particularly to an oral health abnormality early warning and remote inquiry linkage system and use method, comprising an AI detection data receiving module for localizing sensitive areas; a risk assessment engine for multidimensional calculation of risk level; an early warning generation module for generating differentiated early warning; a service provider matching engine for matching specialist doctors / counselors; a consultation single generation module for generating templates following data minimization. The present application automatically triggers the consultation process after AI detects abnormalities, eliminates information gaps, reduces user burden by reducing structured consultation forms, while protecting privacy; enables automatic synchronization of doctor recommendations to observation plans, achieving a diagnosis and treatment closed loop; through a hierarchical response mechanism, both urgent problem handling and user experience are considered, thereby achieving differentiated early warning; HIPAA / individual protection law is followed throughout, data sharing is controllable, transparent and compliant, ensuring compliant data sharing.

Description

Technical Field

[0001] This invention relates to the field of oral health management, and in particular to an oral health abnormality early warning and remote consultation linkage system and its usage method. Background Technology

[0002] Oral health issues have become a significant area of ​​chronic diseases affecting public quality of life. In recent years, technologies related to oral health management have gradually developed, for example:

[0003] Existing conditions, such as the oral health assessment method and system disclosed in announcement number CN120126769A, which is based on an artificial intelligence-assisted model for oral health assessment, can compare and display the shortcomings of artificial intelligence diagnosis in teaching, and continuously retrain the database and auxiliary model during the diagnostic process, so that humans and machines complement each other to improve the effectiveness of oral health assessment. It achieves oral health scoring and risk warning through image recognition, but does not realize the linkage with medical services after detection, resulting in a "detection-reporting" gap; it supports patients to initiate text-based consultations, but requires manual entry of symptoms, relies on subjective descriptions from patients, resulting in low consultation efficiency, and does not automatically connect with oral imaging data; it recommends doctors based on general symptoms, without combining specific dental abnormality types and severity for specialist matching. The following problems are commonly found:

[0004] The detection and consultation processes are disconnected, lacking an automatic triggering and linkage mechanism;

[0005] The doctor's advice could not be automatically synchronized to the family observation plan, resulting in a missing closed loop.

[0006] Data sharing lacks sophisticated authorization and automatic expiration mechanisms, resulting in high privacy compliance risks.

[0007] The risk warning system is too simplistic and lacks differentiated, tiered responses. Summary of the Invention

[0008] The purpose of this invention is to solve the problems in the prior art of triggering graded early warning based on the condition, intelligently matching specialist doctors, and achieving seamless linkage between test data and consultation process. Therefore, this invention proposes an oral health abnormality early warning and remote consultation linkage system and its usage method.

[0009] To achieve the above objectives, the present invention adopts the following technical solution:

[0010] An oral health abnormality early warning and remote consultation linkage system includes an AI detection data receiving module, which is used for localized processing of sensitive areas;

[0011] A risk assessment engine that calculates risk levels across multiple dimensions;

[0012] The early warning generation module is used to generate differentiated early warnings;

[0013] A service provider matching engine that matches specialist doctors / consultants;

[0014] The consultation form generation module is used to generate templates following the data minimization principle.

[0015] The online consultation interface module is used to connect with compliant service providers and verify authorization.

[0016] The service feedback parsing module is used to extract key suggestions based on NLP.

[0017] The observation plan synchronization module is used to update the observation targets / cycles;

[0018] The data authorization management module is used to manage authorization status and ensure compliance.

[0019] Preferably, the AI ​​detection data receiving module first performs tooth instance segmentation and localization on the input oral image using a target detection algorithm, identifies the anatomical location of each tooth, and outputs a health score of 0-100 for each tooth based on the image features of the enamel, dentin, and gingival region through a regression model.

[0020] Preferably, the risk assessment engine proposes a total score of 0-100 based on the health score, and the early warning generation module provides a corresponding level reminder based on the total score.

[0021] Preferably, the service provider's matching engine matches doctors in the corresponding departments based on the total score and corresponding level.

[0022] Preferably, the user confirms the data sharing authorization through the consultation form generation module, generates a HIPAA compliant electronic authorization letter, and establishes a graphic consultation.

[0023] Preferably, doctors provide suggestions through the service feedback analysis module and update the observation plan through the observation plan synchronization module.

[0024] Preferably, the authorization expires automatically 24 hours after the consultation ends, and the service provider is notified to delete the data through the data authorization management module.

[0025] The above-mentioned method of using an oral health abnormality early warning and remote consultation linkage system includes the following steps:

[0026] Step S1, anomaly identification and data reception;

[0027] Step S2, intelligent risk level assessment;

[0028] Step S3 automatically triggers the online consultation service process;

[0029] Step S4, Service feedback received;

[0030] Step S5: Observe the plan to update automatically.

[0031] Preferably, step S1, anomaly identification and data reception, refers to receiving oral health test data and identifying abnormal teeth and their types based on AI analysis results;

[0032] Step S2, intelligent risk level assessment, refers to assessing the risk level by combining the anomaly type and severity indicators, and generating corresponding early warning information. The risk levels are divided into emergency level, attention level and monitoring level.

[0033] Step S3, automatically triggering the online consultation service process, means that the online consultation process is automatically triggered when the risk level reaches the concern level or above;

[0034] Step S4, receiving service feedback, refers to receiving the analysis results and suggestions submitted by the service provider based on the structured consultation form; if connecting with a qualified medical institution, it refers to receiving the doctor's diagnosis and treatment suggestions.

[0035] In step S5, the observation plan is automatically updated to analyze treatment suggestions, extract key elements and synchronize them to the family member observation plan, and automatically adjust the observation targets and detection cycle.

[0036] Preferably, the online consultation process is as follows:

[0037] In phase t1, the doctor's specialty mapping table is queried based on the anomaly type to match the corresponding specialist doctor / consultant;

[0038] In phase t2, a structured consultation form is automatically generated based on the detection data;

[0039] In stage t3, a consultation recommendation and data sharing authorization request are pushed to the user. After the user confirms, a consultation connection is established.

[0040] In stage t4, the system can connect with compliant internet hospital remote consultation service providers or provide non-medical oral health consultation services.

[0041] Compared with the prior art, the present invention has the following advantages:

[0042] 1. This invention automatically triggers a consultation process after detecting anomalies using AI, eliminating information gaps and achieving seamless connection between detection and treatment.

[0043] 2. This invention reduces the burden of filling out structured consultation forms for users, while protecting privacy and helping to improve consultation efficiency.

[0044] 3. This invention matches specialist doctors based on abnormal types, improving service professionalism and enabling accurate doctor matching.

[0045] 4. This invention enables doctors' recommendations to be automatically synchronized with the observation plan, ensuring consistency between home monitoring and professional advice, and achieving a closed loop in diagnosis and treatment.

[0046] 5. This invention achieves differentiated early warning by taking into account both emergency problem handling and user experience through a hierarchical response mechanism.

[0047] 6. This invention complies with HIPAA / Individual Property Protection Act throughout the process, ensuring that data sharing is controllable, transparent, and compliant, and guaranteeing compliant data sharing. Attached Figure Description

[0048] Figure 1 This is a schematic diagram of the structure of an oral health abnormality early warning and remote consultation linkage system proposed in this invention;

[0049] Figure 2 This is a schematic diagram illustrating the steps of using the oral health abnormality early warning and remote consultation linkage system proposed in this invention;

[0050] Figure 3 This is a schematic diagram illustrating the multi-dimensional scoring and judgment rules of an oral health abnormality early warning and remote consultation linkage system and its usage method proposed in this invention.

[0051] Figure 4 This is a schematic diagram of the matching process and consultation form structure of an oral health abnormality early warning and remote consultation linkage system and its usage method proposed in this invention;

[0052] Figure 5 This is a schematic diagram illustrating the analytical and synchronization process of an oral health abnormality early warning and remote consultation linkage system and its usage method proposed in this invention;

[0053] Figure 6 This diagram illustrates the data source layer (de-sensitized), core business layer (encrypted transmission), and data storage layer (encrypted) of the oral health abnormality early warning and remote consultation linkage system and its usage method proposed in this invention.

[0054] In the diagram: 1. AI detection data receiving module; 2. Risk assessment engine; 3. Early warning generation module; 4. Service provider matching engine; 5. Consultation form generation module; 6. Online consultation interface module; 7. Service feedback parsing module; 8. Observation plan synchronization module; 9. Data authorization management module. Detailed Implementation

[0055] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0056] Reference Figures 1-6A system for linking oral health abnormality early warning and remote consultation, and its usage method, includes an AI detection data receiving module 1, a risk assessment engine 2, an early warning generation module 3, a service provider matching engine 4, a consultation form generation module 5, an online consultation interface module 6, a service feedback parsing module 7, an observation plan synchronization module 8, and a data authorization management module 9. The modules communicate with each other using RESTful APIs, message queues (Kafka / RabbitMQ), or event buses.

[0057] AI detection data receiving module 1 is used for localized processing of sensitive areas. It is configured to receive digital oral images captured by front-end image acquisition devices such as intraoral scanners, intraoral endoscopes, or smartphone cameras. This module has a built-in semantic segmentation model based on convolutional neural networks, which can extract regions of interest from the received multimodal image stream. Specifically, it is used for localized processing of sensitive areas, that is, automatically segmenting and cropping non-target soft tissue backgrounds such as lips, cheeks, and tongue, and de-identifying and retaining only local image data of teeth and periodontal tissues to achieve minimal data acquisition.

[0058] It should be noted that the AI ​​model training data comes from anonymized data collected in cooperation with three top-tier hospitals.

[0059] Risk assessment engine 2 is used to calculate risk levels across multiple dimensions. It is configured as the core of a multi-level, configurable clinical decision support system for this purpose. The engine synthesizes scores from the following dimensions using a weighted summation algorithm: the health score of a single tooth obtained from the AI-generated data receiving module 1, a health trend score based on longitudinal historical comparison, a severity score for the abnormality type, and an urgency score for disease progression attributes. The specific calculation rules of this engine are defined based on the following JSON structured configuration document:

[0060] The json function defines a riskAssessmentRules system with the following parameters: riskAssessmentRules: ...

[0061] }, "urgencyScore": { "acute": {"score": 90}, "chronic_progressive": {"score": 50}}, "levelThresholds": { "urgent": {"min": 80, "color": "#EF4444"}, "attention": {"min": 60, "color": "#EAB308"}, "monitor": {"min": 0,"color": "#6B7280"}}}}

[0062] It should be noted that the severityScore dictionary defines static scores corresponding to different abnormality types and their severity. For example, "early_caries" (early caries) has a score of 60 for "mild". The urgencyScore dictionary defines dynamic scores for disease course attributes. For example, "acute" corresponds to 90 points. Finally, the risk assessment engine 2 matches the normalized scores with the threshold ranges in levelThresholds to calculate a comprehensive score of 0-100 and determine its risk level.

[0063] Risk assessment engine 2 uses a weighted summation formula:

[0064] TotalScore = w1 * severityScore + w2 * urgencyScore + w3 * trendScore

[0065] The weights w1, w2, and w3 are obtained through training on a clinical dataset (e.g., w1=0.5, w2=0.3, w3=0.2), and the final total score is normalized to the 0-100 range.

[0066] The early warning generation module 3 is used to generate differentiated early warnings. It communicates and is coupled with the risk assessment engine 2 to generate differentiated early warnings. This module calls different human-computer interaction interface templates and message push protocols based on the risk level and urgency. Its core logic is an event-driven responder that can map "urgent level" to a highly intrusive notification including sound, vibration, and a full-screen pop-up; "attention level" to a standard notification bar message and an in-app red dot marker; and "monitoring level" to silent background recording without actively contacting the user.

[0067] Service provider matching engine 4 is used to match specialist doctors / consultants. Based on anomaly type and risk level, it performs semantic retrieval and matching within the doctor resource knowledge graph to match specialist doctors or health consultants. This engine is based on rule parsing and recommendation priority ranking algorithms. The specific calculation logic is configured in JSON as follows:

[0068] The json file defines a table with the following parameters: `{ “doctorMatchingRules”: { “mappingTable”: [ { “abnormalityType”:“early_caries”, “recommendedDept”: [“Pediatric Dentistry”], “specialtyTags”: [“Caries Prevention”, “Fissure Sealant”], “priority”: 1}, { “abnormalityType”: “double_teeth”, “recommendedDept”: [“Orthodontics”, “Pediatric Dentistry”], “specialtyTags”: [“Early Orthodontics”, “Retained Primary Teeth”], “priority”: 1} ]}}`

[0069] Further explanation:

[0070] The `mappingTable` array in this configuration list establishes a mapping relationship between "abnormalityType" (abnormality type) and "recommendedDept" (recommended department) and "specialtyTags" (specialty tags). By parsing this table, the engine prioritizes matching registered doctors with the highest "priority" and the highest semantic similarity between their specialty tags and the abnormality description, thereby achieving precise matching from medical problems to service resources.

[0071] The consultation form generation module 5 is used to generate templates following data minimization principles. At its core is a data template engine that generates structured consultation forms adhering to the principles of data minimization and purpose constraint. Based on the Personal Information Protection Act and privacy regulations such as HIPAA, it extracts and anonymizes fragments from the original test data, such as "patient nickname," "local image of tooth #16," and "de-identified health score," and dynamically fills them into a predefined HL7 FHIR standard questionnaire resource template. It also explicitly embeds data sharing authorization checkboxes to ensure users have clear control over the scope of use of their personal health information.

[0072] The online consultation interface module 6 is used to connect with compliant service providers and verify authorization. As a secure API gateway, it connects with compliant third-party internet hospital service providers or health consultation platforms. This module enforces an authorization verification process based on the OAuth 2.0 protocol, performing authentication and token verification on every API request, and mandating the use of TLS 1.3 encryption during data transmission to ensure the confidentiality and integrity of the communication channel and data payload.

[0073] The Service Feedback Parsing Module 7 is used to extract key suggestions based on NLP. It deploys a Transformer-based natural language processing pipeline to parse unstructured or semi-structured text from doctor or consultant responses. This module extracts core elements from key medical suggestions, such as diagnostic conclusions, medication recommendations, and follow-up schedules, using named entity recognition technology. It also establishes triple relationships between "teeth-disease-intervention" using relation extraction technology, transforming free text into structured data that can be used by downstream modules.

[0074] The observation plan synchronization module 8 is used to update observation targets / cycles. It receives structured data output from the service feedback parsing module 7 and uses it to update the observation targets, monitoring frequency, and cutoff period in the user's family observation plan. By calling the calendar service API, it automatically creates or updates electronic schedules containing tasks such as "daily oral hygiene monitoring" and "furrow sealant effect review every 14 days," and sets corresponding reminder rules, with user authorization.

[0075] The Data Authorization Management Module 9 manages authorization status and ensures compliance. As the privacy and compliance control center of the entire system, it manages the data authorization status throughout its entire lifecycle, ensuring compliance across the entire data processing chain. It dynamically maintains an authorization lifecycle state machine, recording the status of each shared data item, such as "Pending Authorization," "Authorized," "Expiration Date," "Expired," and "Delete Request Notification Requested." This module triggers an event via an asynchronous message queue when authorization expires, invoking the service provider's data deletion hook and generating audit logs to ensure data accountability.

[0076] In some implementations, the function of the above-mentioned modules is illustrated by the following typical examples:

[0077] Example 1: Early warning and consultation linkage for demineralization (attention level)

[0078] AI detection data receiving module 1 detected early demineralization of tooth #16 (upper left first molar), with a health score of 60 (a decrease of 15% compared to the previous test).

[0079] Risk assessment engine 2 conducts a risk assessment and obtains a comprehensive score of 61 points (attention level). Warning generation module 3 pushes an orange warning.

[0080] The service provider's matching engine 4 matches doctors, matching Dr. Zhang, a pediatric dentist specializing in caries prevention;

[0081] Users confirm data sharing authorization through the front-end interface. The system calls the consultation form generation module 5 to generate a structured consultation form following the data minimization principle and embeds authorization options.

[0082] Dr. Zhang suggested "fissure sealing + follow-up every 14 days, observation for 3 months" through the service feedback analysis module 7. After the observation plan was analyzed in the synchronous module 8, the observation plan was updated (adding oral hygiene monitoring, 14-day cycle, and 3-month period).

[0083] The authorization will automatically expire 24 hours after the consultation ends, and the data authorization management module 9 will notify the service provider to delete the data.

[0084] This embodiment illustrates the complete logic of a closed loop from anomaly detection and identification to online consultation. The AI ​​detection data receiving module 1 analyzes the enamel image obtained by polarized light scattering imaging and detects subsurface demineralization in tooth 16 (left upper first molar), which presents as chalky white spots. The health score output by its pixel-level classification model is 60 points, and compared with the longitudinal time-series data recorded in previous detections, it shows a 15% downward trend, triggering a deterioration warning.

[0085] Based on the early_caries type, the service provider matching engine 4 matched the patient with Dr. Zhang, who specializes in caries prevention and fissure sealing, by querying the registered doctors' specialty tags.

[0086] The system generates a digital digest of the user's authorized content using a hash function (such as SHA-256) and adds a timestamp to form an undeniable electronic authorization record, which meets HIPAA's audit requirements for authorization behavior. This credential is then used as a token to establish a text and image consultation service through an encrypted tunnel.

[0087] 24 hours after the end of this consultation session, the timer in the data authorization management module 9 triggers a status change, and the authorization status is moved from "authorized" to "expired". The system automatically notifies the service provider to execute the data deletion interface through the Webhook callback mechanism. The audit information of the entire process is recorded in an immutable log.

[0088] Example 2: Early Warning and Offline Guidance for Double Rows of Teeth (Attention Level)

[0089] AI detection data receiving module 1 detected double rows of teeth in tooth #26 (60% of permanent teeth have erupted), with a health score of 45.

[0090] Risk assessment engine 2 conducts a risk assessment and obtains a comprehensive score of 68.5 (attention level), and early warning generation module 3 pushes a strong reminder;

[0091] The service provider's matching engine 4 matches doctors, specifically orthodontists (specialty: retained deciduous teeth);

[0092] The doctor suggested "extracting the retained deciduous tooth offline as soon as possible" through the service feedback analysis module 7. The system upgraded to the emergency level through the observation plan synchronization module 8, pushed the offline appointment portal, and the data authorization management module 9 updated the observation plan to "monitoring for 7 days after tooth extraction".

[0093] This embodiment demonstrates how the system guides users to seek precise offline medical care. The AI ​​detection data receiving module 1 analyzes CBCT or intraoral scan data from the 26th tooth region, identifying ectopic eruption of the permanent tooth with an eruption rate of 60%, and retention of the deciduous tooth root due to a resorption rate below 50%, resulting in a "double row of teeth" phenomenon. The overall health assessment score is 45 points. The risk assessment engine 2 calculates a comprehensive score of 68.5 points (attention level), and the warning generation module 3 pushes a strong reminder including a tooth position diagram and the text "Double row of teeth risk, please pay attention."

[0094] Example 3: Treatment of mild gingival redness and swelling (monitoring level)

[0095] AI detection data receiving module 1 detected mild gum redness and swelling in tooth #36, with a health score of 75.

[0096] Risk assessment engine 2 performs a risk assessment and obtains a comprehensive score of 31 points (monitoring level).

[0097] The system stores monitoring-level anomaly records in the background. The observation plan synchronization module 8 marks them in yellow on the tooth location map. This does not trigger the NLP processing flow of the service feedback parsing module 7. The observation plan synchronization module 8 updates the observation plan notes without actively pushing reminders. If the health level is less than 60 in the next detection, it will be upgraded to the attention level.

[0098] This embodiment demonstrates the system's silent monitoring mechanism for low-risk anomalies. The AI ​​detection data receiving module 1, through analysis of multispectral fluorescence images, detected abnormal blood oxygen saturation in the buccal gingival tissue of tooth 36, classifying it as mild gingival redness and swelling, but without significant alveolar bone resorption, resulting in a health score of 75. The risk assessment engine 2 categorized this condition as "gingivitis_mild," with a severity score of only 30 and no urgent indications, ultimately achieving a comprehensive score of 31, falling into the "monitor" range. Since the trigger level was not reached, the warning generation module 3 did not generate any proactive push notifications. The service feedback parsing module 7 simply associated the text tag "mild gingival redness and swelling" with tooth 36 on the tooth position map in the background, forming an observation record. The observation plan synchronization module 8 added a note to the observation report of this test: "Mild gingival redness and swelling in tooth 36, recommended to strengthen interproximal cleaning," and included it in the key focus list for the next routine test, without changing the original examination cycle. This process is completed entirely without the user's awareness. The system will only automatically upgrade the risk level to "attention level" and trigger the intervention process if the dental health score drops below 60 points in the next test.

[0099] The AI ​​detection data receiving module 1 first uses a target detection algorithm (such as YOLOv8 or Faster R-CNN) to segment and locate teeth in the input oral image, identifying the anatomical location (FDI tooth position number) of each tooth. Based on the image features of the enamel, dentin, and gingival regions, a regression model outputs a health score of 0-100 for each tooth. Next, it acquires multi-dimensional appearance features of the labial, lingual, and proximal surfaces of the teeth and inputs them into a health assessment model trained on a dataset of millions of clinical images for inference. Finally, it outputs a quantitative health score of 0-100, which reflects the probability of disease.

[0100] Risk assessment engine 2 proposes a total score of 0-100 based on the health score results, and early warning generation module 3 provides corresponding warning levels based on the total score. Combining the abnormality classification, severity, and rate of change in health across tests, a comprehensive risk score of 0-100 is proposed using a multivariate logistic regression or gradient boosting tree model. Early warning generation module 3 uses this total score as the sole decision criterion, compares it with a preset threshold, and thus provides the corresponding warning level among emergency, attention, and monitoring levels.

[0101] The service provider matching engine 4 matches doctors in the appropriate departments based on the total score and corresponding level alerts. The total score output by the risk assessment engine 2 and the corresponding level alerts generated by the warning generation module 3 are used as input query conditions and matched against the doctor knowledge graph. Its logic includes not only specialty matching but also real-time dynamic factors such as doctor's patient load, positive review rate, and online status, comprehensively calculating a recommendation score and ultimately outputting a ranked list of doctors, accurately matching doctors in the appropriate departments.

[0102] Users confirm data sharing authorization through Consultation Form Generation Module 5, generating a HIPAA-compliant electronic authorization letter and establishing a text and image consultation service. With informed consent, clearly understanding the data's purpose, scope of sharing, and retention period, users actively check the box to confirm data sharing authorization. This operation generates an undeniable HIPAA-compliant electronic authorization letter using digital signature technology. This authorization letter serves as the access token for establishing the text and image consultation service; only after the authorization takes effect is the encrypted communication link established.

[0103] Doctors submit suggestions through the service feedback parsing module 7 and update the observation plan through the observation plan synchronization module 8. After submitting suggestions via the text and image consultation interface, the submitted suggestion text flows through the NLP pipeline of the service feedback parsing module 7 and is transformed into structured medical action instructions. These instructions are then parsed by the rule engine of the observation plan synchronization module 8, extracting parameters such as "follow-up cycle" and "observation goals," and automatically generating or updating the family member observation plan tasks associated with the user account.

[0104] The authorization automatically expires 24 hours after the consultation ends, and the service provider is notified to delete the data via the data authorization management module 9. After the consultation ends, a scheduled task with a 24-hour countdown is activated. After 24 hours, the authorization automatically expires, and the task sends a signed data clearing request to the service provider's system through the event publishing mechanism of the data authorization management module 9, notifying the service provider to delete all cached data associated with this consultation from its database, thus completing the closed-loop management of the data lifecycle.

[0105] It should be noted that the specific model and specifications of the hardware supporting the risk assessment engine 2, as a functional module integrating algorithms and software, need to be selected and determined based on the actual specifications of the device, such as the throughput of the image data stream being processed, the number of concurrent users, and the latency requirements. The specific model and specifications to be adopted need to be selected and determined based on the actual specifications of the device. The specific selection and calculation methods adopt existing technologies in this field, so they will not be elaborated here.

[0106] The functional principle of this invention can be explained through the following operational methods:

[0107] Step S1, Anomaly Identification and Data Reception, refers to receiving oral health detection data and identifying abnormal teeth and their types based on AI analysis results. Through the front-end sensor interface of the AI ​​detection data receiving module 1, multimodal oral health detection data, such as RGB-D images and fluorescence images, are received. Based on the analysis results, the deep convolutional neural network embedded in the module performs instance segmentation of the dentition in the oral cavity, identifies teeth with abnormalities such as caries, malocclusion, and periodontal inflammation, and automatically assigns international dental association (FDI) standard tooth position numbers and abnormality type labels.

[0108] Step S2, intelligent risk level assessment, refers to evaluating the risk level by combining anomaly type and severity indicators, and generating corresponding early warning information. Risk levels are divided into emergency, attention, and monitoring levels. Emergency level corresponds to a comprehensive score ≥80 points, attention level corresponds to 60–79 points, and monitoring level corresponds to <60 points. This involves calling the risk assessment engine 2, combining the anomaly type label output from step S1 with the severity indicator quantified by the fuzzy logic system, and evaluating the risk level through a multi-factor weighted model. This model outputs a comprehensive score of 0-100 points, and based on whether the score falls within the range of [80,100], [60,79], or [0,59], it triggers the early warning generation module 3 to generate differentiated early warning information for the corresponding emergency, attention, and monitoring levels. The internal logic of this step follows a two-tier architecture consisting of a judgment layer and a response layer, as detailed below:

[0109] Mermaidflowchart TD subgraph Risk Level Determination Layer A [Comprehensive Score Calculation Result (Example: 61 points)] --> B {Risk Level Determination} B --> |80-100 points| C [Emergency Level 🔴] B --> |60-79 points| D [Attention Level 🟠] B --> |0-59 points| E [Monitoring Level 🟡] end subgraph Differentiated Response Processing Layer %% Emergency Level Response C --> C1 [Strong Alert Push • Sound + Vibration + Full-Screen Pop-up • Red Highlight] C1 --> C2 [Immediate Medical Treatment Guidance • Nearby Public Medical Institutions • Redirect to Official Registration Platform • Emergency Room Marking] C2 --> C3 [Popular Science Education • Why is it Emergency • Delay Risk] C3 --> C4 [Timeliness Management • Response within 2 Hours • Unprocessed Secondary Alert] C4 --> C5 [Manual Intervention • Customer Service Phone Follow-up] %% Attention Level Response D --> D1 [Standard Alert Push • [Notification bar + in-app red dot • orange marker] D1 --> D2 [Online consultation recommendation (distinguishing qualifications)] D2 --> D21 [Internet hospital • with medical qualifications • can provide diagnostic advice] D2 --> D22 [Health consultation institution • non-medical institution • only provides health guidance] D2 --> D3 [Data authorization prompt • requires user confirmation] D3 --> D4 [Science education • why we care • handling suggestions] D4 --> D5 [Timeliness management • respond within 24 hours • unhandled downgraded to monitoring level] %% Monitoring level response

[0110] E --> E1 [Silent Update • No Active Push Notifications] E1 --> E2 [Tooth Position Map Marking • Abnormal Teeth Marked in Yellow] E2 --> E3 [Plan Update • Included in Observation Plan • Focus on During Next Test] end %% Overall Related Risk Level Determination Layer --> Differentiated Response Processing Layer

[0111] Step S3, Automatically Triggering the Online Consultation Service Process: When the risk level reaches the "Attention" level or above, the online consultation process is automatically triggered. Users can select "Automatic Trigger" or "Manual Confirmation Trigger" mode in the system settings. Automatic triggering is enabled by default, but the system will push a notification and provide a cancellation option before triggering. This process clearly demonstrates the logical mapping from quantitative scoring to qualitative level, and then to specific action strategies. The single numerical output of the judgment layer directly activates the preset sub-process set in the response layer, which includes three dimensions: human-computer interaction, service guidance, and timeliness management, achieving precise risk stratification management.

[0112] In some implementations, when the risk level determined in step S2 reaches the "concern level" or above, the system's event bus will issue a "consultation recommendation" event. This event will automatically trigger the online consultation process orchestration engine, sequentially executing sub-tasks such as doctor matching, consultation form generation, and authorization request, without requiring the user to manually search for the service entry point.

[0113] Furthermore, the internal mechanism of the online consultation process is a microservice orchestration sequence consisting of four key subtasks designed to implement step S3. The online consultation process is as follows:

[0114] In stage t1, the doctor's specialty mapping table is queried based on the anomaly type to match the corresponding specialist doctor / consultant. The service provider's matching engine 4, based on the anomaly type label output in step S1, queries its in-memory doctor's specialty mapping table, calculates the similarity between the anomaly description and the doctor's specialty label using a semantic matching algorithm, and matches the doctor or consultant.

[0115] In stage t2, a structured consultation form is automatically generated based on the test data. The consultation form generation module 5, based on the test data, tooth position, health score, etc. generated in steps S1 and S2, calls the template engine and automatically fills in and generates a structured consultation form in FHIR questionnaire resource format, strictly following the principle of minimum necessity.

[0116] In phase t3, the system pushes a consultation recommendation and data sharing authorization request to the user. After the user confirms, a consultation connection is established. The system pushes a consultation recommendation card and a data sharing authorization request to the user's terminal via a message push channel. This request lists all data items to be shared and their purposes. After the user gives explicit consent through the graphical interface, the system establishes an end-to-end encrypted consultation connection with the selected service provider through the online consultation interface module 6.

[0117] In phase t4, the system can connect to compliant internet hospital remote consultation service providers or provide non-clinical oral health consultation services. The online consultation interface module 6 acts as a unified relay, dynamically selecting the API adapter based on the qualification type of the matching results. It can connect to the webRTC signaling service of compliant internet hospital remote consultation service providers, or to non-clinical oral health consultation services that provide text and image communication capabilities.

[0118] Step S4, Service Feedback Reception, refers to receiving the analysis results and suggestions submitted by the service provider based on the structured consultation form; if connecting with a qualified medical institution, it involves receiving doctors' diagnoses and treatment suggestions. If connecting with an internet hospital with a "Medical Institution Practice License," the received content includes legally valid electronic prescriptions, diagnostic certificates, and treatment suggestions; if connecting with a health consultation institution, it only receives non-medical health guidance information, and the categories are clearly distinguished on the interface. In some implementations, non-medical oral health consultation services only provide health guidance and popular science advice, and may not issue diagnostic conclusions or prescriptions, and the consultation interface prominently displays "This service does not replace offline medical institution diagnosis and treatment."

[0119] Step S5 involves automatically updating the observation plan by parsing treatment recommendations, extracting key elements, synchronizing them to the family member's observation plan, and automatically adjusting observation targets and testing cycles. The service feedback parsing module 7 analyzes the text received in step S4, using dependency parsing technology to analyze the semantic structure of the treatment recommendations, extracting core verbs such as "re-examination," "extraction," and "medication," along with their associated argument roles such as time, location, and method, and transforming them into a structured observation plan protocol. Upon receiving this protocol, the observation plan synchronization module 8 automatically adjusts the observation targets and testing cycles in the family member's oral health record, forming a dynamic and personalized closed-loop health management system.

[0120] A complete, structured consultation form that conforms to the principle of data minimization is shown below. It is generated in stage t2 of the consultation process and presented to the user for confirmation in stage t3:

[0121] Mermaidflowchart TD subgraph Consultation Form Header (Basic Information) A [Structured consultation form from: AI detection intelligent recommendation | Generation time: 2026-05-06 10:35 🔒 Data minimization processing] end subgraph Core Module (based on document 214 table structure) A --> B [📋 Basic Information Area (Anonymized)] B --> B1 [Patient nickname: Xiaoming Age: 8 years old | Gender: Male ☐ Real name (Please authorize if treatment is required) Consultation method: Based on AI detection results intelligent recommendation] B --> B2 [💡 Privacy notice: Data minimization processing has been performed to protect user privacy] A --> C [🎯 Chief complaint area (Auto-fill)] C --> C1 [AI detection found early demineralization, involving tooth position: 16 (left upper first molar)] A --> D [📊 Present illness area (Anonymized)] D --> D1 [Discovery Time: 2026-05-06; Test Results: Health Score 60 (Needs Attention), Early Demineralization, Chalky Spots on Enamel; Historical Comparison: 2026-04-22 (75 points), ↓15% (Deteriorating Trend); Symptom Description: No Obvious Pain / Sensitivity to Hot and Cold] A --> E [📸 Examination Data Area (Localized)] E --> E1 [2D Tooth Map (Tooth Area Only, Yellow Mark on Tooth #16) 🔒 Facial Features Removed]

[0122] E --> E2 [Close-up of tooth #16 | Comparison with historical image from April 22nd 🔒 Image cropped, no facial features] E --> E3 [📄 Complete AI Analysis Report (De-identified Version)] A --> F [🔐 Data Sharing Authorization Area] F --> F1 [Selectable Sharing Items: ☑ Desensitized Nickname, Age, Gender ☑ Health Score ☑ Close-up Image of Tooth #16 ☐ Real Name (Please select if treatment is required)] F --> F2 [Authorization Description: Purpose of Authorization: Oral Health Consultation Service Legal Basis: Article 13 of the Individual Insurance Law Validity Period: Automatically expires 24 hours after consultation] A --> G [🏥 Qualification Information Area] G --> G1 [Service Provider: XX Children's Hospital Internet Hospital (Medical Institution) License Number: XX | Attending Physician: Dr. Zhang (Specialty: Caries Prevention)] end %%Style Optimization (Adhering to document data minimization and compliance requirements) style A fill:#f0f8ff,stroke:#4299e1,stroke-width:2px style F fill:#fef7fb,stroke:#9f7aea,stroke-width:2px style E fill:#e8f4f8,stroke:#38b2ac,stroke-width:1px

[0123] The consultation form structure embodies the concept of "protecting privacy through design," with data in each area being de-identified or localized, and users being given individual control over the authorization of each piece of data.

[0124] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A system for linking early warning of oral health abnormalities with remote consultation, characterized in that, include: AI detection data receiving module (1), which is used for localized processing of sensitive areas; Risk assessment engine (2), which is used to calculate risk levels in multiple dimensions; The early warning generation module (3) is used to generate differentiated early warnings; Service provider matching engine (4), which is used to match specialist doctors / consultants; Consultation form generation module (5), which is used to generate templates following data minimization; The online consultation interface module (6) is used to connect with compliant service providers and verify authorization; The service feedback parsing module (7) is used to extract key suggestions based on NLP. The observation plan synchronization module (8) is used to update the observation target / cycle; The data authorization management module (9) is used to manage the authorization status and ensure compliance.

2. The oral health abnormality early warning and remote consultation linkage system according to claim 1, characterized in that, The AI ​​detection data receiving module (1) first performs tooth instance segmentation and localization on the input oral image through the target detection algorithm, identifies the anatomical position of each tooth, and outputs a health score of 0-100 for each tooth through a regression model based on the image features of the enamel, dentin and gingival regions.

3. The oral health abnormality early warning and remote consultation linkage system according to claim 2, characterized in that, The risk assessment engine (2) proposes a total score of 0-100 based on the health score, and the early warning generation module (3) provides a corresponding level reminder based on the total score.

4. The oral health abnormality early warning and remote consultation linkage system according to claim 3, characterized in that, The service provider matching engine (4) matches doctors in the corresponding departments based on the total score and corresponding level.

5. The oral health abnormality early warning and remote consultation linkage system according to claim 4, characterized in that, Users confirm data sharing authorization through the consultation form generation module (5), generate HIPAA compliant electronic authorization, and establish graphic consultation.

6. The oral health abnormality early warning and remote consultation linkage system according to claim 5, characterized in that, Doctors provide suggestions through the service feedback analysis module (7) and update the observation plan through the observation plan synchronization module (8).

7. The oral health abnormality early warning and remote consultation linkage system according to claim 6, characterized in that, The authorization expires automatically 24 hours after the consultation ends, and the service provider is notified to delete the data through the data authorization management module (9).

8. The method of using the oral health abnormality early warning and remote consultation linkage system as described in claim 7, characterized in that, The method of use includes the following steps: Step S1, anomaly identification and data reception; Step S2, intelligent risk level assessment; Step S3 automatically triggers the online consultation service process; Step S4, Service feedback received; Step S5: Observe the plan to update automatically.

9. The method of using the oral health abnormality early warning and remote consultation linkage system according to claim 8, characterized in that, Step S1, anomaly identification and data reception, refers to receiving oral health test data and identifying abnormal teeth and their types based on AI analysis results; Step S2, intelligent risk level assessment, refers to assessing the risk level by combining the anomaly type and severity indicators, and generating corresponding early warning information. The risk levels are divided into emergency level, attention level and monitoring level. Step S3, automatically triggering the online consultation service process, means that the online consultation process is automatically triggered when the risk level reaches the concern level or above; Step S4, receiving service feedback, refers to receiving the analysis results and suggestions submitted by the service provider based on the structured consultation form; If connected with a qualified medical institution, receive doctors' diagnoses and treatment recommendations; In step S5, the observation plan is automatically updated to analyze treatment suggestions, extract key elements and synchronize them to the family member observation plan, and automatically adjust the observation targets and detection cycle.

10. The method of using the oral health abnormality early warning and remote consultation linkage system according to claim 9, characterized in that, The online consultation process is as follows: In phase t1, the doctor's specialty mapping table is queried based on the anomaly type to match the corresponding specialist doctor / consultant; In phase t2, a structured consultation form is automatically generated based on the detection data; In stage t3, a consultation recommendation and data sharing authorization request are pushed to the user. After the user confirms, a consultation connection is established. In phase t4, the system can connect with compliant internet hospital remote consultation service providers or provide non-medical oral health consultation services.

Images